Hospital Medicine

Nothing changes without personal transformation.

—W Edwards Deming, 1986

Failure is common among quality improvement projects, but also predictable. Health professionals have multiple competing priorities. Improvement projects rarely reduce an individual’s workload. In our experience coaching health professionals, we have found that improvement teams often overlook two important facts: improvement requires behavior change, and everyone resists change.

Quality improvement education focuses on the development of technical skills (eg, process mapping, measure development, data analysis). Technical skills are necessary, but insufficient, to lead change. Process maps and run charts guide improvement work but alone do not motivate frontline staff to change workflows. Rather, soft skills (eg, communication, negotiation, change management, influencing others) convince frontline staff and hospital leaders that change is worth their time and effort.^1,2 Successful improvement teams combine technical skills and soft skills to inspire behavior change.

We propose three practical skills that all improvement teams can adopt to inspire change:

Understand your stakeholders’ needs. Early identification and engagement of stakeholders (individuals or groups who may affect or be affected by the project) is critical. Improvement teams must consider stakeholders at multiple levels in the organization, from frontline staff to executives. The easiest way to understand stakeholders is by talking to them. Often, stakeholders lack time for scheduled meetings, so teams must rely on informal conversations in hallways and elevators. The key is to understand what will motivate the stakeholder to change. Put yourself in the stakeholders’ shoes: What are their needs and priorities? How might their needs and priorities motivate them to change? What potential barriers exist that prevent the stakeholder from making a change?

Tailor your message to establish a rationale for change. Build upon what was learned from stakeholders and decide how the rationale for change will be communicated. What can you say that will influence others to see the problem as important? Recognize that the rationale is different for different stakeholders; a financial rationale may inspire hospital leaders but alienate staff who are driven by patient and staff satisfaction. Even carefully crafted messages may not resonate with stakeholders as intended. Improvement teams must monitor the impact of their message with different stakeholders. Developing a clear, concise, and compelling rationale for change is often challenging and iterative. Multiple communication channels (ie, email, newsletters, formal and informal conversations) must be employed to spread your message.

Share small and large wins. Talking with stakeholders is not a one-time event. Stakeholder interest may decrease over time. Frontline staff can become complacent, falling back into old behaviors. Priorities of hospital leadership can shift. Successful teams maintain lines of communication throughout the project to share successes and sustain stakeholder buy-in. Small and large wins matter. Project outcomes (large wins) may take months to achieve. Teams can maintain stakeholder interest by demonstrating that project processes are feasible and acceptable (small wins). Maintaining regular communication also affords teams the opportunity for early identification of organizational barriers and facilitators that may impact their project. Ongoing communication of project wins sets the stage for sustainment by embedding the change within the local culture.

The goal of any improvement project is to create sustainable change. To do this, improvement teams often need hundreds of people to change the way they work. Change is hard, but improvement teams can overcome resistance to it by strategically engaging stakeholders and thoughtfully communicating the rationale for change.

Nothing changes without personal transformation.

—W Edwards Deming, 1986

Failure is common among quality improvement projects, but also predictable. Health professionals have multiple competing priorities. Improvement projects rarely reduce an individual’s workload. In our experience coaching health professionals, we have found that improvement teams often overlook two important facts: improvement requires behavior change, and everyone resists change.

Quality improvement education focuses on the development of technical skills (eg, process mapping, measure development, data analysis). Technical skills are necessary, but insufficient, to lead change. Process maps and run charts guide improvement work but alone do not motivate frontline staff to change workflows. Rather, soft skills (eg, communication, negotiation, change management, influencing others) convince frontline staff and hospital leaders that change is worth their time and effort.^1,2 Successful improvement teams combine technical skills and soft skills to inspire behavior change.

We propose three practical skills that all improvement teams can adopt to inspire change:

Understand your stakeholders’ needs. Early identification and engagement of stakeholders (individuals or groups who may affect or be affected by the project) is critical. Improvement teams must consider stakeholders at multiple levels in the organization, from frontline staff to executives. The easiest way to understand stakeholders is by talking to them. Often, stakeholders lack time for scheduled meetings, so teams must rely on informal conversations in hallways and elevators. The key is to understand what will motivate the stakeholder to change. Put yourself in the stakeholders’ shoes: What are their needs and priorities? How might their needs and priorities motivate them to change? What potential barriers exist that prevent the stakeholder from making a change?

Tailor your message to establish a rationale for change. Build upon what was learned from stakeholders and decide how the rationale for change will be communicated. What can you say that will influence others to see the problem as important? Recognize that the rationale is different for different stakeholders; a financial rationale may inspire hospital leaders but alienate staff who are driven by patient and staff satisfaction. Even carefully crafted messages may not resonate with stakeholders as intended. Improvement teams must monitor the impact of their message with different stakeholders. Developing a clear, concise, and compelling rationale for change is often challenging and iterative. Multiple communication channels (ie, email, newsletters, formal and informal conversations) must be employed to spread your message.

Share small and large wins. Talking with stakeholders is not a one-time event. Stakeholder interest may decrease over time. Frontline staff can become complacent, falling back into old behaviors. Priorities of hospital leadership can shift. Successful teams maintain lines of communication throughout the project to share successes and sustain stakeholder buy-in. Small and large wins matter. Project outcomes (large wins) may take months to achieve. Teams can maintain stakeholder interest by demonstrating that project processes are feasible and acceptable (small wins). Maintaining regular communication also affords teams the opportunity for early identification of organizational barriers and facilitators that may impact their project. Ongoing communication of project wins sets the stage for sustainment by embedding the change within the local culture.

The goal of any improvement project is to create sustainable change. To do this, improvement teams often need hundreds of people to change the way they work. Change is hard, but improvement teams can overcome resistance to it by strategically engaging stakeholders and thoughtfully communicating the rationale for change.

Nothing changes without personal transformation.

—W Edwards Deming, 1986

Failure is common among quality improvement projects, but also predictable. Health professionals have multiple competing priorities. Improvement projects rarely reduce an individual’s workload. In our experience coaching health professionals, we have found that improvement teams often overlook two important facts: improvement requires behavior change, and everyone resists change.

Quality improvement education focuses on the development of technical skills (eg, process mapping, measure development, data analysis). Technical skills are necessary, but insufficient, to lead change. Process maps and run charts guide improvement work but alone do not motivate frontline staff to change workflows. Rather, soft skills (eg, communication, negotiation, change management, influencing others) convince frontline staff and hospital leaders that change is worth their time and effort.^1,2 Successful improvement teams combine technical skills and soft skills to inspire behavior change.

We propose three practical skills that all improvement teams can adopt to inspire change:

Understand your stakeholders’ needs. Early identification and engagement of stakeholders (individuals or groups who may affect or be affected by the project) is critical. Improvement teams must consider stakeholders at multiple levels in the organization, from frontline staff to executives. The easiest way to understand stakeholders is by talking to them. Often, stakeholders lack time for scheduled meetings, so teams must rely on informal conversations in hallways and elevators. The key is to understand what will motivate the stakeholder to change. Put yourself in the stakeholders’ shoes: What are their needs and priorities? How might their needs and priorities motivate them to change? What potential barriers exist that prevent the stakeholder from making a change?

Tailor your message to establish a rationale for change. Build upon what was learned from stakeholders and decide how the rationale for change will be communicated. What can you say that will influence others to see the problem as important? Recognize that the rationale is different for different stakeholders; a financial rationale may inspire hospital leaders but alienate staff who are driven by patient and staff satisfaction. Even carefully crafted messages may not resonate with stakeholders as intended. Improvement teams must monitor the impact of their message with different stakeholders. Developing a clear, concise, and compelling rationale for change is often challenging and iterative. Multiple communication channels (ie, email, newsletters, formal and informal conversations) must be employed to spread your message.

Share small and large wins. Talking with stakeholders is not a one-time event. Stakeholder interest may decrease over time. Frontline staff can become complacent, falling back into old behaviors. Priorities of hospital leadership can shift. Successful teams maintain lines of communication throughout the project to share successes and sustain stakeholder buy-in. Small and large wins matter. Project outcomes (large wins) may take months to achieve. Teams can maintain stakeholder interest by demonstrating that project processes are feasible and acceptable (small wins). Maintaining regular communication also affords teams the opportunity for early identification of organizational barriers and facilitators that may impact their project. Ongoing communication of project wins sets the stage for sustainment by embedding the change within the local culture.

The goal of any improvement project is to create sustainable change. To do this, improvement teams often need hundreds of people to change the way they work. Change is hard, but improvement teams can overcome resistance to it by strategically engaging stakeholders and thoughtfully communicating the rationale for change.

In the United States, health disparities in COVID-19 outcomes (including morbidity and mortality) based on race and ethnicity have been described in the scientific literature and mainstream media.^1-7 According to the US Centers for Disease Control and Prevention (CDC), Hispanic people are 3.2 times more likely to be hospitalized with COVID-19 than non-Hispanic White people.⁸ Further, Hispanic people diagnosed with COVID-19 are 2.3 times more likely to die, adjusted for age, than non-Hispanic White people.⁹ As the epicenter of the COVID-19 pandemic shifted from the Northeast to the South, the CDC reported that, among people who died from COVID-19 in the United States from May to August 2020, the percentage of Hispanic people increased from 16.3% to 26.4%.¹⁰

Published studies on the effect of ethnicity on critical illness or mortality for hospitalized COVID-19 patients are limited and inconsistent. While some studies reported a higher mortality rate for Hispanic patients,^11-15 others showed no difference.^4,16,17 A recent meta-analysis found that intensive care unit (ICU) utilization and mortality were slightly higher among Hispanic COVID-19 inpatients, but this finding did not reach statistical significance.¹⁸ Past studies from different healthcare systems were limited by the small sample size of hospitalized Hispanic patients and the heterogeneity of patients. A comprehensive analysis from a large healthcare system with sufficient sample size is needed to understand the impact of ethnicity on clinical outcomes of hospitalized COVID-19 patients.

Texas Health Resources (THR) is a large integrated healthcare system serving the Dallas-Fort Worth-Arlington (DFW) metropolitan area. According to the 2019 US Census Bureau American Community Survey, Hispanic people comprise 18.4% of the population of this geographic area.¹⁹ Congruent with the CDC’s findings, Hispanic patients account for a disproportionate share (32.2%) of hospitalized COVID-19 patients at THR relative to the area’s demographic composition. Aware of the increased risk, we undertook an analysis of the clinical outcomes and the clinical, social, and demographic characteristics of Hispanic patients hospitalized at THR with COVID-19. Our primary goal was to investigate whether clinical outcomes differ by ethnicity among patients hospitalized with COVID-19 and, if so, whether inpatient care or preadmission factors contribute to this difference.

Study Setting and Overview

We collected data from the single electronic health record (EHR) used by 20 THR hospitals located across the DFW metropolitan area. THR is the largest faith-based, nonprofit health system in North Texas, operating 20 acute care hospitals. Including all access points, such as outpatient facilities and physician group practices, THR serves 7 million residents in 16 counties in North Texas, of whom 16.8% are Hispanic, 73.3% are non-Hispanic, and 9.9% are unclassified, congruent with demographics in the DFW area.

The institutional review boards at THR and UT Southwestern Medical Center approved the study under a waiver of informed consent (as a minimal-risk medical record review). After collection, all data were de-identified prior to statistical analysis.

Cohort, Outcomes, and Covariables

The study cohort included 6097 adult patients with laboratory-confirmed COVID-19 (age ≥18 years) who were admitted as inpatients from March 3 to November 5, 2020. The primary outcomes included ICU utilization and death during hospitalization. We described demographic characteristics using the following variables: age (18–49, 50–64, 65–79, ≥80 years), sex, self-reported ethnicity, and primary spoken language.

We defined a severe baseline condition as an elevated respiratory subscore parsed from the overall MSOFA (Modified Sequential Organ Failure Assessment),²⁰ an elevated Epic Deterioration Index (EDI),²¹ or an elevated C-reactive protein level (CRP) at baseline (any elevated CRP). Baseline referred to the variable mean during the first available 12-hour window of measurement during the COVID-19 hospital admission, including variables obtained in the emergency department (ED). An elevated MSOFA referred to a score of 4, corresponding to an SpO₂/FiO₂ < 150. Elevated EDI referred to a baseline EDI > 45. An elevated CRP referred to a baseline CRP > 20 mg/dL.²²

Variables reflecting access to healthcare included: THR EHR creation year (representing the first time patients accessed the THR health system), insurance payor type, and presence of a primary care provider (PCP). The federal government established the COVID-19 Claims Reimbursement to Health Care Providers and Facilities for Testing, Treatment, and Vaccine Administration for the Uninsured program. The insurance payor for patients covered by this program is designated as COVID-19 HRSA. Presence of a PCP reflects any documented PCP, regardless of affiliation with THR. We selected these access metrics opportunistically, as they were consistently documented in the EHR and readily available for analysis.

We used 12 variables to describe comorbidities or underlying conditions that, according to the CDC, increased patients’ risk of severe illness from COVID-19²³: diagnoses of diabetes, hypertension, obesity, chronic obstructive pulmonary disease (COPD), asthma, smoking, other lung disease, heart failure, kidney disease without end-stage renal disease (ESRD), ESRD, liver disease, and cancer. We identified comorbidities by mining the structured diagnosis codes documented in the EHR prior to and during the COVID-19 admission. Sources for diagnoses included final billed diagnosis codes, working diagnosis codes, problem list, and reason for visit. The definition of diabetes included previously recorded diabetes or baseline hemoglobin A1c > 9%. We also recorded the presence of four major COVID-19 treatments: steroids, remdesivir, tocilizumab, and fresh frozen plasma (FFP) from convalescent patients.^24-26 Each treatment variable was defined by receipt of one or more doses.

Statistical Analysis

To analyze patient outcomes based on ethnicity, we divided the study cohort into a Hispanic group and a non-Hispanic group based on self-reported ethnicity in the EHR. To study the potential impact of primary language among Hispanic patients, we divided them into English-speaking and non-English-speaking patients based on their self-reported primary language. As a result, we analyzed three groups of patients: (1) non-Hispanic, (2) Hispanic and English speaking, and (3) Hispanic and non-English speaking. We tested differences of a given categorical variable across the three groups using the chi-square test for each age subgroup (18–49, 50–64, 65–79, ≥80 years). The Cochran-Mantel-Haenszel test was used for the overall difference adjusted for age. To assess whether an observed disparity in treatment existed across the three groups, we tested the difference in the administration of four major therapeutics for COVID-19, including steroids, remdesivir, tocilizumab, and convalescent plasma. To determine whether any groups had elevated disease severity at hospital admission (baseline), we tested the difference in four disease-severity metrics across the ethnic-language groups: (1) elevated respiratory MSOFA score, (2) elevated EDI, (3) elevated CRP level, and (4) any of the three conditions.

To study the associations with ICU utilization and death, respectively, we performed a multivariable analysis using a generalized linear mixed model with binomial distribution and a logit link function. In each analysis model, the hospital of admission was included as a random-effect variable to account for the potential treatment variations among different hospitals, while other variables were regarded as fixed effects. In the first multivariable analysis (Model 1), all demographic variables, including age, sex, and ethnicity, and different types of comorbidities and underlying conditions, were included as fixed-effect variables in the initial model, and then backward stepwise variable selection was performed to establish the final model (Model 1). We performed the backward stepwise variable selection separately for the outcome of ICU use or mortality. Based on Akaike information criterion (AIC), during each iteration the fixed-effect variable that led to the largest decrease in the AIC value was removed, and the variable selection process was completed when the AIC value stopped decreasing. In Model 2, we added the disease-severity variable at baseline to the selected variable set derived from Model 1 to explore its effect on the associations between ethnicity and clinical outcomes. In Model 3, we added healthcare access–related variables, including first-time healthsystem access, payor type, and PCP availability to Model 2. We performed all statistical analyses using R, version 4.0.2 (R Foundation for Statistical Computing) in RStudio (version 1.3.1093).

Distinct Demographic and Comorbidity Patterns for Three Ethnic-Language Groups

We identified 6097 adult patients (age ≥18 years) who had confirmed COVID-19 disease and were hospitalized between March 3 and November 5, 2020. Demographic characteristics and comorbidity for these patients are summarized in Table 1. Among these patients, 4139 (67.9%) were non-Hispanic and 1958 (32.1%) were Hispanic. Among the Hispanic patients, 1203 (61.4%) identified English as their primary language and 755 (38.6%) identified a non-English primary language. Age distribution was vastly different among the three ethnic-language groups (Table 1). Unlike the relatively balanced distribution across different age groups in the non-Hispanic group, more than half (55.8%) of the English-speaking Hispanic patients were in the youngest age group (18-49 years). A much lower fraction of Hispanic patients was among the oldest (≥80 years) age group (P < .001). Because COVID-19 clinical outcome is strongly associated with age,²⁷ we used age-stratified analysis when comparing group-level differences in patient outcomes.

Sex distribution also was different among the three groups, with the non-English-speaking Hispanic group having more male patients (53.0%). Diabetes and obesity, which are associated with clinical outcomes of COVID-19 patients, were more prevalent in Hispanic patients (Table 1). Non-English-speaking Hispanic patients had the highest diabetes rate (48.7% with documented diabetes; 15.8% with baseline HbA1c > 9%; P < .001). English-speaking Hispanic patients presented with the highest obesity rate (62.8%; P < .001). Appendix Table 1 provides detailed age-group-specific comorbidity distributions among ethnic-language groups.

Patients of Hispanic Ethnicity Experienced a Higher Rate of ICU Utilization and Mortality

Of the 6097 patients overall, 1365 (22.4%) were admitted to the ICU and 543 (8.9%) died in hospital. For non-Hispanic patients (n = 4139), 883 (21.3%) were admitted to the ICU and 373 (9.0%) died in hospital. For English-speaking Hispanic patients (n = 1203), 241 (20.0%) were admitted to the ICU and 91 (7.6%) died in hospital. For non-English-speaking Hispanic patients (n = 755), 241 (31.9%) were admitted to the ICU and 79 (10.5%) died in hospital. Figure 1 summarizes the age-stratified comparison of ICU utilization and mortality across the three ethnic-language patient groups. In all age groups, non-English-speaking Hispanic patients experienced a significantly higher ICU utilization rate compared to non-Hispanic patients (age-adjusted OR, 1.75; 95% CI, 1.47-2.08; P < .001). English-speaking and non-English-speaking Hispanic patients had a significantly higher mortality rate compared to non-Hispanic patients (age-adjusted OR, 1.53; 95% CI, 1.19-1.98; P = .001 for English-speaking Hispanic patients; age-adjusted OR, 1.43; 95% CI,: 1.10-1.86; P = .01 for non-English-speaking Hispanic patients).

To delineate the risk factors associated with ICU utilization and death, we performed multivariable logistic regression with stepwise variable selection. After adjusting for age, sex, and comorbidity (Model 1), the factors ethnicity and primary language were still strongly associated with ICU utilization and mortality (Appendix Table 2). Non-English-speaking Hispanic patients had an OR of 1.74 (95% CI, 1.41-2.15; P < .001) for ICU utilization and an OR of 1.54 (95% CI, 1.12-2.12; P = .008) for mortality compared to non-Hispanic patients. Similarly, English-speaking Hispanic patients had higher ICU utilization (OR, 1.28; 95% CI, 1.05-1.55; P = .01) and a higher mortality rate (OR, 1.60; 95% CI, 1.19-2.14; P = .002).

No Disparity in COVID-19 Therapeutics Observed Across Three Ethnic-Language Groups

Appendix Figure 1 summarizes the comparison of the administration of four major treatments across the three ethnic-language groups. We did not observe any underuse of COVID-19 therapeutics for Hispanic patients. Usage rates for these therapies were significantly higher, after adjusting for age, in Hispanic groups when compared to non-Hispanic patients (OR ranged from 1.21 to 1.96). Steroids were the most common treatment in all patient groups. Tocilizumab was used almost twice as frequently (OR, 1.96; 95% CI, 1.64-2.33; P < .001) in non-English-speaking Hispanic patients compared to non-Hispanic patients.

Patients of Hispanic Ethnicity Had More Severe Disease at Hospital Admission

Figure 2 shows that non-English-speaking Hispanic patients had a higher rate of severe illness at admission based on each of these metrics: high respiratory MSOFA score (OR, 2.43; 95% CI, 1.77-3.33; P < .001), high EDI (OR, 1.85; 95% CI, 1.41-2.41; P < .001), and high CRP level (OR, 2.06; 95% CI, 1.64-2.58; P < .001). English-speaking Hispanic patients also had a greater rate of high CRP level (OR, 1.48; 95% CI, 1.17-1.86; P = .001) compared to non-Hispanic patients. When considering the presentation of any one of these clinical indicators, the English-speaking and non-English-speaking Hispanic patients had a higher rate of severe baseline condition (OR, 1.33; 95% CI, 1.10-1.61; P = .004 for English-speaking patients; OR, 2.27; 95% CI, 1.89-2.72; P < .001 for non-English-speaking patients).

We then studied how the baseline disease condition affects the association between ethnicity and clinical outcomes. We performed a multivariable analysis including baseline disease severity as a covariable (Model 2, Table 2), which showed that baseline disease severity was strongly associated with ICU admission (OR, 4.52; 95% CI, 3.83-5.33; P < .001) and mortality (OR, 3.32; 95% CI, 2.67-4.13; P < .001). The associations between ethnicity and clinical outcomes were reduced after considering the baseline disease condition. The OR dropped to 1.47 (95% CI, 1.18-1.84; P < .001) and 1.34 (95% CI, 0.97-1.87; P = .08) for ICU utilization and mortality, respectively, when comparing non-English-speaking Hispanic patients to non-Hispanic patients. A similar reduction was observed for English-speaking Hispanic patients. Model comparison showed a significant improvement of Model 2 over Model 1 based on ANOVA test (P < .001) as well as AIC.

Hispanic Patients Had Worse Healthcare Access

To explore the etiology for the more severe disease conditions at hospital admission among Hispanic patients, we analyzed variables related to healthcare access. We found that Hispanic patients were likely to have reduced access to healthcare (Table 1; Appendix Figure 2). For a large proportion (16.9%) of the COVID-19 patients in this study, their medical records were first created at THR in 2020, corresponding to the initial time these patients accessed THR for their healthcare. This surge in 2020, compared to previous years with data (2005–2019), corresponds to the number of new patients seen because of COVID-19 (Appendix Figure 2A). Among this new patient population, the proportion of non-English-speaking Hispanic patients in 2020 was 28.3%, compared to 9.1% from 2005 to 2019 (P < .001). The proportion of new English-speaking Hispanic patients in 2020 was 22.1%, compared to an average of 19.2% from 2005 to 2019 (P < .001). In addition, a much smaller proportion of Hispanic patients had a PCP (P < .001) (Table 1; Appendix Figure 2B), with non-English-speaking Hispanic patients having the smallest proportion (58.5%).

Appendix Figure 2C illustrates the comparison of payor types across the three patient groups. A much higher proportion of Hispanic patients used COVID-19 HRSA (P < .001) compared to non-Hispanic patients. Breaking this down further by primary language, 29.1% of non-English-speaking Hispanic patients relied on COVID-19 HRSA due to otherwise uninsured status, compared to 12.7% of English-speaking Hispanic patients and only 5.1% of non-Hispanic patients. Similarly, non-English-speaking Hispanic patients have the highest self-pay rates (2.3%) compared to English-speaking Hispanic patients (1.4%) and non-Hispanic patients (0.7%). In summary, more Hispanic patients, and especially non-English-speaking Hispanic patients, lacked conventional health insurance and experienced limited access to healthcare.

Further evidence showed a trend of correlation between presentation of severe COVID-19 conditions when arriving at the hospital and each of the healthcare access factors analyzed (Appendix Figure 3).

With a large sample size of hospitalized COVID-19 patients at an integrated health system in the DFW metropolitan area, we observed an increased rate of ICU utilization and mortality among Hispanic inpatients. After adjusting for age, we found that non-English-speaking Hispanic patients were 75% more likely to require critical care compared with non-Hispanic patients. English-speaking and non-English-speaking Hispanic patients had an increased mortality rate (age-adjusted) compared to non-Hispanic patients. The association between ethnicity and clinical outcomes remained significant after adjusting for age, sex, and comorbidities. We did not observe any underuse of major COVID-19 therapeutics in Hispanic patients, and excluded in-hospital treatments from the contributors to the outcome differences.

Hispanic patients, especially non-English-speaking Hispanic patients, had a higher rate of severe COVID-19 disease at the time of hospital admission (Figure 2). After including baseline disease severity into the multivariable analysis (Model 2), the overall model improved (P < .001) while the associations between ethnicity and outcomes decreased (Table 2). This suggests disease severity at admission was a main contributor to the observed associations between ethnicity and clinical outcomes. The higher rate of baseline COVID-19 severity in Hispanic patients might also explain their higher rate of receiving major COVID-19 therapeutics (Appendix Figure 1).

This study found that Hispanic patients were less likely to have a PCP and insurance coverage compared with non-Hispanic patients (P < .001). This disparity was more pronounced among non-English-speaking Hispanic patients (Appendix Figure 2). We also observed that a disproportionately larger proportion (50.4%) of patients who visited the healthcare system for the first time in 2020 (the year of the COVID-19 pandemic) was composed of Hispanic patients, compared to merely 28.4% prior to 2020. While there is a possibility that patients had primary care outside THR, the staggering number of Hispanic patients who were new to the health system in 2020, in conjunction with the fact that immigrants tend to be “healthier” compared to their native-born peers (the so-called immigrant paradox),²⁸ led us to conclude that there were few other primary care options for these patients, making THR’s ED the primary care option of choice. The systemic, structural barriers to routine care might be a possible cause for delayed admission and, in turn, elevated baseline COVID-19 severity for Hispanic patients (Appendix Figure 3).

Recent studies have investigated the impact of socioeconomic factors on racial/ethnic disparities in the COVID-19 pandemic.^7,16,17 To our knowledge, no study has directly analyzed the link between healthcare access metrics, COVID-19 severity at admission, and the Hispanic population stratified by primary language. Studies exist on this subject for other diseases, however. For example, healthcare access factors have been associated with sepsis-related mortality.^29,30 In fact, a recent study that explored the potential effect of language barriers on healthcare access demonstrated an association between limited English proficiency and sepsis-related mortality.³¹ Our study found that Hispanic patients whose primary language is not English had the worst clinical outcomes, including more severe baseline COVID-19 conditions, and the least access to healthcare, highlighting the importance of addressing language barriers in COVID-19 care. Further research is needed to confirm the relationship between limited English proficiency and clinical outcomes, as well as potential factors that contribute to such a relationship in different types of diseases.

Our study has a number of limitations. First, it was limited to only one large healthcare system, which means the results may not be generalizable. Because THR is an open system, comorbidity data may be incomplete, and we cannot exclude the possibility that patients accessed care outside THR prior to or during the pandemic. We may overcome this limitation in the future with cross-system health information exchange data. Second, we did not have data for the time of symptom onset, so we were unable to analyze the direct evidence of the possible delayed care. As a result, we were unable to analyze whether treatments were administered in a timely manner or appropriately. Third, our analysis was not adjusted for other socioeconomic factors (eg, income, education) due to lack of data. We used self-identification for ethnicity, but unlike new approaches by the U.S. Census Bureau,³² our survey allowed only one choice to be selected.

Sociodemographic factors among Hispanic inpatients hospitalized for COVID-19 at a large integrated health system—including a primary non-English language, lack of a PCP, and insurance status—were associated with measures of reduced access to care and more severe illness at admission. Structural barriers to care, which may be associated with reduced health literacy and less access to health insurance, can result in delayed treatment and more severe illness at admission and underdiagnosis of medical conditions, contributing to worse outcomes in this population. Our findings suggest that interventions to promote early recognition of signs and symptoms of COVID-19 and to encourage prompt clinical care at the community level may reduce the burden of COVID-19 deaths in racial or ethnic minority communities with language and socioeconomic barriers.

In the United States, health disparities in COVID-19 outcomes (including morbidity and mortality) based on race and ethnicity have been described in the scientific literature and mainstream media.^1-7 According to the US Centers for Disease Control and Prevention (CDC), Hispanic people are 3.2 times more likely to be hospitalized with COVID-19 than non-Hispanic White people.⁸ Further, Hispanic people diagnosed with COVID-19 are 2.3 times more likely to die, adjusted for age, than non-Hispanic White people.⁹ As the epicenter of the COVID-19 pandemic shifted from the Northeast to the South, the CDC reported that, among people who died from COVID-19 in the United States from May to August 2020, the percentage of Hispanic people increased from 16.3% to 26.4%.¹⁰

Published studies on the effect of ethnicity on critical illness or mortality for hospitalized COVID-19 patients are limited and inconsistent. While some studies reported a higher mortality rate for Hispanic patients,^11-15 others showed no difference.^4,16,17 A recent meta-analysis found that intensive care unit (ICU) utilization and mortality were slightly higher among Hispanic COVID-19 inpatients, but this finding did not reach statistical significance.¹⁸ Past studies from different healthcare systems were limited by the small sample size of hospitalized Hispanic patients and the heterogeneity of patients. A comprehensive analysis from a large healthcare system with sufficient sample size is needed to understand the impact of ethnicity on clinical outcomes of hospitalized COVID-19 patients.

Texas Health Resources (THR) is a large integrated healthcare system serving the Dallas-Fort Worth-Arlington (DFW) metropolitan area. According to the 2019 US Census Bureau American Community Survey, Hispanic people comprise 18.4% of the population of this geographic area.¹⁹ Congruent with the CDC’s findings, Hispanic patients account for a disproportionate share (32.2%) of hospitalized COVID-19 patients at THR relative to the area’s demographic composition. Aware of the increased risk, we undertook an analysis of the clinical outcomes and the clinical, social, and demographic characteristics of Hispanic patients hospitalized at THR with COVID-19. Our primary goal was to investigate whether clinical outcomes differ by ethnicity among patients hospitalized with COVID-19 and, if so, whether inpatient care or preadmission factors contribute to this difference.

Study Setting and Overview

We collected data from the single electronic health record (EHR) used by 20 THR hospitals located across the DFW metropolitan area. THR is the largest faith-based, nonprofit health system in North Texas, operating 20 acute care hospitals. Including all access points, such as outpatient facilities and physician group practices, THR serves 7 million residents in 16 counties in North Texas, of whom 16.8% are Hispanic, 73.3% are non-Hispanic, and 9.9% are unclassified, congruent with demographics in the DFW area.

The institutional review boards at THR and UT Southwestern Medical Center approved the study under a waiver of informed consent (as a minimal-risk medical record review). After collection, all data were de-identified prior to statistical analysis.

Cohort, Outcomes, and Covariables

The study cohort included 6097 adult patients with laboratory-confirmed COVID-19 (age ≥18 years) who were admitted as inpatients from March 3 to November 5, 2020. The primary outcomes included ICU utilization and death during hospitalization. We described demographic characteristics using the following variables: age (18–49, 50–64, 65–79, ≥80 years), sex, self-reported ethnicity, and primary spoken language.

We defined a severe baseline condition as an elevated respiratory subscore parsed from the overall MSOFA (Modified Sequential Organ Failure Assessment),²⁰ an elevated Epic Deterioration Index (EDI),²¹ or an elevated C-reactive protein level (CRP) at baseline (any elevated CRP). Baseline referred to the variable mean during the first available 12-hour window of measurement during the COVID-19 hospital admission, including variables obtained in the emergency department (ED). An elevated MSOFA referred to a score of 4, corresponding to an SpO₂/FiO₂ < 150. Elevated EDI referred to a baseline EDI > 45. An elevated CRP referred to a baseline CRP > 20 mg/dL.²²

Variables reflecting access to healthcare included: THR EHR creation year (representing the first time patients accessed the THR health system), insurance payor type, and presence of a primary care provider (PCP). The federal government established the COVID-19 Claims Reimbursement to Health Care Providers and Facilities for Testing, Treatment, and Vaccine Administration for the Uninsured program. The insurance payor for patients covered by this program is designated as COVID-19 HRSA. Presence of a PCP reflects any documented PCP, regardless of affiliation with THR. We selected these access metrics opportunistically, as they were consistently documented in the EHR and readily available for analysis.

We used 12 variables to describe comorbidities or underlying conditions that, according to the CDC, increased patients’ risk of severe illness from COVID-19²³: diagnoses of diabetes, hypertension, obesity, chronic obstructive pulmonary disease (COPD), asthma, smoking, other lung disease, heart failure, kidney disease without end-stage renal disease (ESRD), ESRD, liver disease, and cancer. We identified comorbidities by mining the structured diagnosis codes documented in the EHR prior to and during the COVID-19 admission. Sources for diagnoses included final billed diagnosis codes, working diagnosis codes, problem list, and reason for visit. The definition of diabetes included previously recorded diabetes or baseline hemoglobin A1c > 9%. We also recorded the presence of four major COVID-19 treatments: steroids, remdesivir, tocilizumab, and fresh frozen plasma (FFP) from convalescent patients.^24-26 Each treatment variable was defined by receipt of one or more doses.

Statistical Analysis

To analyze patient outcomes based on ethnicity, we divided the study cohort into a Hispanic group and a non-Hispanic group based on self-reported ethnicity in the EHR. To study the potential impact of primary language among Hispanic patients, we divided them into English-speaking and non-English-speaking patients based on their self-reported primary language. As a result, we analyzed three groups of patients: (1) non-Hispanic, (2) Hispanic and English speaking, and (3) Hispanic and non-English speaking. We tested differences of a given categorical variable across the three groups using the chi-square test for each age subgroup (18–49, 50–64, 65–79, ≥80 years). The Cochran-Mantel-Haenszel test was used for the overall difference adjusted for age. To assess whether an observed disparity in treatment existed across the three groups, we tested the difference in the administration of four major therapeutics for COVID-19, including steroids, remdesivir, tocilizumab, and convalescent plasma. To determine whether any groups had elevated disease severity at hospital admission (baseline), we tested the difference in four disease-severity metrics across the ethnic-language groups: (1) elevated respiratory MSOFA score, (2) elevated EDI, (3) elevated CRP level, and (4) any of the three conditions.

To study the associations with ICU utilization and death, respectively, we performed a multivariable analysis using a generalized linear mixed model with binomial distribution and a logit link function. In each analysis model, the hospital of admission was included as a random-effect variable to account for the potential treatment variations among different hospitals, while other variables were regarded as fixed effects. In the first multivariable analysis (Model 1), all demographic variables, including age, sex, and ethnicity, and different types of comorbidities and underlying conditions, were included as fixed-effect variables in the initial model, and then backward stepwise variable selection was performed to establish the final model (Model 1). We performed the backward stepwise variable selection separately for the outcome of ICU use or mortality. Based on Akaike information criterion (AIC), during each iteration the fixed-effect variable that led to the largest decrease in the AIC value was removed, and the variable selection process was completed when the AIC value stopped decreasing. In Model 2, we added the disease-severity variable at baseline to the selected variable set derived from Model 1 to explore its effect on the associations between ethnicity and clinical outcomes. In Model 3, we added healthcare access–related variables, including first-time healthsystem access, payor type, and PCP availability to Model 2. We performed all statistical analyses using R, version 4.0.2 (R Foundation for Statistical Computing) in RStudio (version 1.3.1093).

Distinct Demographic and Comorbidity Patterns for Three Ethnic-Language Groups

We identified 6097 adult patients (age ≥18 years) who had confirmed COVID-19 disease and were hospitalized between March 3 and November 5, 2020. Demographic characteristics and comorbidity for these patients are summarized in Table 1. Among these patients, 4139 (67.9%) were non-Hispanic and 1958 (32.1%) were Hispanic. Among the Hispanic patients, 1203 (61.4%) identified English as their primary language and 755 (38.6%) identified a non-English primary language. Age distribution was vastly different among the three ethnic-language groups (Table 1). Unlike the relatively balanced distribution across different age groups in the non-Hispanic group, more than half (55.8%) of the English-speaking Hispanic patients were in the youngest age group (18-49 years). A much lower fraction of Hispanic patients was among the oldest (≥80 years) age group (P < .001). Because COVID-19 clinical outcome is strongly associated with age,²⁷ we used age-stratified analysis when comparing group-level differences in patient outcomes.

Sex distribution also was different among the three groups, with the non-English-speaking Hispanic group having more male patients (53.0%). Diabetes and obesity, which are associated with clinical outcomes of COVID-19 patients, were more prevalent in Hispanic patients (Table 1). Non-English-speaking Hispanic patients had the highest diabetes rate (48.7% with documented diabetes; 15.8% with baseline HbA1c > 9%; P < .001). English-speaking Hispanic patients presented with the highest obesity rate (62.8%; P < .001). Appendix Table 1 provides detailed age-group-specific comorbidity distributions among ethnic-language groups.

Patients of Hispanic Ethnicity Experienced a Higher Rate of ICU Utilization and Mortality

Of the 6097 patients overall, 1365 (22.4%) were admitted to the ICU and 543 (8.9%) died in hospital. For non-Hispanic patients (n = 4139), 883 (21.3%) were admitted to the ICU and 373 (9.0%) died in hospital. For English-speaking Hispanic patients (n = 1203), 241 (20.0%) were admitted to the ICU and 91 (7.6%) died in hospital. For non-English-speaking Hispanic patients (n = 755), 241 (31.9%) were admitted to the ICU and 79 (10.5%) died in hospital. Figure 1 summarizes the age-stratified comparison of ICU utilization and mortality across the three ethnic-language patient groups. In all age groups, non-English-speaking Hispanic patients experienced a significantly higher ICU utilization rate compared to non-Hispanic patients (age-adjusted OR, 1.75; 95% CI, 1.47-2.08; P < .001). English-speaking and non-English-speaking Hispanic patients had a significantly higher mortality rate compared to non-Hispanic patients (age-adjusted OR, 1.53; 95% CI, 1.19-1.98; P = .001 for English-speaking Hispanic patients; age-adjusted OR, 1.43; 95% CI,: 1.10-1.86; P = .01 for non-English-speaking Hispanic patients).

To delineate the risk factors associated with ICU utilization and death, we performed multivariable logistic regression with stepwise variable selection. After adjusting for age, sex, and comorbidity (Model 1), the factors ethnicity and primary language were still strongly associated with ICU utilization and mortality (Appendix Table 2). Non-English-speaking Hispanic patients had an OR of 1.74 (95% CI, 1.41-2.15; P < .001) for ICU utilization and an OR of 1.54 (95% CI, 1.12-2.12; P = .008) for mortality compared to non-Hispanic patients. Similarly, English-speaking Hispanic patients had higher ICU utilization (OR, 1.28; 95% CI, 1.05-1.55; P = .01) and a higher mortality rate (OR, 1.60; 95% CI, 1.19-2.14; P = .002).

No Disparity in COVID-19 Therapeutics Observed Across Three Ethnic-Language Groups

Appendix Figure 1 summarizes the comparison of the administration of four major treatments across the three ethnic-language groups. We did not observe any underuse of COVID-19 therapeutics for Hispanic patients. Usage rates for these therapies were significantly higher, after adjusting for age, in Hispanic groups when compared to non-Hispanic patients (OR ranged from 1.21 to 1.96). Steroids were the most common treatment in all patient groups. Tocilizumab was used almost twice as frequently (OR, 1.96; 95% CI, 1.64-2.33; P < .001) in non-English-speaking Hispanic patients compared to non-Hispanic patients.

Patients of Hispanic Ethnicity Had More Severe Disease at Hospital Admission

Figure 2 shows that non-English-speaking Hispanic patients had a higher rate of severe illness at admission based on each of these metrics: high respiratory MSOFA score (OR, 2.43; 95% CI, 1.77-3.33; P < .001), high EDI (OR, 1.85; 95% CI, 1.41-2.41; P < .001), and high CRP level (OR, 2.06; 95% CI, 1.64-2.58; P < .001). English-speaking Hispanic patients also had a greater rate of high CRP level (OR, 1.48; 95% CI, 1.17-1.86; P = .001) compared to non-Hispanic patients. When considering the presentation of any one of these clinical indicators, the English-speaking and non-English-speaking Hispanic patients had a higher rate of severe baseline condition (OR, 1.33; 95% CI, 1.10-1.61; P = .004 for English-speaking patients; OR, 2.27; 95% CI, 1.89-2.72; P < .001 for non-English-speaking patients).

We then studied how the baseline disease condition affects the association between ethnicity and clinical outcomes. We performed a multivariable analysis including baseline disease severity as a covariable (Model 2, Table 2), which showed that baseline disease severity was strongly associated with ICU admission (OR, 4.52; 95% CI, 3.83-5.33; P < .001) and mortality (OR, 3.32; 95% CI, 2.67-4.13; P < .001). The associations between ethnicity and clinical outcomes were reduced after considering the baseline disease condition. The OR dropped to 1.47 (95% CI, 1.18-1.84; P < .001) and 1.34 (95% CI, 0.97-1.87; P = .08) for ICU utilization and mortality, respectively, when comparing non-English-speaking Hispanic patients to non-Hispanic patients. A similar reduction was observed for English-speaking Hispanic patients. Model comparison showed a significant improvement of Model 2 over Model 1 based on ANOVA test (P < .001) as well as AIC.

Hispanic Patients Had Worse Healthcare Access

To explore the etiology for the more severe disease conditions at hospital admission among Hispanic patients, we analyzed variables related to healthcare access. We found that Hispanic patients were likely to have reduced access to healthcare (Table 1; Appendix Figure 2). For a large proportion (16.9%) of the COVID-19 patients in this study, their medical records were first created at THR in 2020, corresponding to the initial time these patients accessed THR for their healthcare. This surge in 2020, compared to previous years with data (2005–2019), corresponds to the number of new patients seen because of COVID-19 (Appendix Figure 2A). Among this new patient population, the proportion of non-English-speaking Hispanic patients in 2020 was 28.3%, compared to 9.1% from 2005 to 2019 (P < .001). The proportion of new English-speaking Hispanic patients in 2020 was 22.1%, compared to an average of 19.2% from 2005 to 2019 (P < .001). In addition, a much smaller proportion of Hispanic patients had a PCP (P < .001) (Table 1; Appendix Figure 2B), with non-English-speaking Hispanic patients having the smallest proportion (58.5%).

Appendix Figure 2C illustrates the comparison of payor types across the three patient groups. A much higher proportion of Hispanic patients used COVID-19 HRSA (P < .001) compared to non-Hispanic patients. Breaking this down further by primary language, 29.1% of non-English-speaking Hispanic patients relied on COVID-19 HRSA due to otherwise uninsured status, compared to 12.7% of English-speaking Hispanic patients and only 5.1% of non-Hispanic patients. Similarly, non-English-speaking Hispanic patients have the highest self-pay rates (2.3%) compared to English-speaking Hispanic patients (1.4%) and non-Hispanic patients (0.7%). In summary, more Hispanic patients, and especially non-English-speaking Hispanic patients, lacked conventional health insurance and experienced limited access to healthcare.

Further evidence showed a trend of correlation between presentation of severe COVID-19 conditions when arriving at the hospital and each of the healthcare access factors analyzed (Appendix Figure 3).

With a large sample size of hospitalized COVID-19 patients at an integrated health system in the DFW metropolitan area, we observed an increased rate of ICU utilization and mortality among Hispanic inpatients. After adjusting for age, we found that non-English-speaking Hispanic patients were 75% more likely to require critical care compared with non-Hispanic patients. English-speaking and non-English-speaking Hispanic patients had an increased mortality rate (age-adjusted) compared to non-Hispanic patients. The association between ethnicity and clinical outcomes remained significant after adjusting for age, sex, and comorbidities. We did not observe any underuse of major COVID-19 therapeutics in Hispanic patients, and excluded in-hospital treatments from the contributors to the outcome differences.

Hispanic patients, especially non-English-speaking Hispanic patients, had a higher rate of severe COVID-19 disease at the time of hospital admission (Figure 2). After including baseline disease severity into the multivariable analysis (Model 2), the overall model improved (P < .001) while the associations between ethnicity and outcomes decreased (Table 2). This suggests disease severity at admission was a main contributor to the observed associations between ethnicity and clinical outcomes. The higher rate of baseline COVID-19 severity in Hispanic patients might also explain their higher rate of receiving major COVID-19 therapeutics (Appendix Figure 1).

This study found that Hispanic patients were less likely to have a PCP and insurance coverage compared with non-Hispanic patients (P < .001). This disparity was more pronounced among non-English-speaking Hispanic patients (Appendix Figure 2). We also observed that a disproportionately larger proportion (50.4%) of patients who visited the healthcare system for the first time in 2020 (the year of the COVID-19 pandemic) was composed of Hispanic patients, compared to merely 28.4% prior to 2020. While there is a possibility that patients had primary care outside THR, the staggering number of Hispanic patients who were new to the health system in 2020, in conjunction with the fact that immigrants tend to be “healthier” compared to their native-born peers (the so-called immigrant paradox),²⁸ led us to conclude that there were few other primary care options for these patients, making THR’s ED the primary care option of choice. The systemic, structural barriers to routine care might be a possible cause for delayed admission and, in turn, elevated baseline COVID-19 severity for Hispanic patients (Appendix Figure 3).

Recent studies have investigated the impact of socioeconomic factors on racial/ethnic disparities in the COVID-19 pandemic.^7,16,17 To our knowledge, no study has directly analyzed the link between healthcare access metrics, COVID-19 severity at admission, and the Hispanic population stratified by primary language. Studies exist on this subject for other diseases, however. For example, healthcare access factors have been associated with sepsis-related mortality.^29,30 In fact, a recent study that explored the potential effect of language barriers on healthcare access demonstrated an association between limited English proficiency and sepsis-related mortality.³¹ Our study found that Hispanic patients whose primary language is not English had the worst clinical outcomes, including more severe baseline COVID-19 conditions, and the least access to healthcare, highlighting the importance of addressing language barriers in COVID-19 care. Further research is needed to confirm the relationship between limited English proficiency and clinical outcomes, as well as potential factors that contribute to such a relationship in different types of diseases.

Our study has a number of limitations. First, it was limited to only one large healthcare system, which means the results may not be generalizable. Because THR is an open system, comorbidity data may be incomplete, and we cannot exclude the possibility that patients accessed care outside THR prior to or during the pandemic. We may overcome this limitation in the future with cross-system health information exchange data. Second, we did not have data for the time of symptom onset, so we were unable to analyze the direct evidence of the possible delayed care. As a result, we were unable to analyze whether treatments were administered in a timely manner or appropriately. Third, our analysis was not adjusted for other socioeconomic factors (eg, income, education) due to lack of data. We used self-identification for ethnicity, but unlike new approaches by the U.S. Census Bureau,³² our survey allowed only one choice to be selected.

Sociodemographic factors among Hispanic inpatients hospitalized for COVID-19 at a large integrated health system—including a primary non-English language, lack of a PCP, and insurance status—were associated with measures of reduced access to care and more severe illness at admission. Structural barriers to care, which may be associated with reduced health literacy and less access to health insurance, can result in delayed treatment and more severe illness at admission and underdiagnosis of medical conditions, contributing to worse outcomes in this population. Our findings suggest that interventions to promote early recognition of signs and symptoms of COVID-19 and to encourage prompt clinical care at the community level may reduce the burden of COVID-19 deaths in racial or ethnic minority communities with language and socioeconomic barriers.

In the United States, health disparities in COVID-19 outcomes (including morbidity and mortality) based on race and ethnicity have been described in the scientific literature and mainstream media.^1-7 According to the US Centers for Disease Control and Prevention (CDC), Hispanic people are 3.2 times more likely to be hospitalized with COVID-19 than non-Hispanic White people.⁸ Further, Hispanic people diagnosed with COVID-19 are 2.3 times more likely to die, adjusted for age, than non-Hispanic White people.⁹ As the epicenter of the COVID-19 pandemic shifted from the Northeast to the South, the CDC reported that, among people who died from COVID-19 in the United States from May to August 2020, the percentage of Hispanic people increased from 16.3% to 26.4%.¹⁰

Published studies on the effect of ethnicity on critical illness or mortality for hospitalized COVID-19 patients are limited and inconsistent. While some studies reported a higher mortality rate for Hispanic patients,^11-15 others showed no difference.^4,16,17 A recent meta-analysis found that intensive care unit (ICU) utilization and mortality were slightly higher among Hispanic COVID-19 inpatients, but this finding did not reach statistical significance.¹⁸ Past studies from different healthcare systems were limited by the small sample size of hospitalized Hispanic patients and the heterogeneity of patients. A comprehensive analysis from a large healthcare system with sufficient sample size is needed to understand the impact of ethnicity on clinical outcomes of hospitalized COVID-19 patients.

Texas Health Resources (THR) is a large integrated healthcare system serving the Dallas-Fort Worth-Arlington (DFW) metropolitan area. According to the 2019 US Census Bureau American Community Survey, Hispanic people comprise 18.4% of the population of this geographic area.¹⁹ Congruent with the CDC’s findings, Hispanic patients account for a disproportionate share (32.2%) of hospitalized COVID-19 patients at THR relative to the area’s demographic composition. Aware of the increased risk, we undertook an analysis of the clinical outcomes and the clinical, social, and demographic characteristics of Hispanic patients hospitalized at THR with COVID-19. Our primary goal was to investigate whether clinical outcomes differ by ethnicity among patients hospitalized with COVID-19 and, if so, whether inpatient care or preadmission factors contribute to this difference.

Study Setting and Overview

We collected data from the single electronic health record (EHR) used by 20 THR hospitals located across the DFW metropolitan area. THR is the largest faith-based, nonprofit health system in North Texas, operating 20 acute care hospitals. Including all access points, such as outpatient facilities and physician group practices, THR serves 7 million residents in 16 counties in North Texas, of whom 16.8% are Hispanic, 73.3% are non-Hispanic, and 9.9% are unclassified, congruent with demographics in the DFW area.

The institutional review boards at THR and UT Southwestern Medical Center approved the study under a waiver of informed consent (as a minimal-risk medical record review). After collection, all data were de-identified prior to statistical analysis.

Cohort, Outcomes, and Covariables

The study cohort included 6097 adult patients with laboratory-confirmed COVID-19 (age ≥18 years) who were admitted as inpatients from March 3 to November 5, 2020. The primary outcomes included ICU utilization and death during hospitalization. We described demographic characteristics using the following variables: age (18–49, 50–64, 65–79, ≥80 years), sex, self-reported ethnicity, and primary spoken language.

We defined a severe baseline condition as an elevated respiratory subscore parsed from the overall MSOFA (Modified Sequential Organ Failure Assessment),²⁰ an elevated Epic Deterioration Index (EDI),²¹ or an elevated C-reactive protein level (CRP) at baseline (any elevated CRP). Baseline referred to the variable mean during the first available 12-hour window of measurement during the COVID-19 hospital admission, including variables obtained in the emergency department (ED). An elevated MSOFA referred to a score of 4, corresponding to an SpO₂/FiO₂ < 150. Elevated EDI referred to a baseline EDI > 45. An elevated CRP referred to a baseline CRP > 20 mg/dL.²²

Variables reflecting access to healthcare included: THR EHR creation year (representing the first time patients accessed the THR health system), insurance payor type, and presence of a primary care provider (PCP). The federal government established the COVID-19 Claims Reimbursement to Health Care Providers and Facilities for Testing, Treatment, and Vaccine Administration for the Uninsured program. The insurance payor for patients covered by this program is designated as COVID-19 HRSA. Presence of a PCP reflects any documented PCP, regardless of affiliation with THR. We selected these access metrics opportunistically, as they were consistently documented in the EHR and readily available for analysis.

We used 12 variables to describe comorbidities or underlying conditions that, according to the CDC, increased patients’ risk of severe illness from COVID-19²³: diagnoses of diabetes, hypertension, obesity, chronic obstructive pulmonary disease (COPD), asthma, smoking, other lung disease, heart failure, kidney disease without end-stage renal disease (ESRD), ESRD, liver disease, and cancer. We identified comorbidities by mining the structured diagnosis codes documented in the EHR prior to and during the COVID-19 admission. Sources for diagnoses included final billed diagnosis codes, working diagnosis codes, problem list, and reason for visit. The definition of diabetes included previously recorded diabetes or baseline hemoglobin A1c > 9%. We also recorded the presence of four major COVID-19 treatments: steroids, remdesivir, tocilizumab, and fresh frozen plasma (FFP) from convalescent patients.^24-26 Each treatment variable was defined by receipt of one or more doses.

Statistical Analysis

To analyze patient outcomes based on ethnicity, we divided the study cohort into a Hispanic group and a non-Hispanic group based on self-reported ethnicity in the EHR. To study the potential impact of primary language among Hispanic patients, we divided them into English-speaking and non-English-speaking patients based on their self-reported primary language. As a result, we analyzed three groups of patients: (1) non-Hispanic, (2) Hispanic and English speaking, and (3) Hispanic and non-English speaking. We tested differences of a given categorical variable across the three groups using the chi-square test for each age subgroup (18–49, 50–64, 65–79, ≥80 years). The Cochran-Mantel-Haenszel test was used for the overall difference adjusted for age. To assess whether an observed disparity in treatment existed across the three groups, we tested the difference in the administration of four major therapeutics for COVID-19, including steroids, remdesivir, tocilizumab, and convalescent plasma. To determine whether any groups had elevated disease severity at hospital admission (baseline), we tested the difference in four disease-severity metrics across the ethnic-language groups: (1) elevated respiratory MSOFA score, (2) elevated EDI, (3) elevated CRP level, and (4) any of the three conditions.

To study the associations with ICU utilization and death, respectively, we performed a multivariable analysis using a generalized linear mixed model with binomial distribution and a logit link function. In each analysis model, the hospital of admission was included as a random-effect variable to account for the potential treatment variations among different hospitals, while other variables were regarded as fixed effects. In the first multivariable analysis (Model 1), all demographic variables, including age, sex, and ethnicity, and different types of comorbidities and underlying conditions, were included as fixed-effect variables in the initial model, and then backward stepwise variable selection was performed to establish the final model (Model 1). We performed the backward stepwise variable selection separately for the outcome of ICU use or mortality. Based on Akaike information criterion (AIC), during each iteration the fixed-effect variable that led to the largest decrease in the AIC value was removed, and the variable selection process was completed when the AIC value stopped decreasing. In Model 2, we added the disease-severity variable at baseline to the selected variable set derived from Model 1 to explore its effect on the associations between ethnicity and clinical outcomes. In Model 3, we added healthcare access–related variables, including first-time healthsystem access, payor type, and PCP availability to Model 2. We performed all statistical analyses using R, version 4.0.2 (R Foundation for Statistical Computing) in RStudio (version 1.3.1093).

Distinct Demographic and Comorbidity Patterns for Three Ethnic-Language Groups

We identified 6097 adult patients (age ≥18 years) who had confirmed COVID-19 disease and were hospitalized between March 3 and November 5, 2020. Demographic characteristics and comorbidity for these patients are summarized in Table 1. Among these patients, 4139 (67.9%) were non-Hispanic and 1958 (32.1%) were Hispanic. Among the Hispanic patients, 1203 (61.4%) identified English as their primary language and 755 (38.6%) identified a non-English primary language. Age distribution was vastly different among the three ethnic-language groups (Table 1). Unlike the relatively balanced distribution across different age groups in the non-Hispanic group, more than half (55.8%) of the English-speaking Hispanic patients were in the youngest age group (18-49 years). A much lower fraction of Hispanic patients was among the oldest (≥80 years) age group (P < .001). Because COVID-19 clinical outcome is strongly associated with age,²⁷ we used age-stratified analysis when comparing group-level differences in patient outcomes.

Sex distribution also was different among the three groups, with the non-English-speaking Hispanic group having more male patients (53.0%). Diabetes and obesity, which are associated with clinical outcomes of COVID-19 patients, were more prevalent in Hispanic patients (Table 1). Non-English-speaking Hispanic patients had the highest diabetes rate (48.7% with documented diabetes; 15.8% with baseline HbA1c > 9%; P < .001). English-speaking Hispanic patients presented with the highest obesity rate (62.8%; P < .001). Appendix Table 1 provides detailed age-group-specific comorbidity distributions among ethnic-language groups.

Patients of Hispanic Ethnicity Experienced a Higher Rate of ICU Utilization and Mortality

Of the 6097 patients overall, 1365 (22.4%) were admitted to the ICU and 543 (8.9%) died in hospital. For non-Hispanic patients (n = 4139), 883 (21.3%) were admitted to the ICU and 373 (9.0%) died in hospital. For English-speaking Hispanic patients (n = 1203), 241 (20.0%) were admitted to the ICU and 91 (7.6%) died in hospital. For non-English-speaking Hispanic patients (n = 755), 241 (31.9%) were admitted to the ICU and 79 (10.5%) died in hospital. Figure 1 summarizes the age-stratified comparison of ICU utilization and mortality across the three ethnic-language patient groups. In all age groups, non-English-speaking Hispanic patients experienced a significantly higher ICU utilization rate compared to non-Hispanic patients (age-adjusted OR, 1.75; 95% CI, 1.47-2.08; P < .001). English-speaking and non-English-speaking Hispanic patients had a significantly higher mortality rate compared to non-Hispanic patients (age-adjusted OR, 1.53; 95% CI, 1.19-1.98; P = .001 for English-speaking Hispanic patients; age-adjusted OR, 1.43; 95% CI,: 1.10-1.86; P = .01 for non-English-speaking Hispanic patients).

To delineate the risk factors associated with ICU utilization and death, we performed multivariable logistic regression with stepwise variable selection. After adjusting for age, sex, and comorbidity (Model 1), the factors ethnicity and primary language were still strongly associated with ICU utilization and mortality (Appendix Table 2). Non-English-speaking Hispanic patients had an OR of 1.74 (95% CI, 1.41-2.15; P < .001) for ICU utilization and an OR of 1.54 (95% CI, 1.12-2.12; P = .008) for mortality compared to non-Hispanic patients. Similarly, English-speaking Hispanic patients had higher ICU utilization (OR, 1.28; 95% CI, 1.05-1.55; P = .01) and a higher mortality rate (OR, 1.60; 95% CI, 1.19-2.14; P = .002).

No Disparity in COVID-19 Therapeutics Observed Across Three Ethnic-Language Groups

Appendix Figure 1 summarizes the comparison of the administration of four major treatments across the three ethnic-language groups. We did not observe any underuse of COVID-19 therapeutics for Hispanic patients. Usage rates for these therapies were significantly higher, after adjusting for age, in Hispanic groups when compared to non-Hispanic patients (OR ranged from 1.21 to 1.96). Steroids were the most common treatment in all patient groups. Tocilizumab was used almost twice as frequently (OR, 1.96; 95% CI, 1.64-2.33; P < .001) in non-English-speaking Hispanic patients compared to non-Hispanic patients.

Patients of Hispanic Ethnicity Had More Severe Disease at Hospital Admission

Figure 2 shows that non-English-speaking Hispanic patients had a higher rate of severe illness at admission based on each of these metrics: high respiratory MSOFA score (OR, 2.43; 95% CI, 1.77-3.33; P < .001), high EDI (OR, 1.85; 95% CI, 1.41-2.41; P < .001), and high CRP level (OR, 2.06; 95% CI, 1.64-2.58; P < .001). English-speaking Hispanic patients also had a greater rate of high CRP level (OR, 1.48; 95% CI, 1.17-1.86; P = .001) compared to non-Hispanic patients. When considering the presentation of any one of these clinical indicators, the English-speaking and non-English-speaking Hispanic patients had a higher rate of severe baseline condition (OR, 1.33; 95% CI, 1.10-1.61; P = .004 for English-speaking patients; OR, 2.27; 95% CI, 1.89-2.72; P < .001 for non-English-speaking patients).

We then studied how the baseline disease condition affects the association between ethnicity and clinical outcomes. We performed a multivariable analysis including baseline disease severity as a covariable (Model 2, Table 2), which showed that baseline disease severity was strongly associated with ICU admission (OR, 4.52; 95% CI, 3.83-5.33; P < .001) and mortality (OR, 3.32; 95% CI, 2.67-4.13; P < .001). The associations between ethnicity and clinical outcomes were reduced after considering the baseline disease condition. The OR dropped to 1.47 (95% CI, 1.18-1.84; P < .001) and 1.34 (95% CI, 0.97-1.87; P = .08) for ICU utilization and mortality, respectively, when comparing non-English-speaking Hispanic patients to non-Hispanic patients. A similar reduction was observed for English-speaking Hispanic patients. Model comparison showed a significant improvement of Model 2 over Model 1 based on ANOVA test (P < .001) as well as AIC.

Hispanic Patients Had Worse Healthcare Access

To explore the etiology for the more severe disease conditions at hospital admission among Hispanic patients, we analyzed variables related to healthcare access. We found that Hispanic patients were likely to have reduced access to healthcare (Table 1; Appendix Figure 2). For a large proportion (16.9%) of the COVID-19 patients in this study, their medical records were first created at THR in 2020, corresponding to the initial time these patients accessed THR for their healthcare. This surge in 2020, compared to previous years with data (2005–2019), corresponds to the number of new patients seen because of COVID-19 (Appendix Figure 2A). Among this new patient population, the proportion of non-English-speaking Hispanic patients in 2020 was 28.3%, compared to 9.1% from 2005 to 2019 (P < .001). The proportion of new English-speaking Hispanic patients in 2020 was 22.1%, compared to an average of 19.2% from 2005 to 2019 (P < .001). In addition, a much smaller proportion of Hispanic patients had a PCP (P < .001) (Table 1; Appendix Figure 2B), with non-English-speaking Hispanic patients having the smallest proportion (58.5%).

Appendix Figure 2C illustrates the comparison of payor types across the three patient groups. A much higher proportion of Hispanic patients used COVID-19 HRSA (P < .001) compared to non-Hispanic patients. Breaking this down further by primary language, 29.1% of non-English-speaking Hispanic patients relied on COVID-19 HRSA due to otherwise uninsured status, compared to 12.7% of English-speaking Hispanic patients and only 5.1% of non-Hispanic patients. Similarly, non-English-speaking Hispanic patients have the highest self-pay rates (2.3%) compared to English-speaking Hispanic patients (1.4%) and non-Hispanic patients (0.7%). In summary, more Hispanic patients, and especially non-English-speaking Hispanic patients, lacked conventional health insurance and experienced limited access to healthcare.

Further evidence showed a trend of correlation between presentation of severe COVID-19 conditions when arriving at the hospital and each of the healthcare access factors analyzed (Appendix Figure 3).

With a large sample size of hospitalized COVID-19 patients at an integrated health system in the DFW metropolitan area, we observed an increased rate of ICU utilization and mortality among Hispanic inpatients. After adjusting for age, we found that non-English-speaking Hispanic patients were 75% more likely to require critical care compared with non-Hispanic patients. English-speaking and non-English-speaking Hispanic patients had an increased mortality rate (age-adjusted) compared to non-Hispanic patients. The association between ethnicity and clinical outcomes remained significant after adjusting for age, sex, and comorbidities. We did not observe any underuse of major COVID-19 therapeutics in Hispanic patients, and excluded in-hospital treatments from the contributors to the outcome differences.

Hispanic patients, especially non-English-speaking Hispanic patients, had a higher rate of severe COVID-19 disease at the time of hospital admission (Figure 2). After including baseline disease severity into the multivariable analysis (Model 2), the overall model improved (P < .001) while the associations between ethnicity and outcomes decreased (Table 2). This suggests disease severity at admission was a main contributor to the observed associations between ethnicity and clinical outcomes. The higher rate of baseline COVID-19 severity in Hispanic patients might also explain their higher rate of receiving major COVID-19 therapeutics (Appendix Figure 1).

This study found that Hispanic patients were less likely to have a PCP and insurance coverage compared with non-Hispanic patients (P < .001). This disparity was more pronounced among non-English-speaking Hispanic patients (Appendix Figure 2). We also observed that a disproportionately larger proportion (50.4%) of patients who visited the healthcare system for the first time in 2020 (the year of the COVID-19 pandemic) was composed of Hispanic patients, compared to merely 28.4% prior to 2020. While there is a possibility that patients had primary care outside THR, the staggering number of Hispanic patients who were new to the health system in 2020, in conjunction with the fact that immigrants tend to be “healthier” compared to their native-born peers (the so-called immigrant paradox),²⁸ led us to conclude that there were few other primary care options for these patients, making THR’s ED the primary care option of choice. The systemic, structural barriers to routine care might be a possible cause for delayed admission and, in turn, elevated baseline COVID-19 severity for Hispanic patients (Appendix Figure 3).

Recent studies have investigated the impact of socioeconomic factors on racial/ethnic disparities in the COVID-19 pandemic.^7,16,17 To our knowledge, no study has directly analyzed the link between healthcare access metrics, COVID-19 severity at admission, and the Hispanic population stratified by primary language. Studies exist on this subject for other diseases, however. For example, healthcare access factors have been associated with sepsis-related mortality.^29,30 In fact, a recent study that explored the potential effect of language barriers on healthcare access demonstrated an association between limited English proficiency and sepsis-related mortality.³¹ Our study found that Hispanic patients whose primary language is not English had the worst clinical outcomes, including more severe baseline COVID-19 conditions, and the least access to healthcare, highlighting the importance of addressing language barriers in COVID-19 care. Further research is needed to confirm the relationship between limited English proficiency and clinical outcomes, as well as potential factors that contribute to such a relationship in different types of diseases.

Our study has a number of limitations. First, it was limited to only one large healthcare system, which means the results may not be generalizable. Because THR is an open system, comorbidity data may be incomplete, and we cannot exclude the possibility that patients accessed care outside THR prior to or during the pandemic. We may overcome this limitation in the future with cross-system health information exchange data. Second, we did not have data for the time of symptom onset, so we were unable to analyze the direct evidence of the possible delayed care. As a result, we were unable to analyze whether treatments were administered in a timely manner or appropriately. Third, our analysis was not adjusted for other socioeconomic factors (eg, income, education) due to lack of data. We used self-identification for ethnicity, but unlike new approaches by the U.S. Census Bureau,³² our survey allowed only one choice to be selected.

Sociodemographic factors among Hispanic inpatients hospitalized for COVID-19 at a large integrated health system—including a primary non-English language, lack of a PCP, and insurance status—were associated with measures of reduced access to care and more severe illness at admission. Structural barriers to care, which may be associated with reduced health literacy and less access to health insurance, can result in delayed treatment and more severe illness at admission and underdiagnosis of medical conditions, contributing to worse outcomes in this population. Our findings suggest that interventions to promote early recognition of signs and symptoms of COVID-19 and to encourage prompt clinical care at the community level may reduce the burden of COVID-19 deaths in racial or ethnic minority communities with language and socioeconomic barriers.

The practice of internal medicine includes bedside procedures such as paracentesis, thoracentesis, and lumbar puncture (LP). The American Board of Internal Medicine requires graduates of internal medicine residency programs to be competent in the cognitive components of procedural training (eg, indications, contraindications, complications) and considers it essential that trainees have opportunities to perform procedures relevant to their intended career direction.¹ Whether or not the performance of procedures is part of a given hospitalist’s practice, it is necessary that hospitalists understand each procedure’s risks and mitigation strategies to prevent a range of periprocedural complications, including clinically significant bleeding. Numerous recommendations and guidelines exist describing bleeding risk for common procedures. In this Progress Note, we summarize and consolidate this literature, covering a range of scenarios common to the hospital setting, including thrombocytopenia, elevated international normalized ratio (INR), and the use of medications such as antiplatelet and anticoagulant agents (Table 1 and Table 2). We performed electronic searches in PubMed, focusing on literature published since 2016. Key search terms included paracentesis, thoracentesis, lumbar puncture, anticoagulant, antiplatelet, coagulopathy, INR, thrombocytopenia, and guideline. In addition, we used the following MeSH terms: spinal puncture AND blood coagulation disorders, spinal puncture AND platelet aggregation inhibitors, spinal puncture AND anticoagulants, paracentesis AND blood coagulation disorders, paracentesis AND platelet aggregation inhibitors, paracentesis AND anticoagulants, thoracentesis AND blood coagulation disorders, thoracentesis AND platelet aggregation inhibitors, and thoracentesis AND anticoagulants.

Weighing Risks and Benefits

Proceduralists should discuss risks and benefits with patients and the referring service before attempting to mitigate bleeding risk by holding antithrombotic agents or reversing coagulopathy, as these actions come with risks outside the anticipated procedure, in particular, an increased risk of thrombosis. There are many factors that influence an individual patient’s arterial thromboembolism and venous thromboembolism (VTE) risk, including surgical history, genetics, comorbidities, and the underlying indication for antithrombotic therapy. The American College of Chest Physicians updated their clinical practice guidelines describing perioperative thromboembolism risk stratification in 2012.¹² In general, higher-risk individuals include those with any mitral valve prosthesis, recent (within 6 months) stroke or transient ischemic attack (TIA), CHADS₂ score of 5 or 6, recent (within 3 months) VTE, or severe thrombophilia. Individuals at moderate risk include those with bileaflet aortic valve repair (AVR) with at least one major stroke risk factor, CHADS₂ score of 3 or 4, recent (within 3-12 months) or recurrent VTE, or active or recent (treatment within preceding 6 months) cancer. Finally, individuals at low risk include those with bileaflet AVR without major stroke risk factors, CHADS₂ score of 0 to 2, or VTE more than 12 months earlier. There are also patient-specific bleeding risk factors that should be considered, including hypertension, abnormal renal function, abnormal hepatic function, prior stroke, history of major bleeding (especially within the preceding 3 months), and bleeding history with a similar procedure.¹¹

Hepatic and Renal Dysfunction

In the setting of chronic liver disease, thrombocytopenia and elevated INR are generally not reliable indicators of bleeding risk.¹³ The included recommendations for INR and platelet count thresholds in the setting of chronic liver disease are derived from the referenced guidelines and supplemental personal communication with the guideline authors. Many antiplatelet and anticoagulant medications are partially cleared or metabolized by the liver, suggesting that hepatic dysfunction may impact drug clearance, but this has not been well studied. Impaired renal function should also be considered when determining appropriate hold times for antithrombotic drugs that are partially renally cleared. The periprocedural hold and restart times outlined in Table 2 are specific to patients without clinically significant hepatic or renal dysfunction. For patients with these conditions, further information on hold time adjustment can be found in the individual references.

Bridging Therapy

The decision to use bridging therapy prior to a bedside procedure must be individualized and take into account patient-specific factors. However, there is mounting evidence that bridging therapy is associated with higher risk of bleeding with no difference in the risk of thromboembolic events.^2,3,14 If the decision has been made to use bridging therapy with a heparin infusion prior to a bedside procedure, recommendations for hold and restart times can be found in Table 2.

Resuming Therapy

Another key consideration for procedures, especially those associated with a higher risk of bleeding, such as LP, is when to restart medications that have been held prior to the procedure. Table 2 provides a summary of the recommended postprocedural restart times for a variety of agents.

Other Considerations

Some guidelines referenced in this article are based on data collected on procedures performed by interventional radiologists, which may or may not accurately reflect the bleeding risks of bedside procedures performed by hospitalists. In the case of LP, we included some regional anesthesia and pain procedure guidelines based on the assumption that certain procedures are analogous to LP and associated with similar bleeding risks. Some of the guidelines referenced do not provide specific periprocedural INR and platelet thresholds (reported as “No threshold” in Table 1), instead offering statements that elevated INR and thrombocytopenia are not contraindications to bedside procedures and periprocedural transfusion of blood products is generally not recommended, based on the overall low risk of bleeding and lack of evidence for the efficacy of interventions intended to improve INR values and platelet counts in these situations. Patients undergoing paracentesis, thoracentesis, or LP may be on multiple antithrombotic agents, such as dual antiplatelet therapy. There are limited guidelines and studies on how to manage these agents in the periprocedural context; however, one guideline recommends continuing dual antiplatelet therapy for paracentesis, thoracentesis, and LP in patients who have cardiac stents.² There are also limited guidelines on how to handle patients on simultaneous antiplatelet and anticoagulant therapy.

Paracentesis is a common procedure that can be performed safely at the bedside. The overall rate of serious complications is low (1%-2%), with severe hemorrhage accounting for the majority of those complications (0.97%).¹⁵ Bleeding usually occurs from puncture of an abdominal wall vein, a mesenteric varix, or an inferior epigastric artery. Certain techniques may help to mitigate serious bleeding, including the use of ultrasound to avoid overlying vessels. Paracentesis is frequently performed in patients with cirrhosis, a population at increased risk for coagulopathy, although INR and platelet counts may not reflect aggregate bleeding risk in patients with cirrhosis. The American Association for the Study of Liver Diseases released new guidelines in 2021, stating that elevated prothrombin time or thrombocytopenia is not a contraindication to paracentesis.⁶ The most liberal guidelines for patients without chronic liver disease suggest correcting to an INR of 2.0 to 3.0, with multiple societies suggesting that a platelet count as low as 20,000/µL is safe.^2,3 As shown in Table 2, most guidelines recommend continuation of antiplatelet agents such as aspirin and thienopyridines (eg, clopidogrel, prasugrel), whereas recommendations vary regarding continuation of anticoagulant agents.

Akin to paracentesis, thoracentesis is generally considered to be a safe bedside procedure, with an incidence of thoracentesis-associated bleeding of less than 1%.¹⁵ Certain techniques may help to mitigate serious bleeding, including the insertion of the needle over the superior aspect of the rib in an effort to avoid the intercostal neurovascular bundle, which runs along the inferior aspect of each rib. Various clinical societies have proposed INR and platelet thresholds at which the risk of bleeding from thoracentesis is thought to be acceptable. The most liberal guidelines include a target INR of 2.0 to 3.0, although one group recommended an INR of <1.5.^2,5 Thoracentesis is commonly performed in patients with cirrhosis who develop hepatic hydrothorax. In this population, the Society of Interventional Radiology (SIR) guidelines state that there is no INR threshold that necessitates reversal strategies prior to the procedure.² For platelet count, there are multiple recommendations for greater than 20,000/µL and one for greater than 50,000/µL.^2,3,5 The recommendations for continuation or suspension of antiplatelet and anticoagulant medications prior to thoracentesis are similar to those for paracentesis. In general, continuing antiplatelet agents is felt to be safe, whereas there are mixed recommendations for anticoagulants, as described further in Table 2.

Compared to thoracentesis and paracentesis, LP is generally considered to be a higher-risk procedure owing to the rare possibility of spinal hematoma with associated neurologic compromise. In one retrospective review of more than 49,000 patients without coagulopathy who underwent LP, the risk for developing a spinal hematoma by 30 days post procedure was 0.20%.¹⁶ Certain techniques may help to mitigate serious bleeding, including the use of image guidance in patients with large body habitus or those with difficult anatomy. Compared with paracentesis and thoracentesis, guideline recommendations for safe INR and platelet thresholds in patients undergoing LP are based on a more limited body of evidence. Guidelines also suggest a target INR of anywhere from ≤1.5 to the most liberal suggestion of 2.0 to 3.0.^2-4 The SIR guidelines categorize LP as a low–bleeding risk procedure, with a platelet threshold of 20,000/µL but note that most other societies and guidelines regard LP as a high–bleeding risk procedure with more conservative platelet thresholds.² The Association of British Neurologists (ABN), however, allows platelets to be 40,000/µL or greater than 20,000/µL with an additional risk-benefit discussion.⁷ In contrast to paracentesis and thoracentesis, recommendations regarding hold times of antithrombotic medications prior to LP are more variable and sometimes more conservative. For example, some guidelines indicate that the thienopyridines can be continued, whereas others recommend holding them for up to 1 week prior to LP.^2,4,7

A theme throughout the recent literature and recommendations from clinical societies is that it is uncommon for there to be one unifying recommendation for every situation, especially regarding LP. Recent guidelines remain largely based on studies that are decades old. With bedside ultrasound becoming more accessible and established in daily practice, the risk of bleeding has been decreasing, potentially making periprocedural coagulopathies and antithrombotic agents less of a concern. For example, in a retrospective study of 69,859 paracenteses, ultrasound guidance reduced the risk of bleeding complications by 68%, an odds ratio of 0.32 (95% CI, 0.25-0.41).¹⁷ More research is needed to assess procedural bleeding risks in the context of current practice standards. This article focuses on a subset of bedside procedures most commonly performed by hospitalists. Similar references for other common bedside procedures, such as arthrocentesis, central venous catheter, and arterial line placement, would be helpful. Finally, this article does not capture such nuances as needle gauge, operator experience, availability of (and comfort with) ultrasound, and variations in patient anatomy, all of which are factors that can contribute to the complexities and risks of these bedside procedures.

Although not every internal medicine physician performs bedside procedures in their practice, it is vital that all understand the cognitive aspects of common bedside procedures. This necessitates the understanding of periprocedural risks and possible complications and applying that to individual patients. Correcting coagulopathy and stopping or reversing antithrombotic agents are mitigation strategies that are associated with risk. It is therefore important to understand when coagulopathy should be corrected and when antithrombotic agents should be held and for how long. With multiple existing and sometimes conflicting guidelines regarding periprocedural management of coagulopathy and antithrombotic agents, we hope that providing consolidated tables with this information will increase efficiency, aid in risk-benefit discussions between patients and care teams, and enhance patient safety.

The practice of internal medicine includes bedside procedures such as paracentesis, thoracentesis, and lumbar puncture (LP). The American Board of Internal Medicine requires graduates of internal medicine residency programs to be competent in the cognitive components of procedural training (eg, indications, contraindications, complications) and considers it essential that trainees have opportunities to perform procedures relevant to their intended career direction.¹ Whether or not the performance of procedures is part of a given hospitalist’s practice, it is necessary that hospitalists understand each procedure’s risks and mitigation strategies to prevent a range of periprocedural complications, including clinically significant bleeding. Numerous recommendations and guidelines exist describing bleeding risk for common procedures. In this Progress Note, we summarize and consolidate this literature, covering a range of scenarios common to the hospital setting, including thrombocytopenia, elevated international normalized ratio (INR), and the use of medications such as antiplatelet and anticoagulant agents (Table 1 and Table 2). We performed electronic searches in PubMed, focusing on literature published since 2016. Key search terms included paracentesis, thoracentesis, lumbar puncture, anticoagulant, antiplatelet, coagulopathy, INR, thrombocytopenia, and guideline. In addition, we used the following MeSH terms: spinal puncture AND blood coagulation disorders, spinal puncture AND platelet aggregation inhibitors, spinal puncture AND anticoagulants, paracentesis AND blood coagulation disorders, paracentesis AND platelet aggregation inhibitors, paracentesis AND anticoagulants, thoracentesis AND blood coagulation disorders, thoracentesis AND platelet aggregation inhibitors, and thoracentesis AND anticoagulants.

Weighing Risks and Benefits

Proceduralists should discuss risks and benefits with patients and the referring service before attempting to mitigate bleeding risk by holding antithrombotic agents or reversing coagulopathy, as these actions come with risks outside the anticipated procedure, in particular, an increased risk of thrombosis. There are many factors that influence an individual patient’s arterial thromboembolism and venous thromboembolism (VTE) risk, including surgical history, genetics, comorbidities, and the underlying indication for antithrombotic therapy. The American College of Chest Physicians updated their clinical practice guidelines describing perioperative thromboembolism risk stratification in 2012.¹² In general, higher-risk individuals include those with any mitral valve prosthesis, recent (within 6 months) stroke or transient ischemic attack (TIA), CHADS₂ score of 5 or 6, recent (within 3 months) VTE, or severe thrombophilia. Individuals at moderate risk include those with bileaflet aortic valve repair (AVR) with at least one major stroke risk factor, CHADS₂ score of 3 or 4, recent (within 3-12 months) or recurrent VTE, or active or recent (treatment within preceding 6 months) cancer. Finally, individuals at low risk include those with bileaflet AVR without major stroke risk factors, CHADS₂ score of 0 to 2, or VTE more than 12 months earlier. There are also patient-specific bleeding risk factors that should be considered, including hypertension, abnormal renal function, abnormal hepatic function, prior stroke, history of major bleeding (especially within the preceding 3 months), and bleeding history with a similar procedure.¹¹

Hepatic and Renal Dysfunction

In the setting of chronic liver disease, thrombocytopenia and elevated INR are generally not reliable indicators of bleeding risk.¹³ The included recommendations for INR and platelet count thresholds in the setting of chronic liver disease are derived from the referenced guidelines and supplemental personal communication with the guideline authors. Many antiplatelet and anticoagulant medications are partially cleared or metabolized by the liver, suggesting that hepatic dysfunction may impact drug clearance, but this has not been well studied. Impaired renal function should also be considered when determining appropriate hold times for antithrombotic drugs that are partially renally cleared. The periprocedural hold and restart times outlined in Table 2 are specific to patients without clinically significant hepatic or renal dysfunction. For patients with these conditions, further information on hold time adjustment can be found in the individual references.

Bridging Therapy

The decision to use bridging therapy prior to a bedside procedure must be individualized and take into account patient-specific factors. However, there is mounting evidence that bridging therapy is associated with higher risk of bleeding with no difference in the risk of thromboembolic events.^2,3,14 If the decision has been made to use bridging therapy with a heparin infusion prior to a bedside procedure, recommendations for hold and restart times can be found in Table 2.

Resuming Therapy

Another key consideration for procedures, especially those associated with a higher risk of bleeding, such as LP, is when to restart medications that have been held prior to the procedure. Table 2 provides a summary of the recommended postprocedural restart times for a variety of agents.

Other Considerations

Some guidelines referenced in this article are based on data collected on procedures performed by interventional radiologists, which may or may not accurately reflect the bleeding risks of bedside procedures performed by hospitalists. In the case of LP, we included some regional anesthesia and pain procedure guidelines based on the assumption that certain procedures are analogous to LP and associated with similar bleeding risks. Some of the guidelines referenced do not provide specific periprocedural INR and platelet thresholds (reported as “No threshold” in Table 1), instead offering statements that elevated INR and thrombocytopenia are not contraindications to bedside procedures and periprocedural transfusion of blood products is generally not recommended, based on the overall low risk of bleeding and lack of evidence for the efficacy of interventions intended to improve INR values and platelet counts in these situations. Patients undergoing paracentesis, thoracentesis, or LP may be on multiple antithrombotic agents, such as dual antiplatelet therapy. There are limited guidelines and studies on how to manage these agents in the periprocedural context; however, one guideline recommends continuing dual antiplatelet therapy for paracentesis, thoracentesis, and LP in patients who have cardiac stents.² There are also limited guidelines on how to handle patients on simultaneous antiplatelet and anticoagulant therapy.

Paracentesis is a common procedure that can be performed safely at the bedside. The overall rate of serious complications is low (1%-2%), with severe hemorrhage accounting for the majority of those complications (0.97%).¹⁵ Bleeding usually occurs from puncture of an abdominal wall vein, a mesenteric varix, or an inferior epigastric artery. Certain techniques may help to mitigate serious bleeding, including the use of ultrasound to avoid overlying vessels. Paracentesis is frequently performed in patients with cirrhosis, a population at increased risk for coagulopathy, although INR and platelet counts may not reflect aggregate bleeding risk in patients with cirrhosis. The American Association for the Study of Liver Diseases released new guidelines in 2021, stating that elevated prothrombin time or thrombocytopenia is not a contraindication to paracentesis.⁶ The most liberal guidelines for patients without chronic liver disease suggest correcting to an INR of 2.0 to 3.0, with multiple societies suggesting that a platelet count as low as 20,000/µL is safe.^2,3 As shown in Table 2, most guidelines recommend continuation of antiplatelet agents such as aspirin and thienopyridines (eg, clopidogrel, prasugrel), whereas recommendations vary regarding continuation of anticoagulant agents.

Akin to paracentesis, thoracentesis is generally considered to be a safe bedside procedure, with an incidence of thoracentesis-associated bleeding of less than 1%.¹⁵ Certain techniques may help to mitigate serious bleeding, including the insertion of the needle over the superior aspect of the rib in an effort to avoid the intercostal neurovascular bundle, which runs along the inferior aspect of each rib. Various clinical societies have proposed INR and platelet thresholds at which the risk of bleeding from thoracentesis is thought to be acceptable. The most liberal guidelines include a target INR of 2.0 to 3.0, although one group recommended an INR of <1.5.^2,5 Thoracentesis is commonly performed in patients with cirrhosis who develop hepatic hydrothorax. In this population, the Society of Interventional Radiology (SIR) guidelines state that there is no INR threshold that necessitates reversal strategies prior to the procedure.² For platelet count, there are multiple recommendations for greater than 20,000/µL and one for greater than 50,000/µL.^2,3,5 The recommendations for continuation or suspension of antiplatelet and anticoagulant medications prior to thoracentesis are similar to those for paracentesis. In general, continuing antiplatelet agents is felt to be safe, whereas there are mixed recommendations for anticoagulants, as described further in Table 2.

Compared to thoracentesis and paracentesis, LP is generally considered to be a higher-risk procedure owing to the rare possibility of spinal hematoma with associated neurologic compromise. In one retrospective review of more than 49,000 patients without coagulopathy who underwent LP, the risk for developing a spinal hematoma by 30 days post procedure was 0.20%.¹⁶ Certain techniques may help to mitigate serious bleeding, including the use of image guidance in patients with large body habitus or those with difficult anatomy. Compared with paracentesis and thoracentesis, guideline recommendations for safe INR and platelet thresholds in patients undergoing LP are based on a more limited body of evidence. Guidelines also suggest a target INR of anywhere from ≤1.5 to the most liberal suggestion of 2.0 to 3.0.^2-4 The SIR guidelines categorize LP as a low–bleeding risk procedure, with a platelet threshold of 20,000/µL but note that most other societies and guidelines regard LP as a high–bleeding risk procedure with more conservative platelet thresholds.² The Association of British Neurologists (ABN), however, allows platelets to be 40,000/µL or greater than 20,000/µL with an additional risk-benefit discussion.⁷ In contrast to paracentesis and thoracentesis, recommendations regarding hold times of antithrombotic medications prior to LP are more variable and sometimes more conservative. For example, some guidelines indicate that the thienopyridines can be continued, whereas others recommend holding them for up to 1 week prior to LP.^2,4,7

A theme throughout the recent literature and recommendations from clinical societies is that it is uncommon for there to be one unifying recommendation for every situation, especially regarding LP. Recent guidelines remain largely based on studies that are decades old. With bedside ultrasound becoming more accessible and established in daily practice, the risk of bleeding has been decreasing, potentially making periprocedural coagulopathies and antithrombotic agents less of a concern. For example, in a retrospective study of 69,859 paracenteses, ultrasound guidance reduced the risk of bleeding complications by 68%, an odds ratio of 0.32 (95% CI, 0.25-0.41).¹⁷ More research is needed to assess procedural bleeding risks in the context of current practice standards. This article focuses on a subset of bedside procedures most commonly performed by hospitalists. Similar references for other common bedside procedures, such as arthrocentesis, central venous catheter, and arterial line placement, would be helpful. Finally, this article does not capture such nuances as needle gauge, operator experience, availability of (and comfort with) ultrasound, and variations in patient anatomy, all of which are factors that can contribute to the complexities and risks of these bedside procedures.

Although not every internal medicine physician performs bedside procedures in their practice, it is vital that all understand the cognitive aspects of common bedside procedures. This necessitates the understanding of periprocedural risks and possible complications and applying that to individual patients. Correcting coagulopathy and stopping or reversing antithrombotic agents are mitigation strategies that are associated with risk. It is therefore important to understand when coagulopathy should be corrected and when antithrombotic agents should be held and for how long. With multiple existing and sometimes conflicting guidelines regarding periprocedural management of coagulopathy and antithrombotic agents, we hope that providing consolidated tables with this information will increase efficiency, aid in risk-benefit discussions between patients and care teams, and enhance patient safety.

The practice of internal medicine includes bedside procedures such as paracentesis, thoracentesis, and lumbar puncture (LP). The American Board of Internal Medicine requires graduates of internal medicine residency programs to be competent in the cognitive components of procedural training (eg, indications, contraindications, complications) and considers it essential that trainees have opportunities to perform procedures relevant to their intended career direction.¹ Whether or not the performance of procedures is part of a given hospitalist’s practice, it is necessary that hospitalists understand each procedure’s risks and mitigation strategies to prevent a range of periprocedural complications, including clinically significant bleeding. Numerous recommendations and guidelines exist describing bleeding risk for common procedures. In this Progress Note, we summarize and consolidate this literature, covering a range of scenarios common to the hospital setting, including thrombocytopenia, elevated international normalized ratio (INR), and the use of medications such as antiplatelet and anticoagulant agents (Table 1 and Table 2). We performed electronic searches in PubMed, focusing on literature published since 2016. Key search terms included paracentesis, thoracentesis, lumbar puncture, anticoagulant, antiplatelet, coagulopathy, INR, thrombocytopenia, and guideline. In addition, we used the following MeSH terms: spinal puncture AND blood coagulation disorders, spinal puncture AND platelet aggregation inhibitors, spinal puncture AND anticoagulants, paracentesis AND blood coagulation disorders, paracentesis AND platelet aggregation inhibitors, paracentesis AND anticoagulants, thoracentesis AND blood coagulation disorders, thoracentesis AND platelet aggregation inhibitors, and thoracentesis AND anticoagulants.

Weighing Risks and Benefits

Proceduralists should discuss risks and benefits with patients and the referring service before attempting to mitigate bleeding risk by holding antithrombotic agents or reversing coagulopathy, as these actions come with risks outside the anticipated procedure, in particular, an increased risk of thrombosis. There are many factors that influence an individual patient’s arterial thromboembolism and venous thromboembolism (VTE) risk, including surgical history, genetics, comorbidities, and the underlying indication for antithrombotic therapy. The American College of Chest Physicians updated their clinical practice guidelines describing perioperative thromboembolism risk stratification in 2012.¹² In general, higher-risk individuals include those with any mitral valve prosthesis, recent (within 6 months) stroke or transient ischemic attack (TIA), CHADS₂ score of 5 or 6, recent (within 3 months) VTE, or severe thrombophilia. Individuals at moderate risk include those with bileaflet aortic valve repair (AVR) with at least one major stroke risk factor, CHADS₂ score of 3 or 4, recent (within 3-12 months) or recurrent VTE, or active or recent (treatment within preceding 6 months) cancer. Finally, individuals at low risk include those with bileaflet AVR without major stroke risk factors, CHADS₂ score of 0 to 2, or VTE more than 12 months earlier. There are also patient-specific bleeding risk factors that should be considered, including hypertension, abnormal renal function, abnormal hepatic function, prior stroke, history of major bleeding (especially within the preceding 3 months), and bleeding history with a similar procedure.¹¹

Hepatic and Renal Dysfunction

In the setting of chronic liver disease, thrombocytopenia and elevated INR are generally not reliable indicators of bleeding risk.¹³ The included recommendations for INR and platelet count thresholds in the setting of chronic liver disease are derived from the referenced guidelines and supplemental personal communication with the guideline authors. Many antiplatelet and anticoagulant medications are partially cleared or metabolized by the liver, suggesting that hepatic dysfunction may impact drug clearance, but this has not been well studied. Impaired renal function should also be considered when determining appropriate hold times for antithrombotic drugs that are partially renally cleared. The periprocedural hold and restart times outlined in Table 2 are specific to patients without clinically significant hepatic or renal dysfunction. For patients with these conditions, further information on hold time adjustment can be found in the individual references.

Bridging Therapy

The decision to use bridging therapy prior to a bedside procedure must be individualized and take into account patient-specific factors. However, there is mounting evidence that bridging therapy is associated with higher risk of bleeding with no difference in the risk of thromboembolic events.^2,3,14 If the decision has been made to use bridging therapy with a heparin infusion prior to a bedside procedure, recommendations for hold and restart times can be found in Table 2.

Resuming Therapy

Another key consideration for procedures, especially those associated with a higher risk of bleeding, such as LP, is when to restart medications that have been held prior to the procedure. Table 2 provides a summary of the recommended postprocedural restart times for a variety of agents.

Other Considerations

Some guidelines referenced in this article are based on data collected on procedures performed by interventional radiologists, which may or may not accurately reflect the bleeding risks of bedside procedures performed by hospitalists. In the case of LP, we included some regional anesthesia and pain procedure guidelines based on the assumption that certain procedures are analogous to LP and associated with similar bleeding risks. Some of the guidelines referenced do not provide specific periprocedural INR and platelet thresholds (reported as “No threshold” in Table 1), instead offering statements that elevated INR and thrombocytopenia are not contraindications to bedside procedures and periprocedural transfusion of blood products is generally not recommended, based on the overall low risk of bleeding and lack of evidence for the efficacy of interventions intended to improve INR values and platelet counts in these situations. Patients undergoing paracentesis, thoracentesis, or LP may be on multiple antithrombotic agents, such as dual antiplatelet therapy. There are limited guidelines and studies on how to manage these agents in the periprocedural context; however, one guideline recommends continuing dual antiplatelet therapy for paracentesis, thoracentesis, and LP in patients who have cardiac stents.² There are also limited guidelines on how to handle patients on simultaneous antiplatelet and anticoagulant therapy.

Paracentesis is a common procedure that can be performed safely at the bedside. The overall rate of serious complications is low (1%-2%), with severe hemorrhage accounting for the majority of those complications (0.97%).¹⁵ Bleeding usually occurs from puncture of an abdominal wall vein, a mesenteric varix, or an inferior epigastric artery. Certain techniques may help to mitigate serious bleeding, including the use of ultrasound to avoid overlying vessels. Paracentesis is frequently performed in patients with cirrhosis, a population at increased risk for coagulopathy, although INR and platelet counts may not reflect aggregate bleeding risk in patients with cirrhosis. The American Association for the Study of Liver Diseases released new guidelines in 2021, stating that elevated prothrombin time or thrombocytopenia is not a contraindication to paracentesis.⁶ The most liberal guidelines for patients without chronic liver disease suggest correcting to an INR of 2.0 to 3.0, with multiple societies suggesting that a platelet count as low as 20,000/µL is safe.^2,3 As shown in Table 2, most guidelines recommend continuation of antiplatelet agents such as aspirin and thienopyridines (eg, clopidogrel, prasugrel), whereas recommendations vary regarding continuation of anticoagulant agents.

Akin to paracentesis, thoracentesis is generally considered to be a safe bedside procedure, with an incidence of thoracentesis-associated bleeding of less than 1%.¹⁵ Certain techniques may help to mitigate serious bleeding, including the insertion of the needle over the superior aspect of the rib in an effort to avoid the intercostal neurovascular bundle, which runs along the inferior aspect of each rib. Various clinical societies have proposed INR and platelet thresholds at which the risk of bleeding from thoracentesis is thought to be acceptable. The most liberal guidelines include a target INR of 2.0 to 3.0, although one group recommended an INR of <1.5.^2,5 Thoracentesis is commonly performed in patients with cirrhosis who develop hepatic hydrothorax. In this population, the Society of Interventional Radiology (SIR) guidelines state that there is no INR threshold that necessitates reversal strategies prior to the procedure.² For platelet count, there are multiple recommendations for greater than 20,000/µL and one for greater than 50,000/µL.^2,3,5 The recommendations for continuation or suspension of antiplatelet and anticoagulant medications prior to thoracentesis are similar to those for paracentesis. In general, continuing antiplatelet agents is felt to be safe, whereas there are mixed recommendations for anticoagulants, as described further in Table 2.

Compared to thoracentesis and paracentesis, LP is generally considered to be a higher-risk procedure owing to the rare possibility of spinal hematoma with associated neurologic compromise. In one retrospective review of more than 49,000 patients without coagulopathy who underwent LP, the risk for developing a spinal hematoma by 30 days post procedure was 0.20%.¹⁶ Certain techniques may help to mitigate serious bleeding, including the use of image guidance in patients with large body habitus or those with difficult anatomy. Compared with paracentesis and thoracentesis, guideline recommendations for safe INR and platelet thresholds in patients undergoing LP are based on a more limited body of evidence. Guidelines also suggest a target INR of anywhere from ≤1.5 to the most liberal suggestion of 2.0 to 3.0.^2-4 The SIR guidelines categorize LP as a low–bleeding risk procedure, with a platelet threshold of 20,000/µL but note that most other societies and guidelines regard LP as a high–bleeding risk procedure with more conservative platelet thresholds.² The Association of British Neurologists (ABN), however, allows platelets to be 40,000/µL or greater than 20,000/µL with an additional risk-benefit discussion.⁷ In contrast to paracentesis and thoracentesis, recommendations regarding hold times of antithrombotic medications prior to LP are more variable and sometimes more conservative. For example, some guidelines indicate that the thienopyridines can be continued, whereas others recommend holding them for up to 1 week prior to LP.^2,4,7

A theme throughout the recent literature and recommendations from clinical societies is that it is uncommon for there to be one unifying recommendation for every situation, especially regarding LP. Recent guidelines remain largely based on studies that are decades old. With bedside ultrasound becoming more accessible and established in daily practice, the risk of bleeding has been decreasing, potentially making periprocedural coagulopathies and antithrombotic agents less of a concern. For example, in a retrospective study of 69,859 paracenteses, ultrasound guidance reduced the risk of bleeding complications by 68%, an odds ratio of 0.32 (95% CI, 0.25-0.41).¹⁷ More research is needed to assess procedural bleeding risks in the context of current practice standards. This article focuses on a subset of bedside procedures most commonly performed by hospitalists. Similar references for other common bedside procedures, such as arthrocentesis, central venous catheter, and arterial line placement, would be helpful. Finally, this article does not capture such nuances as needle gauge, operator experience, availability of (and comfort with) ultrasound, and variations in patient anatomy, all of which are factors that can contribute to the complexities and risks of these bedside procedures.

Although not every internal medicine physician performs bedside procedures in their practice, it is vital that all understand the cognitive aspects of common bedside procedures. This necessitates the understanding of periprocedural risks and possible complications and applying that to individual patients. Correcting coagulopathy and stopping or reversing antithrombotic agents are mitigation strategies that are associated with risk. It is therefore important to understand when coagulopathy should be corrected and when antithrombotic agents should be held and for how long. With multiple existing and sometimes conflicting guidelines regarding periprocedural management of coagulopathy and antithrombotic agents, we hope that providing consolidated tables with this information will increase efficiency, aid in risk-benefit discussions between patients and care teams, and enhance patient safety.

Most hospital pediatricians can recall cases where an abnormal result in one unnecessary test led to a cascade of multiple further unnecessary treatments, procedures, and tests. These cases are well described in the literature and written off as a side effect of delivering high-quality, comprehensive pediatric care.¹ Unfortunately, however, these frequent events are not without consequence and can cause significant harm to patients, as well as stress and fear for parents and families, and indirectly waste valuable resources.

As we look forward to recovering from the COVID-19 pandemic, there are calls to prioritize high-value and more equitable care in the postpandemic world.² Choosing Wisely is a global movement comprised of clinician-led campaigns that partner with national specialty societies to develop lists of evidence-based recommendations of tests, treatments, and procedures that offer no added clinical value and may cause harm.³

In pediatrics, there is a growing recognition and published literature on the harms of overdiagnosis and unnecessary care in children.^4-6 Choosing Wisely recommendations are being used as a resource to drive healthcare prioritization and ensure low-value care is avoided so that greater focus can be placed on areas of need exacerbated by the pandemic. Using a Choosing Wisely perspective can drive quality and help inform a shift in practice, creating a roadmap for reducing testing or treatment cascades that harm patients and waste resources as we move toward the goal of high-value pediatric care. However, adoption of Choosing Wisely recommendations in pediatrics has been slow. For example, the pediatric working group of the Society of Hospital Medicine released a Choosing Wisely^® recommendation in 2013 against the use of continuous pulse oximetry monitoring in children with acute respiratory illness who are not on supplementary oxygen.⁷ Data from a cross-sectional study across 56 hospitals 6 years later found significant variation in this practice for infants hospitalized with bronchiolitis and not receiving supplemental oxygen; 46% were continuously monitored with pulse oximetry (range, 2%-92%).⁸

Traditionally, attention in children’s healthcare has focused on underuse (eg, immunizations or mental health) rather than overuse. Further, the weakness of the evidence base, with very few randomized controlled trials in children, limits our ability to provide sufficient confidence in the evidence supporting some of our recommendations.⁹

Second, there is also tremendous anxiety for both parents and frontline clinicians around diagnostic uncertainty of any kind when it comes to children. We endeavour to reassure ourselves and patients’ families by leaving no stone unturned. This approach can lead to unnecessary care, including false-positive test results, “incidentalomas,” and adverse effects from unnecessary medications. Despite the best intentions of assuaging caregivers’ anxiety, overuse of invasive and uncomfortable tests can have the opposite effect of increasing stress and trauma for both children and parents.

Third, there is compelling evidence that practice habits, once established, are difficult to break.¹⁰ Particularly in the high-stakes practice of hospital pediatric medicine, where we are conditioned to expect the worst and anticipate the unexpected. This “do everything to everyone” approach, however, can lead to significant harms for pediatric patients. For example, the exposure to ionizing radiation through unnecessary computed tomography (CT) scans can increase a child’s lifetime cancer risk.¹¹

The perpetuation of unnecessary care needs to change in pediatrics, especially for the most vulnerable young patients seeking hospital care. Implementation is a necessary next step to introduce recommendations into practice, and the Choosing Wisely efforts of the Hospital for Sick Children in Toronto, Canada, can offer insights into opportunities to embed this approach across similar quaternary care teaching hospitals, as well as general hospitals and the systems they support.

Creating Lists of Recommendations Aligned With Quality Metrics

The Hospital for Sick Children developed a hospital-specific Choosing Wisely list in 2016 to address a gap in existing Choosing Wisely Canada campaign recommendations related to pediatric hospitals.¹² Choosing Wisely Canada was initially focused on adult medicine, and a list of recommendations developed by the Canadian Paediatric Society relates mostly to overuse in pediatric outpatient settings and is not applicable to hospital-based practice.¹³ The Society of Hospital Medicine-Pediatric Hospital Medicine Choosing Wisely^® list predominantly pertained to unnecessary care of infants with bronchiolitis (eg, not to order chest radiographs in uncomplicated asthma and bronchiolitis). We had measured our compliance with this recommendation and found it was already well below the achievable benchmark of care in the United States,¹⁴ so we preferred to create a list that would resonate with our clinicians. Since the original list was created at the Hospital for Sick Children,¹² we have developed two subsequent lists of recommendations, which were released in 2018 and 2021 (Table).

The approach to list development used by staff pediatricians and trainees, with input from hospital staff and family advisors, has been described elsewhere.¹² The goal was to self-identify five local practices that we felt would help us reduce unnecessary care. This list served as the foundation of an organization-wide quality initiative driven by a steering committee that consisted of the clinician champions as well as representation from various groups at the hospital, including decision support, information services, the family advisory committee, and public affairs.

Each recommendation needed to be evidence-based and measurable, have a clinician champion to implement the recommendation, and have the potential to improve the quality and safety of the care we provided. “Balancing” measures needed to be carefully monitored to ensure that no diagnoses were being missed or negative effects resulted from decreasing these interventions. In order for a recommendation to be considered, a subgroup of the pediatric department’s clinical advisory committee reviewed the references provided to ensure that what was being suggested was based on published evidence and part of current national guidelines. The clinician champion needed to agree to lead the implementation project, and specific outcomes, including appropriate balancing measures, needed to be identified a priori, in addition to an appropriate mechanism to collect the data. Hospital executive leaders were supportive of the initiative and facilitated access to “in-kind” hospital resources as required, although no financial budget was provided. After some early success, the Department of Paediatrics provided part-time project management support to help coordinate the growth and administration of the initiative.

Measuring and Supporting Practice Change

The main implementation principles included targeted education/awareness, transparent measurement with audit/feedback, and, most importantly, embedding changes in the ordering process, essentially making it easy for frontline clinicians to do the right thing (and trickier to do the “wrong” thing). Audit and feedback have been used at both the individual provider level (eg, respiratory viral testing–ordering practices) and the divisional level (eg, ordering of postoperative antibiotics). These quality improvement initiatives have had a compelling impact. Scorecards have been developed and results shared internally using local divisional as well as hospital-wide tools, varying from staff meetings to screensavers across hospital computers and television screens and the hospital intranet. Evaluation is ongoing, but many of the initial results have been encouraging.^15-17

For example, the 2016 list includes recommendations related to emergency department (ED) test ordering. Implementation efforts to address unnecessary nasopharyngeal swabs for viral testing in bronchiolitis reduced this practice by 80%,¹⁵ and there has been a 50% reduction in ankle X-rays in children with acute ankle injuries who meet criteria for a low-risk examination.¹⁶ The 2016 list also included a recommendation related to inappropriate intravenous immunoglobulin (IVIG) use in children with typical acute immune thrombocytopenic purpura (ITP), and a targeted quality improvement initiative reduced inappropriate IVIG use by 50%, with no detectable increase in bleeding complications or readmission to hospital.¹⁷ These results have been sustained over a period of 3 or more years. Examples from the 2018 list include a 40% decrease in inappropriate urinary tract infection diagnosis and treatment in the ED and a four-fold decrease in the CT abdomen/pelvis imaging rate for low-risk trauma.¹⁸

The steering committee meets every 2 months and includes all of the clinician champions as well as representatives from strategic hospital resources and two family advisors (NGS). These meetings are chaired by the Associate Pediatrician-in-Chief (JNF) and the project manager. The progress of the active projects is discussed, and the experience of the group is used to problem-solve, plan ahead, and encourage academic presentation and publication of the various projects. Patient partnership and participation in committees has ensured that improvements to patient experience, satisfaction, and education are considered in the outcomes of implementation. Moreover, it has safeguarded that this effort is not misperceived as limiting care and remains focused on advancing quality, safety, and the patient experience.

While most projects have surpassed expectations, not all have proceeded as anticipated. The biggest challenge is finding a reliable and practical source for data collection. For example, at the time of initiation of the voiding cystourethrography (VCUG) recommendation, practice had presumably changed over the recent years, and compliance already exceeded the goal, illustrating the importance of current accurate data. The oxygen saturation–monitoring recommendation highlighted the challenge presented by data collection that requires manual audits; the inability to find staff to do this regularly significantly hampered this project. The critical role of the clinician champion was highlighted in a few projects when a lead was absent for a prolonged period of time (eg, due to a parental leave or change in job), with no willing replacement. There does seem to be a strong correlation between the commitment and passion of the clinician lead and the success of the project. We have incorporated the lessons learned into the development and rollout of the 2018 and 2021 lists.

The challenge is to scale up these successes to impact and change practice across the hospital pediatrics community. After 5 years, awareness of and engagement with this process are still not uniform across our hospital campus. Nevertheless, anecdotally, at the Hospital for Sick Children, there is a shift in culture where clinicians have processed the imperative to reduce overuse and unnecessary tests and treatments, with phrases such as “this is not very Choosing Wisely” entering the vernacular. It is becoming part of the culture. Second, the new generation of medical school trainees and residents has displayed a tremendous appetite and passion for stewardship and a sense that practice can change from the ground up. The SickKids Choosing Wisely efforts have been a hub for resident-led quality improvement projects and leadership for implementation of recommendations.¹⁹ As we continue to engage all providers at our hospital, we are also reaching out to the other community hospitals in our region, and all children’s hospitals in Canada, to share the principles and lessons learned from our program through a national community of practice.

Practicing pediatric medicine in a well-resourced hospital setting should not drive us to overuse in practice “just because we can.” The harms of this approach to our patients and health systems, coupled with the pressures of the pandemic, are compelling reasons to be responsible stewards. There are opportunities to reshape and rethink practice patterns and habits.²⁰ Overuse and overdiagnosis harm our patients and families physically and emotionally and indirectly waste resources urgently needed for investment upstream. Providing safe, quality, high-value care to our young patients requires constant critical thinking. The time is here to advance Choosing Wisely into pediatric hospital practice.

Most hospital pediatricians can recall cases where an abnormal result in one unnecessary test led to a cascade of multiple further unnecessary treatments, procedures, and tests. These cases are well described in the literature and written off as a side effect of delivering high-quality, comprehensive pediatric care.¹ Unfortunately, however, these frequent events are not without consequence and can cause significant harm to patients, as well as stress and fear for parents and families, and indirectly waste valuable resources.

As we look forward to recovering from the COVID-19 pandemic, there are calls to prioritize high-value and more equitable care in the postpandemic world.² Choosing Wisely is a global movement comprised of clinician-led campaigns that partner with national specialty societies to develop lists of evidence-based recommendations of tests, treatments, and procedures that offer no added clinical value and may cause harm.³

In pediatrics, there is a growing recognition and published literature on the harms of overdiagnosis and unnecessary care in children.^4-6 Choosing Wisely recommendations are being used as a resource to drive healthcare prioritization and ensure low-value care is avoided so that greater focus can be placed on areas of need exacerbated by the pandemic. Using a Choosing Wisely perspective can drive quality and help inform a shift in practice, creating a roadmap for reducing testing or treatment cascades that harm patients and waste resources as we move toward the goal of high-value pediatric care. However, adoption of Choosing Wisely recommendations in pediatrics has been slow. For example, the pediatric working group of the Society of Hospital Medicine released a Choosing Wisely^® recommendation in 2013 against the use of continuous pulse oximetry monitoring in children with acute respiratory illness who are not on supplementary oxygen.⁷ Data from a cross-sectional study across 56 hospitals 6 years later found significant variation in this practice for infants hospitalized with bronchiolitis and not receiving supplemental oxygen; 46% were continuously monitored with pulse oximetry (range, 2%-92%).⁸

Traditionally, attention in children’s healthcare has focused on underuse (eg, immunizations or mental health) rather than overuse. Further, the weakness of the evidence base, with very few randomized controlled trials in children, limits our ability to provide sufficient confidence in the evidence supporting some of our recommendations.⁹

Second, there is also tremendous anxiety for both parents and frontline clinicians around diagnostic uncertainty of any kind when it comes to children. We endeavour to reassure ourselves and patients’ families by leaving no stone unturned. This approach can lead to unnecessary care, including false-positive test results, “incidentalomas,” and adverse effects from unnecessary medications. Despite the best intentions of assuaging caregivers’ anxiety, overuse of invasive and uncomfortable tests can have the opposite effect of increasing stress and trauma for both children and parents.

Third, there is compelling evidence that practice habits, once established, are difficult to break.¹⁰ Particularly in the high-stakes practice of hospital pediatric medicine, where we are conditioned to expect the worst and anticipate the unexpected. This “do everything to everyone” approach, however, can lead to significant harms for pediatric patients. For example, the exposure to ionizing radiation through unnecessary computed tomography (CT) scans can increase a child’s lifetime cancer risk.¹¹

The perpetuation of unnecessary care needs to change in pediatrics, especially for the most vulnerable young patients seeking hospital care. Implementation is a necessary next step to introduce recommendations into practice, and the Choosing Wisely efforts of the Hospital for Sick Children in Toronto, Canada, can offer insights into opportunities to embed this approach across similar quaternary care teaching hospitals, as well as general hospitals and the systems they support.

Creating Lists of Recommendations Aligned With Quality Metrics

The Hospital for Sick Children developed a hospital-specific Choosing Wisely list in 2016 to address a gap in existing Choosing Wisely Canada campaign recommendations related to pediatric hospitals.¹² Choosing Wisely Canada was initially focused on adult medicine, and a list of recommendations developed by the Canadian Paediatric Society relates mostly to overuse in pediatric outpatient settings and is not applicable to hospital-based practice.¹³ The Society of Hospital Medicine-Pediatric Hospital Medicine Choosing Wisely^® list predominantly pertained to unnecessary care of infants with bronchiolitis (eg, not to order chest radiographs in uncomplicated asthma and bronchiolitis). We had measured our compliance with this recommendation and found it was already well below the achievable benchmark of care in the United States,¹⁴ so we preferred to create a list that would resonate with our clinicians. Since the original list was created at the Hospital for Sick Children,¹² we have developed two subsequent lists of recommendations, which were released in 2018 and 2021 (Table).

The approach to list development used by staff pediatricians and trainees, with input from hospital staff and family advisors, has been described elsewhere.¹² The goal was to self-identify five local practices that we felt would help us reduce unnecessary care. This list served as the foundation of an organization-wide quality initiative driven by a steering committee that consisted of the clinician champions as well as representation from various groups at the hospital, including decision support, information services, the family advisory committee, and public affairs.

Each recommendation needed to be evidence-based and measurable, have a clinician champion to implement the recommendation, and have the potential to improve the quality and safety of the care we provided. “Balancing” measures needed to be carefully monitored to ensure that no diagnoses were being missed or negative effects resulted from decreasing these interventions. In order for a recommendation to be considered, a subgroup of the pediatric department’s clinical advisory committee reviewed the references provided to ensure that what was being suggested was based on published evidence and part of current national guidelines. The clinician champion needed to agree to lead the implementation project, and specific outcomes, including appropriate balancing measures, needed to be identified a priori, in addition to an appropriate mechanism to collect the data. Hospital executive leaders were supportive of the initiative and facilitated access to “in-kind” hospital resources as required, although no financial budget was provided. After some early success, the Department of Paediatrics provided part-time project management support to help coordinate the growth and administration of the initiative.

Measuring and Supporting Practice Change

The main implementation principles included targeted education/awareness, transparent measurement with audit/feedback, and, most importantly, embedding changes in the ordering process, essentially making it easy for frontline clinicians to do the right thing (and trickier to do the “wrong” thing). Audit and feedback have been used at both the individual provider level (eg, respiratory viral testing–ordering practices) and the divisional level (eg, ordering of postoperative antibiotics). These quality improvement initiatives have had a compelling impact. Scorecards have been developed and results shared internally using local divisional as well as hospital-wide tools, varying from staff meetings to screensavers across hospital computers and television screens and the hospital intranet. Evaluation is ongoing, but many of the initial results have been encouraging.^15-17

For example, the 2016 list includes recommendations related to emergency department (ED) test ordering. Implementation efforts to address unnecessary nasopharyngeal swabs for viral testing in bronchiolitis reduced this practice by 80%,¹⁵ and there has been a 50% reduction in ankle X-rays in children with acute ankle injuries who meet criteria for a low-risk examination.¹⁶ The 2016 list also included a recommendation related to inappropriate intravenous immunoglobulin (IVIG) use in children with typical acute immune thrombocytopenic purpura (ITP), and a targeted quality improvement initiative reduced inappropriate IVIG use by 50%, with no detectable increase in bleeding complications or readmission to hospital.¹⁷ These results have been sustained over a period of 3 or more years. Examples from the 2018 list include a 40% decrease in inappropriate urinary tract infection diagnosis and treatment in the ED and a four-fold decrease in the CT abdomen/pelvis imaging rate for low-risk trauma.¹⁸

The steering committee meets every 2 months and includes all of the clinician champions as well as representatives from strategic hospital resources and two family advisors (NGS). These meetings are chaired by the Associate Pediatrician-in-Chief (JNF) and the project manager. The progress of the active projects is discussed, and the experience of the group is used to problem-solve, plan ahead, and encourage academic presentation and publication of the various projects. Patient partnership and participation in committees has ensured that improvements to patient experience, satisfaction, and education are considered in the outcomes of implementation. Moreover, it has safeguarded that this effort is not misperceived as limiting care and remains focused on advancing quality, safety, and the patient experience.

While most projects have surpassed expectations, not all have proceeded as anticipated. The biggest challenge is finding a reliable and practical source for data collection. For example, at the time of initiation of the voiding cystourethrography (VCUG) recommendation, practice had presumably changed over the recent years, and compliance already exceeded the goal, illustrating the importance of current accurate data. The oxygen saturation–monitoring recommendation highlighted the challenge presented by data collection that requires manual audits; the inability to find staff to do this regularly significantly hampered this project. The critical role of the clinician champion was highlighted in a few projects when a lead was absent for a prolonged period of time (eg, due to a parental leave or change in job), with no willing replacement. There does seem to be a strong correlation between the commitment and passion of the clinician lead and the success of the project. We have incorporated the lessons learned into the development and rollout of the 2018 and 2021 lists.

The challenge is to scale up these successes to impact and change practice across the hospital pediatrics community. After 5 years, awareness of and engagement with this process are still not uniform across our hospital campus. Nevertheless, anecdotally, at the Hospital for Sick Children, there is a shift in culture where clinicians have processed the imperative to reduce overuse and unnecessary tests and treatments, with phrases such as “this is not very Choosing Wisely” entering the vernacular. It is becoming part of the culture. Second, the new generation of medical school trainees and residents has displayed a tremendous appetite and passion for stewardship and a sense that practice can change from the ground up. The SickKids Choosing Wisely efforts have been a hub for resident-led quality improvement projects and leadership for implementation of recommendations.¹⁹ As we continue to engage all providers at our hospital, we are also reaching out to the other community hospitals in our region, and all children’s hospitals in Canada, to share the principles and lessons learned from our program through a national community of practice.

Practicing pediatric medicine in a well-resourced hospital setting should not drive us to overuse in practice “just because we can.” The harms of this approach to our patients and health systems, coupled with the pressures of the pandemic, are compelling reasons to be responsible stewards. There are opportunities to reshape and rethink practice patterns and habits.²⁰ Overuse and overdiagnosis harm our patients and families physically and emotionally and indirectly waste resources urgently needed for investment upstream. Providing safe, quality, high-value care to our young patients requires constant critical thinking. The time is here to advance Choosing Wisely into pediatric hospital practice.

Most hospital pediatricians can recall cases where an abnormal result in one unnecessary test led to a cascade of multiple further unnecessary treatments, procedures, and tests. These cases are well described in the literature and written off as a side effect of delivering high-quality, comprehensive pediatric care.¹ Unfortunately, however, these frequent events are not without consequence and can cause significant harm to patients, as well as stress and fear for parents and families, and indirectly waste valuable resources.

As we look forward to recovering from the COVID-19 pandemic, there are calls to prioritize high-value and more equitable care in the postpandemic world.² Choosing Wisely is a global movement comprised of clinician-led campaigns that partner with national specialty societies to develop lists of evidence-based recommendations of tests, treatments, and procedures that offer no added clinical value and may cause harm.³

In pediatrics, there is a growing recognition and published literature on the harms of overdiagnosis and unnecessary care in children.^4-6 Choosing Wisely recommendations are being used as a resource to drive healthcare prioritization and ensure low-value care is avoided so that greater focus can be placed on areas of need exacerbated by the pandemic. Using a Choosing Wisely perspective can drive quality and help inform a shift in practice, creating a roadmap for reducing testing or treatment cascades that harm patients and waste resources as we move toward the goal of high-value pediatric care. However, adoption of Choosing Wisely recommendations in pediatrics has been slow. For example, the pediatric working group of the Society of Hospital Medicine released a Choosing Wisely^® recommendation in 2013 against the use of continuous pulse oximetry monitoring in children with acute respiratory illness who are not on supplementary oxygen.⁷ Data from a cross-sectional study across 56 hospitals 6 years later found significant variation in this practice for infants hospitalized with bronchiolitis and not receiving supplemental oxygen; 46% were continuously monitored with pulse oximetry (range, 2%-92%).⁸

Traditionally, attention in children’s healthcare has focused on underuse (eg, immunizations or mental health) rather than overuse. Further, the weakness of the evidence base, with very few randomized controlled trials in children, limits our ability to provide sufficient confidence in the evidence supporting some of our recommendations.⁹

Second, there is also tremendous anxiety for both parents and frontline clinicians around diagnostic uncertainty of any kind when it comes to children. We endeavour to reassure ourselves and patients’ families by leaving no stone unturned. This approach can lead to unnecessary care, including false-positive test results, “incidentalomas,” and adverse effects from unnecessary medications. Despite the best intentions of assuaging caregivers’ anxiety, overuse of invasive and uncomfortable tests can have the opposite effect of increasing stress and trauma for both children and parents.

Third, there is compelling evidence that practice habits, once established, are difficult to break.¹⁰ Particularly in the high-stakes practice of hospital pediatric medicine, where we are conditioned to expect the worst and anticipate the unexpected. This “do everything to everyone” approach, however, can lead to significant harms for pediatric patients. For example, the exposure to ionizing radiation through unnecessary computed tomography (CT) scans can increase a child’s lifetime cancer risk.¹¹

The perpetuation of unnecessary care needs to change in pediatrics, especially for the most vulnerable young patients seeking hospital care. Implementation is a necessary next step to introduce recommendations into practice, and the Choosing Wisely efforts of the Hospital for Sick Children in Toronto, Canada, can offer insights into opportunities to embed this approach across similar quaternary care teaching hospitals, as well as general hospitals and the systems they support.

Creating Lists of Recommendations Aligned With Quality Metrics

The Hospital for Sick Children developed a hospital-specific Choosing Wisely list in 2016 to address a gap in existing Choosing Wisely Canada campaign recommendations related to pediatric hospitals.¹² Choosing Wisely Canada was initially focused on adult medicine, and a list of recommendations developed by the Canadian Paediatric Society relates mostly to overuse in pediatric outpatient settings and is not applicable to hospital-based practice.¹³ The Society of Hospital Medicine-Pediatric Hospital Medicine Choosing Wisely^® list predominantly pertained to unnecessary care of infants with bronchiolitis (eg, not to order chest radiographs in uncomplicated asthma and bronchiolitis). We had measured our compliance with this recommendation and found it was already well below the achievable benchmark of care in the United States,¹⁴ so we preferred to create a list that would resonate with our clinicians. Since the original list was created at the Hospital for Sick Children,¹² we have developed two subsequent lists of recommendations, which were released in 2018 and 2021 (Table).

The approach to list development used by staff pediatricians and trainees, with input from hospital staff and family advisors, has been described elsewhere.¹² The goal was to self-identify five local practices that we felt would help us reduce unnecessary care. This list served as the foundation of an organization-wide quality initiative driven by a steering committee that consisted of the clinician champions as well as representation from various groups at the hospital, including decision support, information services, the family advisory committee, and public affairs.

Each recommendation needed to be evidence-based and measurable, have a clinician champion to implement the recommendation, and have the potential to improve the quality and safety of the care we provided. “Balancing” measures needed to be carefully monitored to ensure that no diagnoses were being missed or negative effects resulted from decreasing these interventions. In order for a recommendation to be considered, a subgroup of the pediatric department’s clinical advisory committee reviewed the references provided to ensure that what was being suggested was based on published evidence and part of current national guidelines. The clinician champion needed to agree to lead the implementation project, and specific outcomes, including appropriate balancing measures, needed to be identified a priori, in addition to an appropriate mechanism to collect the data. Hospital executive leaders were supportive of the initiative and facilitated access to “in-kind” hospital resources as required, although no financial budget was provided. After some early success, the Department of Paediatrics provided part-time project management support to help coordinate the growth and administration of the initiative.

Measuring and Supporting Practice Change

The main implementation principles included targeted education/awareness, transparent measurement with audit/feedback, and, most importantly, embedding changes in the ordering process, essentially making it easy for frontline clinicians to do the right thing (and trickier to do the “wrong” thing). Audit and feedback have been used at both the individual provider level (eg, respiratory viral testing–ordering practices) and the divisional level (eg, ordering of postoperative antibiotics). These quality improvement initiatives have had a compelling impact. Scorecards have been developed and results shared internally using local divisional as well as hospital-wide tools, varying from staff meetings to screensavers across hospital computers and television screens and the hospital intranet. Evaluation is ongoing, but many of the initial results have been encouraging.^15-17

For example, the 2016 list includes recommendations related to emergency department (ED) test ordering. Implementation efforts to address unnecessary nasopharyngeal swabs for viral testing in bronchiolitis reduced this practice by 80%,¹⁵ and there has been a 50% reduction in ankle X-rays in children with acute ankle injuries who meet criteria for a low-risk examination.¹⁶ The 2016 list also included a recommendation related to inappropriate intravenous immunoglobulin (IVIG) use in children with typical acute immune thrombocytopenic purpura (ITP), and a targeted quality improvement initiative reduced inappropriate IVIG use by 50%, with no detectable increase in bleeding complications or readmission to hospital.¹⁷ These results have been sustained over a period of 3 or more years. Examples from the 2018 list include a 40% decrease in inappropriate urinary tract infection diagnosis and treatment in the ED and a four-fold decrease in the CT abdomen/pelvis imaging rate for low-risk trauma.¹⁸

The steering committee meets every 2 months and includes all of the clinician champions as well as representatives from strategic hospital resources and two family advisors (NGS). These meetings are chaired by the Associate Pediatrician-in-Chief (JNF) and the project manager. The progress of the active projects is discussed, and the experience of the group is used to problem-solve, plan ahead, and encourage academic presentation and publication of the various projects. Patient partnership and participation in committees has ensured that improvements to patient experience, satisfaction, and education are considered in the outcomes of implementation. Moreover, it has safeguarded that this effort is not misperceived as limiting care and remains focused on advancing quality, safety, and the patient experience.

While most projects have surpassed expectations, not all have proceeded as anticipated. The biggest challenge is finding a reliable and practical source for data collection. For example, at the time of initiation of the voiding cystourethrography (VCUG) recommendation, practice had presumably changed over the recent years, and compliance already exceeded the goal, illustrating the importance of current accurate data. The oxygen saturation–monitoring recommendation highlighted the challenge presented by data collection that requires manual audits; the inability to find staff to do this regularly significantly hampered this project. The critical role of the clinician champion was highlighted in a few projects when a lead was absent for a prolonged period of time (eg, due to a parental leave or change in job), with no willing replacement. There does seem to be a strong correlation between the commitment and passion of the clinician lead and the success of the project. We have incorporated the lessons learned into the development and rollout of the 2018 and 2021 lists.

The challenge is to scale up these successes to impact and change practice across the hospital pediatrics community. After 5 years, awareness of and engagement with this process are still not uniform across our hospital campus. Nevertheless, anecdotally, at the Hospital for Sick Children, there is a shift in culture where clinicians have processed the imperative to reduce overuse and unnecessary tests and treatments, with phrases such as “this is not very Choosing Wisely” entering the vernacular. It is becoming part of the culture. Second, the new generation of medical school trainees and residents has displayed a tremendous appetite and passion for stewardship and a sense that practice can change from the ground up. The SickKids Choosing Wisely efforts have been a hub for resident-led quality improvement projects and leadership for implementation of recommendations.¹⁹ As we continue to engage all providers at our hospital, we are also reaching out to the other community hospitals in our region, and all children’s hospitals in Canada, to share the principles and lessons learned from our program through a national community of practice.

Practicing pediatric medicine in a well-resourced hospital setting should not drive us to overuse in practice “just because we can.” The harms of this approach to our patients and health systems, coupled with the pressures of the pandemic, are compelling reasons to be responsible stewards. There are opportunities to reshape and rethink practice patterns and habits.²⁰ Overuse and overdiagnosis harm our patients and families physically and emotionally and indirectly waste resources urgently needed for investment upstream. Providing safe, quality, high-value care to our young patients requires constant critical thinking. The time is here to advance Choosing Wisely into pediatric hospital practice.

A 59-year-old man is observed in the hospital for substernal chest pain initially concerning for angina. Serial troponin testing is negative, and based on additional history of intermittent dysphagia, an elective upper endoscopy is recommended after discharge. The patient does not have health insurance and expresses anxiety about the cost of endoscopy. He asks how he could compare the costs at different hospitals. How do federal price transparency rules assist the hospitalist in addressing this patient’s question?

Healthcare costs continue to rise in the United States despite mounting concerns about wasteful spending and unaffordability.¹ One contributor is a lack of price transparency.² In theory, price transparency allows individuals to shop for services, spurring competition and lower prices. However, healthcare prices have historically been opaque to both physicians and patients; unlike other licensed professionals who provide clients estimates for their work (eg, lawyers, electricians), physicians are rarely able to offer patients real-time insight or guidance about costs, which most patients discover only when the bill arrives. The situation is particularly problematic for patients who bear higher out-of-pocket costs, such as the uninsured or those with high-deductible health plans.³

Decades of work to improve healthcare price transparency have unfortunately borne little fruit. Multiple states and organizations have attempted to disseminate price information on comparison websites.⁴ These efforts only modestly reduced some prices, with benefits confined to elective, single-episode, commodifiable services such as magnetic resonance imaging scans.⁵ The Affordable Care Act required hospitals to publish standard charges, also called a chargemaster (Table).⁶ However, chargemaster fees are notoriously inflated and inaccessible at the point of service, undercutting transparency.

Beginning January 2021, the Centers for Medicare & Medicaid Services (CMS) required all hospitals to publish negotiated prices—including payor-specific negotiated charges—for 300 “shoppable services” (Table).⁶ The list must include 70 common CMS-specified services, such as a basic metabolic panel, upper endoscopy, and prostate biopsy, as well as another 230 services that each hospital determines relevant to its patient population.

In circumstances where hospitals have negotiated different prices for a service, they must list each third-party payor and their payor-specific charge. The information must be prominently displayed, accessible without requiring the patient to enter personal information, and provided in a machine-readable file. CMS may impose a $300 daily penalty on hospitals failing to comply with the policy. Of note, the policy does not apply to clinics or ambulatory surgery centers.

As more hospitals share data, this policy will directly benefit both patients and physicians. It can benefit patients with the time, foresight, and ability to search for the lowest price for shoppable services. Other patients may also benefit indirectly, to the extent that insurers and other purchasers apply this information to negotiate lower and more uniform prices. Decreased price variation may also encourage hospitals to compete on quality to distinguish the value of their services. Hospitalists could benefit through the ability to directly help patients locate price information.

Despite these potential benefits, the policy has limitations. Price information about shoppable services is most useful for discharge planning, and other solutions are needed to address transparency before and during unplanned admissions. Patients who prioritize continuity with a hospital or physician may be less price sensitive, particularly for more complex services. Patients with commercial insurance may be shielded from cost considerations and personal incentives to comparison shop. Interpreting hospitals’ estimates remains difficult, as it can be unclear if professional fees are included or if certain prices are offered to outpatients.⁷ Price information is not accompanied by corresponding quality data. Additionally, price transparency may also fail to lower prices in heavily concentrated payor or provider markets, and it remains unknown whether some providers may actually raise prices after learning about higher rates negotiated by competitors.^8,9

Another issue is hospital participation. Early evidence suggests that most hospitals have not complied with the letter or spirit of the regulation.^7,10 A sample of the country’s 100 largest hospitals in February 2021 found 18 lacked downloadable files and 46 did not display payor-specific rates.¹¹ In addition, some hospitals posted prices on websites designed to block discovery by search engines, a tactic deemed illegal by CMS.¹² Thus far, enforcement efforts have consisted of warnings rather than financial penalties.

Despite its limitations, this policy represents a meaningful advance for healthcare competition and patient empowerment. Additionally, it signals federal willingness to address the lack of price transparency as a source of widespread patient and clinician frustration—a commitment that will be needed to sustain this policy and implement additional measures in the future.

CMS could consider five steps to augment the policy and maximize transparency and value for patients.

First, CMS could consider increasing daily nonparticipation penalties. Hospitals, particularly those in areas with less competition, have less incentive to participate given meager current penalties. Because the magnitude needed to compel action remains unknown, CMS could gradually escalate penalties over time until there is broader participation across hospitals.

Second, policymakers could aggregate price information centrally, organize the data around patients’ clinical scenarios, and advertise its availability. Currently, this information is scattered and time-consuming for hospitalists and patients to gather for decision-making. Additionally, CMS could encourage the development of third-party tools that aggregate and analyze machine-readable price data or require that prices be posted at the point of service.

Third, CMS could revise the policy to include quality as well as price information. Price alone does not offer a full enough picture of what consumers can expect from hospitals for shoppable services. Pairing price and quality information is better aligned to addressing costs in the context of value, rather than cost-cutting for its own purposes.

Fourth, over time, CMS could expand the list of services and sites required to report (eg, clinics and ambulatory surgical centers as well as hospitals).

Fifth, CMS rule-makers could set reporting standards and contextualize price information in common clinical scenarios. Patients may have difficulty shopping for complex healthcare services without understanding how they apply in different clinical situations. Decision-making would also be aided by reporting standards—for instance, for how prices are displayed and whether they include certain fees (eg, professional fees, pathology studies).

Hospitalists planning follow-up care should inform patients that price information is increasingly available and encourage them to search on the internet or contact hospital billing offices to request information (eg, discounted cash prices and minimum negotiated charges) before obtaining elective services after discharge. Hospitalists can also encourage patients to discuss shoppable services with their primary care physicians to understand the clinical context and make high-value decisions. Hospitalists who wish to build communication skills discussing costs with patients can increasingly find resources for these conversations and request that prices be displayed in the electronic health record for this purpose.^13,14 As conversations occur, hospitalists should seek to understand other factors, such as convenience and continuity relationships, that might influence choices.

Starting in 2021, CMS policy requires that hospitals report prices for services such as the endoscopy recommended for the patient in the scenario. Though the policy gives patients new hope for greater transparency and better prices, additional steps are needed to help patients and hospitalists achieve these benefits.

A 59-year-old man is observed in the hospital for substernal chest pain initially concerning for angina. Serial troponin testing is negative, and based on additional history of intermittent dysphagia, an elective upper endoscopy is recommended after discharge. The patient does not have health insurance and expresses anxiety about the cost of endoscopy. He asks how he could compare the costs at different hospitals. How do federal price transparency rules assist the hospitalist in addressing this patient’s question?

Healthcare costs continue to rise in the United States despite mounting concerns about wasteful spending and unaffordability.¹ One contributor is a lack of price transparency.² In theory, price transparency allows individuals to shop for services, spurring competition and lower prices. However, healthcare prices have historically been opaque to both physicians and patients; unlike other licensed professionals who provide clients estimates for their work (eg, lawyers, electricians), physicians are rarely able to offer patients real-time insight or guidance about costs, which most patients discover only when the bill arrives. The situation is particularly problematic for patients who bear higher out-of-pocket costs, such as the uninsured or those with high-deductible health plans.³

Decades of work to improve healthcare price transparency have unfortunately borne little fruit. Multiple states and organizations have attempted to disseminate price information on comparison websites.⁴ These efforts only modestly reduced some prices, with benefits confined to elective, single-episode, commodifiable services such as magnetic resonance imaging scans.⁵ The Affordable Care Act required hospitals to publish standard charges, also called a chargemaster (Table).⁶ However, chargemaster fees are notoriously inflated and inaccessible at the point of service, undercutting transparency.

Beginning January 2021, the Centers for Medicare & Medicaid Services (CMS) required all hospitals to publish negotiated prices—including payor-specific negotiated charges—for 300 “shoppable services” (Table).⁶ The list must include 70 common CMS-specified services, such as a basic metabolic panel, upper endoscopy, and prostate biopsy, as well as another 230 services that each hospital determines relevant to its patient population.

In circumstances where hospitals have negotiated different prices for a service, they must list each third-party payor and their payor-specific charge. The information must be prominently displayed, accessible without requiring the patient to enter personal information, and provided in a machine-readable file. CMS may impose a $300 daily penalty on hospitals failing to comply with the policy. Of note, the policy does not apply to clinics or ambulatory surgery centers.

As more hospitals share data, this policy will directly benefit both patients and physicians. It can benefit patients with the time, foresight, and ability to search for the lowest price for shoppable services. Other patients may also benefit indirectly, to the extent that insurers and other purchasers apply this information to negotiate lower and more uniform prices. Decreased price variation may also encourage hospitals to compete on quality to distinguish the value of their services. Hospitalists could benefit through the ability to directly help patients locate price information.

Despite these potential benefits, the policy has limitations. Price information about shoppable services is most useful for discharge planning, and other solutions are needed to address transparency before and during unplanned admissions. Patients who prioritize continuity with a hospital or physician may be less price sensitive, particularly for more complex services. Patients with commercial insurance may be shielded from cost considerations and personal incentives to comparison shop. Interpreting hospitals’ estimates remains difficult, as it can be unclear if professional fees are included or if certain prices are offered to outpatients.⁷ Price information is not accompanied by corresponding quality data. Additionally, price transparency may also fail to lower prices in heavily concentrated payor or provider markets, and it remains unknown whether some providers may actually raise prices after learning about higher rates negotiated by competitors.^8,9

Another issue is hospital participation. Early evidence suggests that most hospitals have not complied with the letter or spirit of the regulation.^7,10 A sample of the country’s 100 largest hospitals in February 2021 found 18 lacked downloadable files and 46 did not display payor-specific rates.¹¹ In addition, some hospitals posted prices on websites designed to block discovery by search engines, a tactic deemed illegal by CMS.¹² Thus far, enforcement efforts have consisted of warnings rather than financial penalties.

Despite its limitations, this policy represents a meaningful advance for healthcare competition and patient empowerment. Additionally, it signals federal willingness to address the lack of price transparency as a source of widespread patient and clinician frustration—a commitment that will be needed to sustain this policy and implement additional measures in the future.

CMS could consider five steps to augment the policy and maximize transparency and value for patients.

First, CMS could consider increasing daily nonparticipation penalties. Hospitals, particularly those in areas with less competition, have less incentive to participate given meager current penalties. Because the magnitude needed to compel action remains unknown, CMS could gradually escalate penalties over time until there is broader participation across hospitals.

Second, policymakers could aggregate price information centrally, organize the data around patients’ clinical scenarios, and advertise its availability. Currently, this information is scattered and time-consuming for hospitalists and patients to gather for decision-making. Additionally, CMS could encourage the development of third-party tools that aggregate and analyze machine-readable price data or require that prices be posted at the point of service.

Third, CMS could revise the policy to include quality as well as price information. Price alone does not offer a full enough picture of what consumers can expect from hospitals for shoppable services. Pairing price and quality information is better aligned to addressing costs in the context of value, rather than cost-cutting for its own purposes.

Fourth, over time, CMS could expand the list of services and sites required to report (eg, clinics and ambulatory surgical centers as well as hospitals).

Fifth, CMS rule-makers could set reporting standards and contextualize price information in common clinical scenarios. Patients may have difficulty shopping for complex healthcare services without understanding how they apply in different clinical situations. Decision-making would also be aided by reporting standards—for instance, for how prices are displayed and whether they include certain fees (eg, professional fees, pathology studies).

Hospitalists planning follow-up care should inform patients that price information is increasingly available and encourage them to search on the internet or contact hospital billing offices to request information (eg, discounted cash prices and minimum negotiated charges) before obtaining elective services after discharge. Hospitalists can also encourage patients to discuss shoppable services with their primary care physicians to understand the clinical context and make high-value decisions. Hospitalists who wish to build communication skills discussing costs with patients can increasingly find resources for these conversations and request that prices be displayed in the electronic health record for this purpose.^13,14 As conversations occur, hospitalists should seek to understand other factors, such as convenience and continuity relationships, that might influence choices.

Starting in 2021, CMS policy requires that hospitals report prices for services such as the endoscopy recommended for the patient in the scenario. Though the policy gives patients new hope for greater transparency and better prices, additional steps are needed to help patients and hospitalists achieve these benefits.

A 59-year-old man is observed in the hospital for substernal chest pain initially concerning for angina. Serial troponin testing is negative, and based on additional history of intermittent dysphagia, an elective upper endoscopy is recommended after discharge. The patient does not have health insurance and expresses anxiety about the cost of endoscopy. He asks how he could compare the costs at different hospitals. How do federal price transparency rules assist the hospitalist in addressing this patient’s question?

Healthcare costs continue to rise in the United States despite mounting concerns about wasteful spending and unaffordability.¹ One contributor is a lack of price transparency.² In theory, price transparency allows individuals to shop for services, spurring competition and lower prices. However, healthcare prices have historically been opaque to both physicians and patients; unlike other licensed professionals who provide clients estimates for their work (eg, lawyers, electricians), physicians are rarely able to offer patients real-time insight or guidance about costs, which most patients discover only when the bill arrives. The situation is particularly problematic for patients who bear higher out-of-pocket costs, such as the uninsured or those with high-deductible health plans.³

Decades of work to improve healthcare price transparency have unfortunately borne little fruit. Multiple states and organizations have attempted to disseminate price information on comparison websites.⁴ These efforts only modestly reduced some prices, with benefits confined to elective, single-episode, commodifiable services such as magnetic resonance imaging scans.⁵ The Affordable Care Act required hospitals to publish standard charges, also called a chargemaster (Table).⁶ However, chargemaster fees are notoriously inflated and inaccessible at the point of service, undercutting transparency.

Beginning January 2021, the Centers for Medicare & Medicaid Services (CMS) required all hospitals to publish negotiated prices—including payor-specific negotiated charges—for 300 “shoppable services” (Table).⁶ The list must include 70 common CMS-specified services, such as a basic metabolic panel, upper endoscopy, and prostate biopsy, as well as another 230 services that each hospital determines relevant to its patient population.

In circumstances where hospitals have negotiated different prices for a service, they must list each third-party payor and their payor-specific charge. The information must be prominently displayed, accessible without requiring the patient to enter personal information, and provided in a machine-readable file. CMS may impose a $300 daily penalty on hospitals failing to comply with the policy. Of note, the policy does not apply to clinics or ambulatory surgery centers.

As more hospitals share data, this policy will directly benefit both patients and physicians. It can benefit patients with the time, foresight, and ability to search for the lowest price for shoppable services. Other patients may also benefit indirectly, to the extent that insurers and other purchasers apply this information to negotiate lower and more uniform prices. Decreased price variation may also encourage hospitals to compete on quality to distinguish the value of their services. Hospitalists could benefit through the ability to directly help patients locate price information.

Despite these potential benefits, the policy has limitations. Price information about shoppable services is most useful for discharge planning, and other solutions are needed to address transparency before and during unplanned admissions. Patients who prioritize continuity with a hospital or physician may be less price sensitive, particularly for more complex services. Patients with commercial insurance may be shielded from cost considerations and personal incentives to comparison shop. Interpreting hospitals’ estimates remains difficult, as it can be unclear if professional fees are included or if certain prices are offered to outpatients.⁷ Price information is not accompanied by corresponding quality data. Additionally, price transparency may also fail to lower prices in heavily concentrated payor or provider markets, and it remains unknown whether some providers may actually raise prices after learning about higher rates negotiated by competitors.^8,9

Another issue is hospital participation. Early evidence suggests that most hospitals have not complied with the letter or spirit of the regulation.^7,10 A sample of the country’s 100 largest hospitals in February 2021 found 18 lacked downloadable files and 46 did not display payor-specific rates.¹¹ In addition, some hospitals posted prices on websites designed to block discovery by search engines, a tactic deemed illegal by CMS.¹² Thus far, enforcement efforts have consisted of warnings rather than financial penalties.

Despite its limitations, this policy represents a meaningful advance for healthcare competition and patient empowerment. Additionally, it signals federal willingness to address the lack of price transparency as a source of widespread patient and clinician frustration—a commitment that will be needed to sustain this policy and implement additional measures in the future.

CMS could consider five steps to augment the policy and maximize transparency and value for patients.

First, CMS could consider increasing daily nonparticipation penalties. Hospitals, particularly those in areas with less competition, have less incentive to participate given meager current penalties. Because the magnitude needed to compel action remains unknown, CMS could gradually escalate penalties over time until there is broader participation across hospitals.

Second, policymakers could aggregate price information centrally, organize the data around patients’ clinical scenarios, and advertise its availability. Currently, this information is scattered and time-consuming for hospitalists and patients to gather for decision-making. Additionally, CMS could encourage the development of third-party tools that aggregate and analyze machine-readable price data or require that prices be posted at the point of service.

Third, CMS could revise the policy to include quality as well as price information. Price alone does not offer a full enough picture of what consumers can expect from hospitals for shoppable services. Pairing price and quality information is better aligned to addressing costs in the context of value, rather than cost-cutting for its own purposes.

Fourth, over time, CMS could expand the list of services and sites required to report (eg, clinics and ambulatory surgical centers as well as hospitals).

Fifth, CMS rule-makers could set reporting standards and contextualize price information in common clinical scenarios. Patients may have difficulty shopping for complex healthcare services without understanding how they apply in different clinical situations. Decision-making would also be aided by reporting standards—for instance, for how prices are displayed and whether they include certain fees (eg, professional fees, pathology studies).

Hospitalists planning follow-up care should inform patients that price information is increasingly available and encourage them to search on the internet or contact hospital billing offices to request information (eg, discounted cash prices and minimum negotiated charges) before obtaining elective services after discharge. Hospitalists can also encourage patients to discuss shoppable services with their primary care physicians to understand the clinical context and make high-value decisions. Hospitalists who wish to build communication skills discussing costs with patients can increasingly find resources for these conversations and request that prices be displayed in the electronic health record for this purpose.^13,14 As conversations occur, hospitalists should seek to understand other factors, such as convenience and continuity relationships, that might influence choices.

Starting in 2021, CMS policy requires that hospitals report prices for services such as the endoscopy recommended for the patient in the scenario. Though the policy gives patients new hope for greater transparency and better prices, additional steps are needed to help patients and hospitalists achieve these benefits.

Orbital cellulitis/abscess (OCA) is a potential complication of sinusitis. If not treated promptly, it can result in vision loss, intracranial infection, or cavernous sinus thrombosis.^1,2 In 1970, Chandler et al³ classified orbital complications of acute sinusitis into five groups: inflammatory edema (group 1); orbital cellulitis (group 2); subperiosteal abscess (SPA) (group 3); orbital abscess (group 4); and cavernous sinus thrombosis (group 5). Group 1, or preseptal cellulitis, is significantly different from groups 2, 3, and 4, collectively referred to as OCA, which affect the actual orbital content.

Children with OCA are generally hospitalized so they can be treated with intravenous antibiotics. While orbital abscesses (group 4) are typically treated surgically, successful medical management has been reported for cases of orbital cellulitis and SPA (groups 2 and 3).^4,5 No widely accepted guidelines exist for the evaluation and medical management of OCA, resulting in significant variation in care.⁶ The purpose of this systematic review is to summarize existing evidence guiding the medical management of OCA regarding laboratory testing, imaging, and microbiology. This review does not address surgical considerations.

The review protocol has been registered in the PROSPERO International Prospective Register of Systematic Reviews (crd.york.ac.uk/prospero/index.asp; identifier: CRD42020158463), and the review was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.⁷

Search Strategy

A systematic search of the literature was designed and conducted by a medical librarian (ES), with input from the research team (AB, SM). The search strategy included Medical Subject Headings (MeSH) terms and keywords related to orbital or subperiosteal cellulitis/abscess and children; see Appendix Table 1 for the complete search strategy. Searches were conducted in MEDLINE (Ovid), Web of Science Core Collection, Scopus, CINAHL (EBSCO), and Cochrane Central Register of Controlled Trials (CENTRAL) using advanced search techniques relative to each database. Searches were last conducted on February 9, 2021.

Eligibility Criteria

The study designs (retrospective and prospective) included in the search were limited to randomized clinical trials, cohort studies, case-control studies, and case series with participants <18 years of age. Case reports describing fewer than 5 patients and literature reviews were excluded. Studies including a combination of adult and pediatric patients were included if pediatric outcomes were reported separately. Only studies available in English were included.

Outcome Measures

The outcome measures were determined a priori based on three clinical questions:

• Q1. What is the role of inflammatory markers—white blood cell (WBC) count, C-reactive protein (CRP), and fever—in distinguishing between the following: preseptal cellulitis (group 1) and OCA (groups 2, 3, and 4); orbital cellulitis (group 2) and abscess (groups 3 and 4); and patients who do and do not require surgery?
• Q2. What is the role of imaging in the evaluation of OCA?
• Q3. What is the microbiology of OCA over the past 2 decades? What is the prevalence of methicillin-resistant Staphylococcus aureus (MRSA)?

Screening

Two review authors (AB, SM) performed both the title/abstract and full-text screen, independently applying the eligibility criteria. Disagreements were discussed, and conflicts were resolved with input from a third reviewer author (ES). Duplications were removed. When two studies had overlapping patient data, the study with fewer data points was excluded.

Data Extraction and Synthesis

All studies included after the full-text screen were divided based on the clinical question they answered (Q1, Q2, Q3 above). Some studies reported outcomes pertinent to more than one question. Two review authors were assigned to each clinical question. They independently reviewed each article and extracted the pertinent data into question-specific extraction sheets. Articles assigned to Q2 were reviewed by two pediatric neuroradiologists. For each study, the following details were extracted: authors, location, year, study type, study period, population, and number and ages of participants. Details that were question-specific included: (Q1) values and/or percentages for inflammatory markers; (Q2) reasons for imaging or type of imaging; and (Q3) participants managed surgically and culture results. The data were then synthesized in table and/or narrative format. For Q3, the organisms identified from intraoperative and blood cultures in each study were mathematically combined. When possible, prevalence was calculated using the number of patients with at least one pathogen recovered as the denominator. If this number was not available, the number of patients who underwent surgery was used as the denominator.

Quality Assessment

No randomized controlled trials were identified. More than 90% of the studies identified and included were retrospective descriptive studies. By the nature of the case series design, the study quality was felt to be poor, with high risk of bias. The Joanna Briggs Institute Critical Appraisal tools for systematic reviews were used to appraise each individual study included (Appendix Table 2).⁸ The Grading of Recommendations, Assessment, Development and Evaluations (GRADE) criteria were used in rating the quality of evidence for each question.⁹

A summary of the search strategy and study selection is provided in the Figure (PRISMA flow diagram). The initial search identified 3007 studies. After duplicates were removed and general eligibility criteria applied, 94 articles remained. Question-specific eligibility criteria, discussed in the following sections, were then applied, resulting in 63 articles included in the review.

Q1: Are Inflammatory Markers, Including Fever, WBC, and CRP, Useful in Distinguishing Preseptal Cellulitis (group 1) From OCA (Groups 2, 3, and 4); Orbital Cellulitis (group 2) From Abscess (Groups 3 and 4); or Identifying Patients Who Require Surgical Intervention?

Fever and elevation of the WBC count and CRP have been used to assess the severity of certain pediatric infections^10,11 and therefore may be helpful in distinguishing severity of illness in OCA. Studies included in this section provided numerical values for at least one of the following: WBC count, CRP, or percentage of patients with fever for at least one type of orbital infection. Included studies had at least five patients per group.

Thirty-three articles were screened for the inflammatory marker section. Thirteen were excluded for the following reasons: no numbers reported for inflammatory markers (n = 6); group 1 and groups 2, 3, and 4 results combined (n = 6); fewer than five patients with orbital cellulitis included (n = 1). Twenty studies were included: 18 case series and 2 retrospective cohorts. Appendix Table 3 summarizes the data from studies included. Based on GRADE criteria, the body of evidence included in this section is of low quality.⁹

Distinguishing Between Preseptal and OCA

Eleven studies were included in this section (Table 1). WBC count was significantly higher in patients with groups 2, 3, and 4 than group 1 in two studies (Devrim et al,¹²P < .01; Santos et al,¹³P = .025). CRP was significantly higher in patients with groups 2, 3, and 4 than group 1 in four studies (Öcal Demir et al,¹⁴P = .02; Devrim et al,¹²P < .01; Ohana-Sarna-Cahan et al,¹⁸P < .001; Santos et al,¹³P < .001). Patients with groups 2, 3, and 4 had a significantly higher fever rate in three studies (Botting et al,²¹P < .001; Ohana-Sarna-Cahan et al,¹⁸P = .0001; Santos et al,¹³ P = .029).

Distinguishing Between Orbital Cellulitis and Abscess

Seven studies were included in this section (Appendix Table 3). One study showed significantly higher WBC count in group 3 than group 2 (P = .004), although results were reported as percentage of patients above a cutoff number calculated to distinguish between cellulitis and abscess (Appendix Table 3).²² CRP was not significantly different between group 2 and groups 3 and 4. One study found a significantly higher fever rate in patients with group 3 compared to patients with group 2 (P < .001).²²

Identifying Patients Requiring Surgery

Six studies were included in this section (Appendix Table 3). One study found a significantly higher WBC count in patients treated surgically (Tabarino et al,²⁴P < .05). Patients treated surgically had a significantly higher CRP in two studies (Cohen et al,²⁵P = .02; Friling et al,²⁶ P = .04). Fever was inconsistently reported in the studies, with some using mean presenting temperatures and some using rates of fever. One study found a significantly higher mean presenting temperature in patients treated surgically (P = .027), but the difference between the two groups was 0.7 °C.²³

Summary

Most studies found no significant difference in WBC count, CRP, or fever between preseptal and OCA, cellulitis and abscess, or patients receiving medical and surgical interventions.

Q2: What Is the Role of Imaging in Evaluation of OCA?

Twenty-five articles were selected for the imaging section review. All the included studies were retrospective descriptive studies. Quantitative data extraction and analysis of these studies could not be performed because of their heterogeneous methodologies and lack of objective data. Therefore, the information gleaned from these studies is summarized in narrative format. Per GRADE criteria, the body of evidence included in this section is of low quality.

Who Needs Imaging?

Proptosis, ophthalmoplegia, decreased vision, and pain with eye movements are widely agreed-upon indications for imaging evaluation.^21,27,28 Because of concern for radiation exposure in pediatric patients, some authors suggested that computed tomography (CT) should only be obtained if patients fail to respond to medical therapy or if surgery is being considered.^17,29,30 However, Rudloe et al³¹ found that half of the patients with group 3 or higher disease on CT did not have proptosis, ophthalmoplegia, or pain with extraocular movement. In addition, evaluation of young children with acute periorbital swelling can be difficult, so a lower threshold for imaging is likely warranted in these younger patients.

What Type of Imaging Should Be Obtained?

The American College of Radiology 2018 Appropriateness Criteria (ACR criteria) for orbital imaging state that orbital CT is usually indicated for patients with suspected Chandler groups 2, 3, and 4 infections.³² CT with contrast is useful for evaluating the extent of orbital infection and size of the abscess and for delineating the adjacent osseous anatomy, which is essential for cases in which surgical intervention is planned.^{20,21,26,27,30,31,33,34} Distinguishing abscess from cellulitis on CT sometimes can be challenging; therefore, serial clinical examinations and, occasionally, surgical exploration may be required.^35,36

Magnetic resonance imaging (MRI) is helpful for evaluating intracranial complications (eg, epidural abscess),^27,37 but it is limited for evaluating the osseous components of the paranasal sinuses. Although one study suggested that rapid MRI is comparable to contrast CT for differentiating group 1 infections from groups 2, 3, and 4 infections, it provided limited assessment of other complications.³⁸ With no definitive studies comparing CT with MRI for orbital infections, adherence to the ACR criteria is recommended.

Orbital ultrasound is limited by its small field of view and artifact produced by the surrounding bony interface, both of which can obscure posterior intraorbital pathologies.^29,39,40 Plain radiographs are not helpful for evaluating OCA due to limited soft-tissue contrast.⁴¹

When Should Repeat Imaging Be Obtained?

Children with group 3 OCA have been successfully managed medically in a carefully monitored setting.⁴² Repeat CT imaging is sometimes useful in these patients, particularly if the clinical examination is difficult.^42-44 However, improvement in CT findings may lag behind clinical improvement.³⁹

Summary

Per ACR criteria, orbital CT with contrast is recommended to evaluate patients with suspected Chandler groups 2, 3, and 4 OCA. MRI is reserved for evaluating intracranial complications.

Q3: What Is the Microbiology of OCA? What Is the MRSA Prevalence?

Knowledge of the microbiology of OCA is essential for the appropriate selection of empiric antibiotics. Because fewer children with groups 2 and 3 OCA undergo surgery, intraoperative cultures often are not available to guide antibiotic selection.⁴⁵ As a result, significant variation exists in antibiotic prescribing.⁶

Studies discussing the microbiology of OCA were included only if they were published in the past 2 decades (2000-2020) and were excluded if the study period was before 1990, as microbiology changes over time and new vaccines are introduced. To be included, the majority of cultures reported had to be intraoperative (orbital or sinus) specimens. Studies reporting only nasal, conjunctival, or other surface cultures were excluded. When studies included patients with group 1 OCA, only microbiology data for groups 2, 3, and 4 OCA were extracted. The pattern of resistance for S aureus was not always explicitly reported; however, when non-MRSA active antibiotics were used, methicillin-susceptible S aureus was assumed. 

A total of 63 studies were screened for the microbiology section; 32 were excluded for the following reasons: published before 2000 or study period before 1990 (n = 18), reported surface cultures or culture site not clearly stated (n = 4), microbiology mixed between preseptal and orbital (n = 6), wrong study type (n = 2), and study group overlaps with a different article included (n = 2). Of the 32 studies included, 3 were prospective observational, 4 were retrospective cohort, and 25 were case series. Based on GRADE criteria, the body of evidence included in this section is of low quality.⁴²

Appendix Table 4 summarizes the microbiologic data from the studies included. In the group of children that had a positive culture (orbital, sinus, or blood), the most commonly recovered organisms reported were S aureus (median, 22%; range, 0%-100%), Streptococcus anginosus group (median, 16%; range, 0%-100%), group A Streptococcus (median, 12%; range, 0%-80%), and Streptococcus pneumoniae (median, 8%; range, 0%-100%). Streptococcus as a group had a median prevalence of 57%, ranging from 0% to 100%. MRSA prevalence had a median of 3% (interquartile range [IQR], 0%-13%). Median prevalence of polymicrobial cultures was 20%, and median prevalence of anaerobic organisms was 14% (Table 2). Orbital and sinus cultures had the highest yield, with an average return of an organism of 72% (median, 75%; IQR, 64%-84%). Blood culture results were reported in 14 studies and usually obtained in a subgroup of the study population. When obtained, blood cultures rarely yielded an organism (median, 10%; IQR, 5%-15%); the rate of identified bacteremia in the total population had a median of 5% (IQR, 5%-7%) across studies.

Microbiology was compared between studies completed in the United States and in other countries (Table 2). Based on median prevalence across studies, both S anginosus group and MRSA were more prevalent in the United States than internationally (28% vs 0% and 11% vs 0%, respectively). No clear trend in MRSA prevalence was evident over the 2 decades; however, the studies included were heterogeneous and did not have the power to detect such a trend.

Two reports suggest a difference of MRSA prevalence by patient age. Hsu et al⁴⁶ found that three of eight MRSA infections were in infants age <1 year, which accounted for 50% (3/6) of infants included in the study. Miller et al⁴⁷ reported MRSA in 4 of 9 (44%) infants with OCA. Age <1 year may be associated with increased frequency of MRSA infection in OCA.

Summary

Blood cultures have low yield. The most common organisms recovered from OCA are Streptococcus species (most commonly S anginosus group, group A Streptococcus, and pneumococcus) and S aureus. Polymicrobial infections including anaerobes are common. MRSA prevalence is low globally but varies significantly among geographic areas.

Our systematic review of the literature for the medical management of OCA revealed predominantly descriptive studies and only a limited number of comparison-based studies, likely reflecting the rarity of advanced forms of OCA. Given the lack of high-quality evidence and the level of heterogeneity among studies, the conclusions that can be drawn are limited.

Distinguishing between disease severity and OCA requiring surgical intervention remains challenging. Although studies in our review suggest a trend toward markers of inflammation (fever, elevated WBC count and CRP) being more common in more severe presentations, the results were mixed, and studies were low quality and underpowered to detect meaningful differences. For example, most studies do not define what constitutes a fever in their cohort. Our review suggests that markers of inflammation cannot be used to distinguish between Chandler groups or to identify patients requiring surgery. Of note, the presence of fever and elevated inflammatory markers may have influenced the decision to obtain imaging or to proceed to surgery, thereby also potentially biasing these clinical indicators toward predictors for more severe disease. Decisions regarding surgery should therefore be based on the entire clinical picture, including response to appropriate antibiotics.

We found a lack of high-quality evidence regarding the role of imaging in OCA, and the studies reviewed were heterogeneous. Recommendations for imaging therefore remain at the level of expert opinion (ACR criteria). CT imaging is the first-line modality for imaging in suspected OCA given the limitations of alternative imaging modalities, but the sensitivity and specificity of CT imaging remain unknown for diagnosis of orbital abscesses.

Our review of the published microbiology confirmed that Staphylococcus and Streptococcus species are the most common pathogens identified in OCA. Prevalence across the different studies varied greatly. Owing to the significant heterogeneity in studies, calculation of pooled prevalence was not possible. By using the number of positive cultures as our denominator (or total surgeries if number of positive cultures was unavailable), we likely overestimated the prevalence of S aureus. S aureus is generally recognized as a pyogenic pathogen, more likely to be associated with abscess formation.⁴⁸ Therefore, culture results obtained predominantly from abscesses likely result in an overestimate of S aureus in OCA (groups 2, 3, and 4). Regardless, MRSA prevalence was generally low, both nationally and internationally. The MRSA results from the study by McKinley at el⁴⁹ (Texas) was a notable outlier in the United States, with MRSA prevalence as high as 44% compared with the median prevalence of 3% (IQR, 0-13), highlighting the importance of local resistance patterns when choosing empiric antibiotics.

Limitations to the microbiology review included significant heterogeneity in both the types of cultures included and the reporting of results. Although we excluded studies that reported only surface culture results or did not specify culture type, we did include studies that had surface culture results combined with intraoperative culture results, making it impossible to separate the two. Since most of the cultures included in combined results reported organisms based on intraoperative cultures, we felt they provided valuable information that should be included. In most studies, blood cultures were not obtained in all participants, so the yield of blood cultures is likely an overestimate, as blood cultures are more likely to be obtained in higher-acuity patients.

Although the available evidence regarding the medical management of OCA remains low quality, certain limited conclusions can be drawn, as presented in this review. Further high-quality studies are needed to better inform the medical management of OCA.

The authors thank Dr Kyle Pronko for his help with data extraction for the imaging section.

Orbital cellulitis/abscess (OCA) is a potential complication of sinusitis. If not treated promptly, it can result in vision loss, intracranial infection, or cavernous sinus thrombosis.^1,2 In 1970, Chandler et al³ classified orbital complications of acute sinusitis into five groups: inflammatory edema (group 1); orbital cellulitis (group 2); subperiosteal abscess (SPA) (group 3); orbital abscess (group 4); and cavernous sinus thrombosis (group 5). Group 1, or preseptal cellulitis, is significantly different from groups 2, 3, and 4, collectively referred to as OCA, which affect the actual orbital content.

Children with OCA are generally hospitalized so they can be treated with intravenous antibiotics. While orbital abscesses (group 4) are typically treated surgically, successful medical management has been reported for cases of orbital cellulitis and SPA (groups 2 and 3).^4,5 No widely accepted guidelines exist for the evaluation and medical management of OCA, resulting in significant variation in care.⁶ The purpose of this systematic review is to summarize existing evidence guiding the medical management of OCA regarding laboratory testing, imaging, and microbiology. This review does not address surgical considerations.

The review protocol has been registered in the PROSPERO International Prospective Register of Systematic Reviews (crd.york.ac.uk/prospero/index.asp; identifier: CRD42020158463), and the review was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.⁷

Search Strategy

A systematic search of the literature was designed and conducted by a medical librarian (ES), with input from the research team (AB, SM). The search strategy included Medical Subject Headings (MeSH) terms and keywords related to orbital or subperiosteal cellulitis/abscess and children; see Appendix Table 1 for the complete search strategy. Searches were conducted in MEDLINE (Ovid), Web of Science Core Collection, Scopus, CINAHL (EBSCO), and Cochrane Central Register of Controlled Trials (CENTRAL) using advanced search techniques relative to each database. Searches were last conducted on February 9, 2021.

Eligibility Criteria

The study designs (retrospective and prospective) included in the search were limited to randomized clinical trials, cohort studies, case-control studies, and case series with participants <18 years of age. Case reports describing fewer than 5 patients and literature reviews were excluded. Studies including a combination of adult and pediatric patients were included if pediatric outcomes were reported separately. Only studies available in English were included.

Outcome Measures

The outcome measures were determined a priori based on three clinical questions:

• Q1. What is the role of inflammatory markers—white blood cell (WBC) count, C-reactive protein (CRP), and fever—in distinguishing between the following: preseptal cellulitis (group 1) and OCA (groups 2, 3, and 4); orbital cellulitis (group 2) and abscess (groups 3 and 4); and patients who do and do not require surgery?
• Q2. What is the role of imaging in the evaluation of OCA?
• Q3. What is the microbiology of OCA over the past 2 decades? What is the prevalence of methicillin-resistant Staphylococcus aureus (MRSA)?

Screening

Two review authors (AB, SM) performed both the title/abstract and full-text screen, independently applying the eligibility criteria. Disagreements were discussed, and conflicts were resolved with input from a third reviewer author (ES). Duplications were removed. When two studies had overlapping patient data, the study with fewer data points was excluded.

Data Extraction and Synthesis

All studies included after the full-text screen were divided based on the clinical question they answered (Q1, Q2, Q3 above). Some studies reported outcomes pertinent to more than one question. Two review authors were assigned to each clinical question. They independently reviewed each article and extracted the pertinent data into question-specific extraction sheets. Articles assigned to Q2 were reviewed by two pediatric neuroradiologists. For each study, the following details were extracted: authors, location, year, study type, study period, population, and number and ages of participants. Details that were question-specific included: (Q1) values and/or percentages for inflammatory markers; (Q2) reasons for imaging or type of imaging; and (Q3) participants managed surgically and culture results. The data were then synthesized in table and/or narrative format. For Q3, the organisms identified from intraoperative and blood cultures in each study were mathematically combined. When possible, prevalence was calculated using the number of patients with at least one pathogen recovered as the denominator. If this number was not available, the number of patients who underwent surgery was used as the denominator.

Quality Assessment

No randomized controlled trials were identified. More than 90% of the studies identified and included were retrospective descriptive studies. By the nature of the case series design, the study quality was felt to be poor, with high risk of bias. The Joanna Briggs Institute Critical Appraisal tools for systematic reviews were used to appraise each individual study included (Appendix Table 2).⁸ The Grading of Recommendations, Assessment, Development and Evaluations (GRADE) criteria were used in rating the quality of evidence for each question.⁹

A summary of the search strategy and study selection is provided in the Figure (PRISMA flow diagram). The initial search identified 3007 studies. After duplicates were removed and general eligibility criteria applied, 94 articles remained. Question-specific eligibility criteria, discussed in the following sections, were then applied, resulting in 63 articles included in the review.

Q1: Are Inflammatory Markers, Including Fever, WBC, and CRP, Useful in Distinguishing Preseptal Cellulitis (group 1) From OCA (Groups 2, 3, and 4); Orbital Cellulitis (group 2) From Abscess (Groups 3 and 4); or Identifying Patients Who Require Surgical Intervention?

Fever and elevation of the WBC count and CRP have been used to assess the severity of certain pediatric infections^10,11 and therefore may be helpful in distinguishing severity of illness in OCA. Studies included in this section provided numerical values for at least one of the following: WBC count, CRP, or percentage of patients with fever for at least one type of orbital infection. Included studies had at least five patients per group.

Thirty-three articles were screened for the inflammatory marker section. Thirteen were excluded for the following reasons: no numbers reported for inflammatory markers (n = 6); group 1 and groups 2, 3, and 4 results combined (n = 6); fewer than five patients with orbital cellulitis included (n = 1). Twenty studies were included: 18 case series and 2 retrospective cohorts. Appendix Table 3 summarizes the data from studies included. Based on GRADE criteria, the body of evidence included in this section is of low quality.⁹

Distinguishing Between Preseptal and OCA

Eleven studies were included in this section (Table 1). WBC count was significantly higher in patients with groups 2, 3, and 4 than group 1 in two studies (Devrim et al,¹²P < .01; Santos et al,¹³P = .025). CRP was significantly higher in patients with groups 2, 3, and 4 than group 1 in four studies (Öcal Demir et al,¹⁴P = .02; Devrim et al,¹²P < .01; Ohana-Sarna-Cahan et al,¹⁸P < .001; Santos et al,¹³P < .001). Patients with groups 2, 3, and 4 had a significantly higher fever rate in three studies (Botting et al,²¹P < .001; Ohana-Sarna-Cahan et al,¹⁸P = .0001; Santos et al,¹³ P = .029).

Distinguishing Between Orbital Cellulitis and Abscess

Seven studies were included in this section (Appendix Table 3). One study showed significantly higher WBC count in group 3 than group 2 (P = .004), although results were reported as percentage of patients above a cutoff number calculated to distinguish between cellulitis and abscess (Appendix Table 3).²² CRP was not significantly different between group 2 and groups 3 and 4. One study found a significantly higher fever rate in patients with group 3 compared to patients with group 2 (P < .001).²²

Identifying Patients Requiring Surgery

Six studies were included in this section (Appendix Table 3). One study found a significantly higher WBC count in patients treated surgically (Tabarino et al,²⁴P < .05). Patients treated surgically had a significantly higher CRP in two studies (Cohen et al,²⁵P = .02; Friling et al,²⁶ P = .04). Fever was inconsistently reported in the studies, with some using mean presenting temperatures and some using rates of fever. One study found a significantly higher mean presenting temperature in patients treated surgically (P = .027), but the difference between the two groups was 0.7 °C.²³

Summary

Most studies found no significant difference in WBC count, CRP, or fever between preseptal and OCA, cellulitis and abscess, or patients receiving medical and surgical interventions.

Q2: What Is the Role of Imaging in Evaluation of OCA?

Twenty-five articles were selected for the imaging section review. All the included studies were retrospective descriptive studies. Quantitative data extraction and analysis of these studies could not be performed because of their heterogeneous methodologies and lack of objective data. Therefore, the information gleaned from these studies is summarized in narrative format. Per GRADE criteria, the body of evidence included in this section is of low quality.

Who Needs Imaging?

Proptosis, ophthalmoplegia, decreased vision, and pain with eye movements are widely agreed-upon indications for imaging evaluation.^21,27,28 Because of concern for radiation exposure in pediatric patients, some authors suggested that computed tomography (CT) should only be obtained if patients fail to respond to medical therapy or if surgery is being considered.^17,29,30 However, Rudloe et al³¹ found that half of the patients with group 3 or higher disease on CT did not have proptosis, ophthalmoplegia, or pain with extraocular movement. In addition, evaluation of young children with acute periorbital swelling can be difficult, so a lower threshold for imaging is likely warranted in these younger patients.

What Type of Imaging Should Be Obtained?

The American College of Radiology 2018 Appropriateness Criteria (ACR criteria) for orbital imaging state that orbital CT is usually indicated for patients with suspected Chandler groups 2, 3, and 4 infections.³² CT with contrast is useful for evaluating the extent of orbital infection and size of the abscess and for delineating the adjacent osseous anatomy, which is essential for cases in which surgical intervention is planned.^{20,21,26,27,30,31,33,34} Distinguishing abscess from cellulitis on CT sometimes can be challenging; therefore, serial clinical examinations and, occasionally, surgical exploration may be required.^35,36

Magnetic resonance imaging (MRI) is helpful for evaluating intracranial complications (eg, epidural abscess),^27,37 but it is limited for evaluating the osseous components of the paranasal sinuses. Although one study suggested that rapid MRI is comparable to contrast CT for differentiating group 1 infections from groups 2, 3, and 4 infections, it provided limited assessment of other complications.³⁸ With no definitive studies comparing CT with MRI for orbital infections, adherence to the ACR criteria is recommended.

Orbital ultrasound is limited by its small field of view and artifact produced by the surrounding bony interface, both of which can obscure posterior intraorbital pathologies.^29,39,40 Plain radiographs are not helpful for evaluating OCA due to limited soft-tissue contrast.⁴¹

When Should Repeat Imaging Be Obtained?

Children with group 3 OCA have been successfully managed medically in a carefully monitored setting.⁴² Repeat CT imaging is sometimes useful in these patients, particularly if the clinical examination is difficult.^42-44 However, improvement in CT findings may lag behind clinical improvement.³⁹

Summary

Per ACR criteria, orbital CT with contrast is recommended to evaluate patients with suspected Chandler groups 2, 3, and 4 OCA. MRI is reserved for evaluating intracranial complications.

Q3: What Is the Microbiology of OCA? What Is the MRSA Prevalence?

Knowledge of the microbiology of OCA is essential for the appropriate selection of empiric antibiotics. Because fewer children with groups 2 and 3 OCA undergo surgery, intraoperative cultures often are not available to guide antibiotic selection.⁴⁵ As a result, significant variation exists in antibiotic prescribing.⁶

Studies discussing the microbiology of OCA were included only if they were published in the past 2 decades (2000-2020) and were excluded if the study period was before 1990, as microbiology changes over time and new vaccines are introduced. To be included, the majority of cultures reported had to be intraoperative (orbital or sinus) specimens. Studies reporting only nasal, conjunctival, or other surface cultures were excluded. When studies included patients with group 1 OCA, only microbiology data for groups 2, 3, and 4 OCA were extracted. The pattern of resistance for S aureus was not always explicitly reported; however, when non-MRSA active antibiotics were used, methicillin-susceptible S aureus was assumed. 

A total of 63 studies were screened for the microbiology section; 32 were excluded for the following reasons: published before 2000 or study period before 1990 (n = 18), reported surface cultures or culture site not clearly stated (n = 4), microbiology mixed between preseptal and orbital (n = 6), wrong study type (n = 2), and study group overlaps with a different article included (n = 2). Of the 32 studies included, 3 were prospective observational, 4 were retrospective cohort, and 25 were case series. Based on GRADE criteria, the body of evidence included in this section is of low quality.⁴²

Appendix Table 4 summarizes the microbiologic data from the studies included. In the group of children that had a positive culture (orbital, sinus, or blood), the most commonly recovered organisms reported were S aureus (median, 22%; range, 0%-100%), Streptococcus anginosus group (median, 16%; range, 0%-100%), group A Streptococcus (median, 12%; range, 0%-80%), and Streptococcus pneumoniae (median, 8%; range, 0%-100%). Streptococcus as a group had a median prevalence of 57%, ranging from 0% to 100%. MRSA prevalence had a median of 3% (interquartile range [IQR], 0%-13%). Median prevalence of polymicrobial cultures was 20%, and median prevalence of anaerobic organisms was 14% (Table 2). Orbital and sinus cultures had the highest yield, with an average return of an organism of 72% (median, 75%; IQR, 64%-84%). Blood culture results were reported in 14 studies and usually obtained in a subgroup of the study population. When obtained, blood cultures rarely yielded an organism (median, 10%; IQR, 5%-15%); the rate of identified bacteremia in the total population had a median of 5% (IQR, 5%-7%) across studies.

Microbiology was compared between studies completed in the United States and in other countries (Table 2). Based on median prevalence across studies, both S anginosus group and MRSA were more prevalent in the United States than internationally (28% vs 0% and 11% vs 0%, respectively). No clear trend in MRSA prevalence was evident over the 2 decades; however, the studies included were heterogeneous and did not have the power to detect such a trend.

Two reports suggest a difference of MRSA prevalence by patient age. Hsu et al⁴⁶ found that three of eight MRSA infections were in infants age <1 year, which accounted for 50% (3/6) of infants included in the study. Miller et al⁴⁷ reported MRSA in 4 of 9 (44%) infants with OCA. Age <1 year may be associated with increased frequency of MRSA infection in OCA.

Summary

Blood cultures have low yield. The most common organisms recovered from OCA are Streptococcus species (most commonly S anginosus group, group A Streptococcus, and pneumococcus) and S aureus. Polymicrobial infections including anaerobes are common. MRSA prevalence is low globally but varies significantly among geographic areas.

Our systematic review of the literature for the medical management of OCA revealed predominantly descriptive studies and only a limited number of comparison-based studies, likely reflecting the rarity of advanced forms of OCA. Given the lack of high-quality evidence and the level of heterogeneity among studies, the conclusions that can be drawn are limited.

Distinguishing between disease severity and OCA requiring surgical intervention remains challenging. Although studies in our review suggest a trend toward markers of inflammation (fever, elevated WBC count and CRP) being more common in more severe presentations, the results were mixed, and studies were low quality and underpowered to detect meaningful differences. For example, most studies do not define what constitutes a fever in their cohort. Our review suggests that markers of inflammation cannot be used to distinguish between Chandler groups or to identify patients requiring surgery. Of note, the presence of fever and elevated inflammatory markers may have influenced the decision to obtain imaging or to proceed to surgery, thereby also potentially biasing these clinical indicators toward predictors for more severe disease. Decisions regarding surgery should therefore be based on the entire clinical picture, including response to appropriate antibiotics.

We found a lack of high-quality evidence regarding the role of imaging in OCA, and the studies reviewed were heterogeneous. Recommendations for imaging therefore remain at the level of expert opinion (ACR criteria). CT imaging is the first-line modality for imaging in suspected OCA given the limitations of alternative imaging modalities, but the sensitivity and specificity of CT imaging remain unknown for diagnosis of orbital abscesses.

Our review of the published microbiology confirmed that Staphylococcus and Streptococcus species are the most common pathogens identified in OCA. Prevalence across the different studies varied greatly. Owing to the significant heterogeneity in studies, calculation of pooled prevalence was not possible. By using the number of positive cultures as our denominator (or total surgeries if number of positive cultures was unavailable), we likely overestimated the prevalence of S aureus. S aureus is generally recognized as a pyogenic pathogen, more likely to be associated with abscess formation.⁴⁸ Therefore, culture results obtained predominantly from abscesses likely result in an overestimate of S aureus in OCA (groups 2, 3, and 4). Regardless, MRSA prevalence was generally low, both nationally and internationally. The MRSA results from the study by McKinley at el⁴⁹ (Texas) was a notable outlier in the United States, with MRSA prevalence as high as 44% compared with the median prevalence of 3% (IQR, 0-13), highlighting the importance of local resistance patterns when choosing empiric antibiotics.

Limitations to the microbiology review included significant heterogeneity in both the types of cultures included and the reporting of results. Although we excluded studies that reported only surface culture results or did not specify culture type, we did include studies that had surface culture results combined with intraoperative culture results, making it impossible to separate the two. Since most of the cultures included in combined results reported organisms based on intraoperative cultures, we felt they provided valuable information that should be included. In most studies, blood cultures were not obtained in all participants, so the yield of blood cultures is likely an overestimate, as blood cultures are more likely to be obtained in higher-acuity patients.

Although the available evidence regarding the medical management of OCA remains low quality, certain limited conclusions can be drawn, as presented in this review. Further high-quality studies are needed to better inform the medical management of OCA.

The authors thank Dr Kyle Pronko for his help with data extraction for the imaging section.

Orbital cellulitis/abscess (OCA) is a potential complication of sinusitis. If not treated promptly, it can result in vision loss, intracranial infection, or cavernous sinus thrombosis.^1,2 In 1970, Chandler et al³ classified orbital complications of acute sinusitis into five groups: inflammatory edema (group 1); orbital cellulitis (group 2); subperiosteal abscess (SPA) (group 3); orbital abscess (group 4); and cavernous sinus thrombosis (group 5). Group 1, or preseptal cellulitis, is significantly different from groups 2, 3, and 4, collectively referred to as OCA, which affect the actual orbital content.

Children with OCA are generally hospitalized so they can be treated with intravenous antibiotics. While orbital abscesses (group 4) are typically treated surgically, successful medical management has been reported for cases of orbital cellulitis and SPA (groups 2 and 3).^4,5 No widely accepted guidelines exist for the evaluation and medical management of OCA, resulting in significant variation in care.⁶ The purpose of this systematic review is to summarize existing evidence guiding the medical management of OCA regarding laboratory testing, imaging, and microbiology. This review does not address surgical considerations.

The review protocol has been registered in the PROSPERO International Prospective Register of Systematic Reviews (crd.york.ac.uk/prospero/index.asp; identifier: CRD42020158463), and the review was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.⁷

Search Strategy

A systematic search of the literature was designed and conducted by a medical librarian (ES), with input from the research team (AB, SM). The search strategy included Medical Subject Headings (MeSH) terms and keywords related to orbital or subperiosteal cellulitis/abscess and children; see Appendix Table 1 for the complete search strategy. Searches were conducted in MEDLINE (Ovid), Web of Science Core Collection, Scopus, CINAHL (EBSCO), and Cochrane Central Register of Controlled Trials (CENTRAL) using advanced search techniques relative to each database. Searches were last conducted on February 9, 2021.

Eligibility Criteria

The study designs (retrospective and prospective) included in the search were limited to randomized clinical trials, cohort studies, case-control studies, and case series with participants <18 years of age. Case reports describing fewer than 5 patients and literature reviews were excluded. Studies including a combination of adult and pediatric patients were included if pediatric outcomes were reported separately. Only studies available in English were included.

Outcome Measures

The outcome measures were determined a priori based on three clinical questions:

• Q1. What is the role of inflammatory markers—white blood cell (WBC) count, C-reactive protein (CRP), and fever—in distinguishing between the following: preseptal cellulitis (group 1) and OCA (groups 2, 3, and 4); orbital cellulitis (group 2) and abscess (groups 3 and 4); and patients who do and do not require surgery?
• Q2. What is the role of imaging in the evaluation of OCA?
• Q3. What is the microbiology of OCA over the past 2 decades? What is the prevalence of methicillin-resistant Staphylococcus aureus (MRSA)?

Screening

Two review authors (AB, SM) performed both the title/abstract and full-text screen, independently applying the eligibility criteria. Disagreements were discussed, and conflicts were resolved with input from a third reviewer author (ES). Duplications were removed. When two studies had overlapping patient data, the study with fewer data points was excluded.

Data Extraction and Synthesis

All studies included after the full-text screen were divided based on the clinical question they answered (Q1, Q2, Q3 above). Some studies reported outcomes pertinent to more than one question. Two review authors were assigned to each clinical question. They independently reviewed each article and extracted the pertinent data into question-specific extraction sheets. Articles assigned to Q2 were reviewed by two pediatric neuroradiologists. For each study, the following details were extracted: authors, location, year, study type, study period, population, and number and ages of participants. Details that were question-specific included: (Q1) values and/or percentages for inflammatory markers; (Q2) reasons for imaging or type of imaging; and (Q3) participants managed surgically and culture results. The data were then synthesized in table and/or narrative format. For Q3, the organisms identified from intraoperative and blood cultures in each study were mathematically combined. When possible, prevalence was calculated using the number of patients with at least one pathogen recovered as the denominator. If this number was not available, the number of patients who underwent surgery was used as the denominator.

Quality Assessment

No randomized controlled trials were identified. More than 90% of the studies identified and included were retrospective descriptive studies. By the nature of the case series design, the study quality was felt to be poor, with high risk of bias. The Joanna Briggs Institute Critical Appraisal tools for systematic reviews were used to appraise each individual study included (Appendix Table 2).⁸ The Grading of Recommendations, Assessment, Development and Evaluations (GRADE) criteria were used in rating the quality of evidence for each question.⁹

A summary of the search strategy and study selection is provided in the Figure (PRISMA flow diagram). The initial search identified 3007 studies. After duplicates were removed and general eligibility criteria applied, 94 articles remained. Question-specific eligibility criteria, discussed in the following sections, were then applied, resulting in 63 articles included in the review.

Q1: Are Inflammatory Markers, Including Fever, WBC, and CRP, Useful in Distinguishing Preseptal Cellulitis (group 1) From OCA (Groups 2, 3, and 4); Orbital Cellulitis (group 2) From Abscess (Groups 3 and 4); or Identifying Patients Who Require Surgical Intervention?

Fever and elevation of the WBC count and CRP have been used to assess the severity of certain pediatric infections^10,11 and therefore may be helpful in distinguishing severity of illness in OCA. Studies included in this section provided numerical values for at least one of the following: WBC count, CRP, or percentage of patients with fever for at least one type of orbital infection. Included studies had at least five patients per group.

Thirty-three articles were screened for the inflammatory marker section. Thirteen were excluded for the following reasons: no numbers reported for inflammatory markers (n = 6); group 1 and groups 2, 3, and 4 results combined (n = 6); fewer than five patients with orbital cellulitis included (n = 1). Twenty studies were included: 18 case series and 2 retrospective cohorts. Appendix Table 3 summarizes the data from studies included. Based on GRADE criteria, the body of evidence included in this section is of low quality.⁹

Distinguishing Between Preseptal and OCA

Eleven studies were included in this section (Table 1). WBC count was significantly higher in patients with groups 2, 3, and 4 than group 1 in two studies (Devrim et al,¹²P < .01; Santos et al,¹³P = .025). CRP was significantly higher in patients with groups 2, 3, and 4 than group 1 in four studies (Öcal Demir et al,¹⁴P = .02; Devrim et al,¹²P < .01; Ohana-Sarna-Cahan et al,¹⁸P < .001; Santos et al,¹³P < .001). Patients with groups 2, 3, and 4 had a significantly higher fever rate in three studies (Botting et al,²¹P < .001; Ohana-Sarna-Cahan et al,¹⁸P = .0001; Santos et al,¹³ P = .029).

Distinguishing Between Orbital Cellulitis and Abscess

Seven studies were included in this section (Appendix Table 3). One study showed significantly higher WBC count in group 3 than group 2 (P = .004), although results were reported as percentage of patients above a cutoff number calculated to distinguish between cellulitis and abscess (Appendix Table 3).²² CRP was not significantly different between group 2 and groups 3 and 4. One study found a significantly higher fever rate in patients with group 3 compared to patients with group 2 (P < .001).²²

Identifying Patients Requiring Surgery

Six studies were included in this section (Appendix Table 3). One study found a significantly higher WBC count in patients treated surgically (Tabarino et al,²⁴P < .05). Patients treated surgically had a significantly higher CRP in two studies (Cohen et al,²⁵P = .02; Friling et al,²⁶ P = .04). Fever was inconsistently reported in the studies, with some using mean presenting temperatures and some using rates of fever. One study found a significantly higher mean presenting temperature in patients treated surgically (P = .027), but the difference between the two groups was 0.7 °C.²³

Summary

Most studies found no significant difference in WBC count, CRP, or fever between preseptal and OCA, cellulitis and abscess, or patients receiving medical and surgical interventions.

Q2: What Is the Role of Imaging in Evaluation of OCA?

Twenty-five articles were selected for the imaging section review. All the included studies were retrospective descriptive studies. Quantitative data extraction and analysis of these studies could not be performed because of their heterogeneous methodologies and lack of objective data. Therefore, the information gleaned from these studies is summarized in narrative format. Per GRADE criteria, the body of evidence included in this section is of low quality.

Who Needs Imaging?

Proptosis, ophthalmoplegia, decreased vision, and pain with eye movements are widely agreed-upon indications for imaging evaluation.^21,27,28 Because of concern for radiation exposure in pediatric patients, some authors suggested that computed tomography (CT) should only be obtained if patients fail to respond to medical therapy or if surgery is being considered.^17,29,30 However, Rudloe et al³¹ found that half of the patients with group 3 or higher disease on CT did not have proptosis, ophthalmoplegia, or pain with extraocular movement. In addition, evaluation of young children with acute periorbital swelling can be difficult, so a lower threshold for imaging is likely warranted in these younger patients.

What Type of Imaging Should Be Obtained?

The American College of Radiology 2018 Appropriateness Criteria (ACR criteria) for orbital imaging state that orbital CT is usually indicated for patients with suspected Chandler groups 2, 3, and 4 infections.³² CT with contrast is useful for evaluating the extent of orbital infection and size of the abscess and for delineating the adjacent osseous anatomy, which is essential for cases in which surgical intervention is planned.^{20,21,26,27,30,31,33,34} Distinguishing abscess from cellulitis on CT sometimes can be challenging; therefore, serial clinical examinations and, occasionally, surgical exploration may be required.^35,36

Magnetic resonance imaging (MRI) is helpful for evaluating intracranial complications (eg, epidural abscess),^27,37 but it is limited for evaluating the osseous components of the paranasal sinuses. Although one study suggested that rapid MRI is comparable to contrast CT for differentiating group 1 infections from groups 2, 3, and 4 infections, it provided limited assessment of other complications.³⁸ With no definitive studies comparing CT with MRI for orbital infections, adherence to the ACR criteria is recommended.

Orbital ultrasound is limited by its small field of view and artifact produced by the surrounding bony interface, both of which can obscure posterior intraorbital pathologies.^29,39,40 Plain radiographs are not helpful for evaluating OCA due to limited soft-tissue contrast.⁴¹

When Should Repeat Imaging Be Obtained?

Children with group 3 OCA have been successfully managed medically in a carefully monitored setting.⁴² Repeat CT imaging is sometimes useful in these patients, particularly if the clinical examination is difficult.^42-44 However, improvement in CT findings may lag behind clinical improvement.³⁹

Summary

Per ACR criteria, orbital CT with contrast is recommended to evaluate patients with suspected Chandler groups 2, 3, and 4 OCA. MRI is reserved for evaluating intracranial complications.

Q3: What Is the Microbiology of OCA? What Is the MRSA Prevalence?

Knowledge of the microbiology of OCA is essential for the appropriate selection of empiric antibiotics. Because fewer children with groups 2 and 3 OCA undergo surgery, intraoperative cultures often are not available to guide antibiotic selection.⁴⁵ As a result, significant variation exists in antibiotic prescribing.⁶

Studies discussing the microbiology of OCA were included only if they were published in the past 2 decades (2000-2020) and were excluded if the study period was before 1990, as microbiology changes over time and new vaccines are introduced. To be included, the majority of cultures reported had to be intraoperative (orbital or sinus) specimens. Studies reporting only nasal, conjunctival, or other surface cultures were excluded. When studies included patients with group 1 OCA, only microbiology data for groups 2, 3, and 4 OCA were extracted. The pattern of resistance for S aureus was not always explicitly reported; however, when non-MRSA active antibiotics were used, methicillin-susceptible S aureus was assumed. 

A total of 63 studies were screened for the microbiology section; 32 were excluded for the following reasons: published before 2000 or study period before 1990 (n = 18), reported surface cultures or culture site not clearly stated (n = 4), microbiology mixed between preseptal and orbital (n = 6), wrong study type (n = 2), and study group overlaps with a different article included (n = 2). Of the 32 studies included, 3 were prospective observational, 4 were retrospective cohort, and 25 were case series. Based on GRADE criteria, the body of evidence included in this section is of low quality.⁴²

Appendix Table 4 summarizes the microbiologic data from the studies included. In the group of children that had a positive culture (orbital, sinus, or blood), the most commonly recovered organisms reported were S aureus (median, 22%; range, 0%-100%), Streptococcus anginosus group (median, 16%; range, 0%-100%), group A Streptococcus (median, 12%; range, 0%-80%), and Streptococcus pneumoniae (median, 8%; range, 0%-100%). Streptococcus as a group had a median prevalence of 57%, ranging from 0% to 100%. MRSA prevalence had a median of 3% (interquartile range [IQR], 0%-13%). Median prevalence of polymicrobial cultures was 20%, and median prevalence of anaerobic organisms was 14% (Table 2). Orbital and sinus cultures had the highest yield, with an average return of an organism of 72% (median, 75%; IQR, 64%-84%). Blood culture results were reported in 14 studies and usually obtained in a subgroup of the study population. When obtained, blood cultures rarely yielded an organism (median, 10%; IQR, 5%-15%); the rate of identified bacteremia in the total population had a median of 5% (IQR, 5%-7%) across studies.

Microbiology was compared between studies completed in the United States and in other countries (Table 2). Based on median prevalence across studies, both S anginosus group and MRSA were more prevalent in the United States than internationally (28% vs 0% and 11% vs 0%, respectively). No clear trend in MRSA prevalence was evident over the 2 decades; however, the studies included were heterogeneous and did not have the power to detect such a trend.

Two reports suggest a difference of MRSA prevalence by patient age. Hsu et al⁴⁶ found that three of eight MRSA infections were in infants age <1 year, which accounted for 50% (3/6) of infants included in the study. Miller et al⁴⁷ reported MRSA in 4 of 9 (44%) infants with OCA. Age <1 year may be associated with increased frequency of MRSA infection in OCA.

Summary

Blood cultures have low yield. The most common organisms recovered from OCA are Streptococcus species (most commonly S anginosus group, group A Streptococcus, and pneumococcus) and S aureus. Polymicrobial infections including anaerobes are common. MRSA prevalence is low globally but varies significantly among geographic areas.

Our systematic review of the literature for the medical management of OCA revealed predominantly descriptive studies and only a limited number of comparison-based studies, likely reflecting the rarity of advanced forms of OCA. Given the lack of high-quality evidence and the level of heterogeneity among studies, the conclusions that can be drawn are limited.

Distinguishing between disease severity and OCA requiring surgical intervention remains challenging. Although studies in our review suggest a trend toward markers of inflammation (fever, elevated WBC count and CRP) being more common in more severe presentations, the results were mixed, and studies were low quality and underpowered to detect meaningful differences. For example, most studies do not define what constitutes a fever in their cohort. Our review suggests that markers of inflammation cannot be used to distinguish between Chandler groups or to identify patients requiring surgery. Of note, the presence of fever and elevated inflammatory markers may have influenced the decision to obtain imaging or to proceed to surgery, thereby also potentially biasing these clinical indicators toward predictors for more severe disease. Decisions regarding surgery should therefore be based on the entire clinical picture, including response to appropriate antibiotics.

We found a lack of high-quality evidence regarding the role of imaging in OCA, and the studies reviewed were heterogeneous. Recommendations for imaging therefore remain at the level of expert opinion (ACR criteria). CT imaging is the first-line modality for imaging in suspected OCA given the limitations of alternative imaging modalities, but the sensitivity and specificity of CT imaging remain unknown for diagnosis of orbital abscesses.

Our review of the published microbiology confirmed that Staphylococcus and Streptococcus species are the most common pathogens identified in OCA. Prevalence across the different studies varied greatly. Owing to the significant heterogeneity in studies, calculation of pooled prevalence was not possible. By using the number of positive cultures as our denominator (or total surgeries if number of positive cultures was unavailable), we likely overestimated the prevalence of S aureus. S aureus is generally recognized as a pyogenic pathogen, more likely to be associated with abscess formation.⁴⁸ Therefore, culture results obtained predominantly from abscesses likely result in an overestimate of S aureus in OCA (groups 2, 3, and 4). Regardless, MRSA prevalence was generally low, both nationally and internationally. The MRSA results from the study by McKinley at el⁴⁹ (Texas) was a notable outlier in the United States, with MRSA prevalence as high as 44% compared with the median prevalence of 3% (IQR, 0-13), highlighting the importance of local resistance patterns when choosing empiric antibiotics.

Limitations to the microbiology review included significant heterogeneity in both the types of cultures included and the reporting of results. Although we excluded studies that reported only surface culture results or did not specify culture type, we did include studies that had surface culture results combined with intraoperative culture results, making it impossible to separate the two. Since most of the cultures included in combined results reported organisms based on intraoperative cultures, we felt they provided valuable information that should be included. In most studies, blood cultures were not obtained in all participants, so the yield of blood cultures is likely an overestimate, as blood cultures are more likely to be obtained in higher-acuity patients.

Although the available evidence regarding the medical management of OCA remains low quality, certain limited conclusions can be drawn, as presented in this review. Further high-quality studies are needed to better inform the medical management of OCA.

The authors thank Dr Kyle Pronko for his help with data extraction for the imaging section.

The ability to read and correctly interpret research is an essential skill, but most hospitalists—and physicians in general—do not receive formal training in biostatistics during their medical education.^1-3 In addition to straightforward statistical tests that compare a single exposure and outcome, researchers commonly use statistical models to identify and quantify complex relationships among many exposures (eg, demographics, clinical characteristics, interventions, or other variables) and an outcome. Understanding statistical models can be challenging. Still, it is important to recognize the advantages and limitations of statistical models, how to interpret their results, and the potential implications of findings on current clinical practice.

In the article “Rates and Characteristics of Medical Malpractice Claims Against Hospitalists” published in the July 2021 issue of the Journal of Hospital Medicine, Schaffer et al⁴ used the Comparative Benchmarking System database, which is maintained by a malpractice insurer, to characterize malpractice claims against hospitalists. The authors used multiple logistic regression models to understand the relationship among clinical factors and indemnity payments. In this Progress Note, we describe situations in which logistic regression is the proper statistical method to analyze a data set, explain results from logistic regression analyses, and equip readers with skills to critically appraise conclusions drawn from these models.

Statistical models often are used to describe the relationship among one or more exposure variables (ie, independent variables) and an outcome (ie, dependent variable). These models allow researchers to evaluate the effects of multiple exposure variables simultaneously, which in turn allows them to “isolate” the effect of each variable; in other words, models facilitate an understanding of the relationship between each exposure variable and the outcome, adjusted for (ie, independent of) the other exposure variables in the model.

Several statistical models can be used to quantify relationships within the data, but each type of model has certain assumptions that must be satisfied. Two important assumptions include characteristics of the outcome (eg, the type and distribution) and the nature of the relationships among the outcome and independent variables (eg, linear vs nonlinear). Simple linear regression, one of the most basic statistical models used in research,⁵ assumes that (a) the outcome is continuous (ie, any numeric value is possible) and normally distributed (ie, its histogram is a bell-shaped curve) and (b) the relationship between the independent variable and the outcome is linear (ie, follows a straight line). If an investigator wanted to understand how weight is related to height, a simple linear regression could be used to develop a mathematical equation that tells us how the outcome (weight) generally increases as the independent variable (height) increases.

Often, the outcome in a study is not a continuous variable but a simple success/failure variable (ie, dichotomous variable that can be one of two possible values). Schaffer et al⁴ examined the binary outcome of whether a malpractice claim case would end in an indemnity payment or no payment. Linear regression models are not equipped to handle dichotomous outcomes. Instead, we need to use a different statistical model: logistic regression. In logistic regression, the probability (p) of a defined outcome event is estimated by creating a regression model.

A probability (p) is a measure of how likely an event (eg, a malpractice claim ends in an indemnity payment or not) is to occur. It is always between 0 (ie, the event will definitely not occur) and 1 (ie, the event will definitely occur). A p of 0.5 means there is a 50/50 chance that the event will occur (ie, equivalent to a coin flip). Because p is a probability, we need to make sure it is always between 0 and 1. If we were to try to model p with a linear regression, the model would assume that p could extend beyond 0 and 1. What can we do?

Applying a transformation is a commonly used tool in statistics to make data work better within statistical models.⁶ In this case, we will transform the variable p. In logistic regression, we model the probability of experiencing the outcome through a transformation called a logit. The logit represents the natural logarithm (ln) of the ratio of the probability of experiencing the outcome (p) vs the probability of not experiencing the outcome (1 – p), with the ratio being the odds of the event occurring.

This transformation works well for dichotomous outcomes because the logit transformation approximates a straight line as long as p is not too large or too small (between 0.05 and 0.95).

If we are performing a logistic regression with only one independent variable (x) and want to understand the relationship between this variable (x) and the probability of an outcome event (p), then our model is the equation of a line. The equation for the base model of logistic regression with one independent variable (x) is

where β₀ is the y-intercept and β₁ is the slope of the line. Equation (2) is identical to the algebraic equation y = mx + b for a line, just rearranged slightly. In this algebraic equation, m is the slope (the same as β₁) and b is the y-intercept (the same as β₀). We will see that β₀ and β₁ are estimated (ie, assigned numeric values) from the data collected to help us understand how x and

are related and are the basis for estimating odds ratios.

We can build more complex models using multivariable logistic regression by adding more independent variables to the right side of equation (2). Essentially, this is what Schaffer et al⁴ did when, for example, they described clinical factors associated with indemnity payments (Schaffer et al, Table 3).

There are two notable techniques used frequently with multivariable logistic regression models. The first involves choosing which independent variables to include in the model. One way to select variables for multivariable models is defining them a priori, that is deciding which variables are clinically or conceptually associated with the outcome before looking at the data. With this approach, we can test specific hypotheses about the relationships between the independent variables and the outcome. Another common approach is to look at the data and identify the variables that vary significantly between the two outcome groups. Schaffer et al⁴ used an a priori approach to define variables in their multivariable model (ie, “variables for inclusion into the multivariable model were determined a priori”).

A second technique is the evaluation of collinearity, which helps us understand whether the independent variables are related to each other. It is important to consider collinearity between independent variables because the inclusion of two (or more) variables that are highly correlated can cause interference between the two and create misleading results from the model. There are techniques to assess collinear relationships before modeling or as part of the model-building process to determine which variables should be excluded. If there are two (or more) independent variables that are similar, one (or more) must be removed from the model.

Fitting the model is the process by which statistical software (eg, SAS, Stata, R, SPSS) estimates the relationships among independent variables in the model and the outcome within a specific dataset. In equation (2), this essentially means that the software will evaluate the data and provide us with the best estimates for β₀ (the y-intercept) and β₁ (the slope) that describe the relationship between the variable x and

Modeling can be iterative, and part of the process may include removing variables from the model that are not significantly associated with the outcome to create a simpler solution, a process known as model reduction. The results from models describe the independent association between a specific characteristic and the outcome, meaning that the relationship has been adjusted for all the other characteristics in the model.

The relationships among the independent variables and outcome are most often represented as an odds ratio (OR), which quantifies the strength of the association between two variables and is directly calculated from the β values in the model. As the name suggests, an OR is a ratio of odds. But what are odds? Simply, the odds of an outcome (such as mortality) is the probability of experiencing the event divided by the probability of not experiencing that event; in other words, it is the ratio:

The concept of odds is often unfamiliar, so it can be helpful to consider the definition in the context of games of chance. For example, in horse race betting, the outcome of interest is that a horse will lose a race. Imagine that the probability of a horse losing a race is 0.8 and the probability of winning is 0.2. The odds of losing are

These odds usually are listed as 4-to-1, meaning that out of 5 races (ie, 4 + 1) the horse is expected to lose 4 times and win once. When odds are listed this way, we can easily calculate the associated probability by recognizing that the total number of expected races is the sum of two numbers (probability of losing: 4 races out of 5, or 0.80 vs probability of winning: 1 race out of 5, or 0.20).

In medical research, the OR typically represents the odds for one group of patients (A) compared with the odds for another group of patients (B) experiencing an outcome. If the odds of the outcome are the same for group A and group B, then OR = 1.0, meaning that the probability of the outcome is the same between the two groups. If the patients in group A have greater odds of experiencing the outcome compared with group B patients (and a greater probability of the outcome), then the OR will be >1. If the opposite is true, then the OR will be <1.

Schaffer et al⁴ estimated that the OR of an indemnity payment in malpractice cases involving errors in clinical judgment as a contributing factor was 5.01 (95% CI, 3.37-7.45). This means that malpractice cases involving errors in clinical judgement had a 5.01 times greater odds of indemnity payment compared with those without these errors after adjusting for all other variables in the model (eg, age, severity). Note that the 95% CI does not include 1.0. This indicates that the OR is statistically >1, and we can conclude that there is a significant relationship between errors in clinical judgment and payment that is unlikely to be attributed to chance alone.

In logistic regression for categorical independent variables, all categories are compared with a reference group within that variable, with the reference group serving as the denominator of the OR. The authors⁴ did not incorporate continuous independent variables in their multivariable logistic regression model. However, if the authors examined length of hospitalization as a contributing factor in indemnity payments, for example, the OR would represent a 1-unit increase in this variable (eg, 1-day increase in length of stay).

Logistic regression describes the relationships in data and is an important statistical model across many types of research. This Progress Note emphasizes the importance of weighing the advantages and limitations of logistic regression, provides a common approach to data transformation, and guides the correct interpretation of logistic regression model results.

The ability to read and correctly interpret research is an essential skill, but most hospitalists—and physicians in general—do not receive formal training in biostatistics during their medical education.^1-3 In addition to straightforward statistical tests that compare a single exposure and outcome, researchers commonly use statistical models to identify and quantify complex relationships among many exposures (eg, demographics, clinical characteristics, interventions, or other variables) and an outcome. Understanding statistical models can be challenging. Still, it is important to recognize the advantages and limitations of statistical models, how to interpret their results, and the potential implications of findings on current clinical practice.

In the article “Rates and Characteristics of Medical Malpractice Claims Against Hospitalists” published in the July 2021 issue of the Journal of Hospital Medicine, Schaffer et al⁴ used the Comparative Benchmarking System database, which is maintained by a malpractice insurer, to characterize malpractice claims against hospitalists. The authors used multiple logistic regression models to understand the relationship among clinical factors and indemnity payments. In this Progress Note, we describe situations in which logistic regression is the proper statistical method to analyze a data set, explain results from logistic regression analyses, and equip readers with skills to critically appraise conclusions drawn from these models.

Statistical models often are used to describe the relationship among one or more exposure variables (ie, independent variables) and an outcome (ie, dependent variable). These models allow researchers to evaluate the effects of multiple exposure variables simultaneously, which in turn allows them to “isolate” the effect of each variable; in other words, models facilitate an understanding of the relationship between each exposure variable and the outcome, adjusted for (ie, independent of) the other exposure variables in the model.

Several statistical models can be used to quantify relationships within the data, but each type of model has certain assumptions that must be satisfied. Two important assumptions include characteristics of the outcome (eg, the type and distribution) and the nature of the relationships among the outcome and independent variables (eg, linear vs nonlinear). Simple linear regression, one of the most basic statistical models used in research,⁵ assumes that (a) the outcome is continuous (ie, any numeric value is possible) and normally distributed (ie, its histogram is a bell-shaped curve) and (b) the relationship between the independent variable and the outcome is linear (ie, follows a straight line). If an investigator wanted to understand how weight is related to height, a simple linear regression could be used to develop a mathematical equation that tells us how the outcome (weight) generally increases as the independent variable (height) increases.

Often, the outcome in a study is not a continuous variable but a simple success/failure variable (ie, dichotomous variable that can be one of two possible values). Schaffer et al⁴ examined the binary outcome of whether a malpractice claim case would end in an indemnity payment or no payment. Linear regression models are not equipped to handle dichotomous outcomes. Instead, we need to use a different statistical model: logistic regression. In logistic regression, the probability (p) of a defined outcome event is estimated by creating a regression model.

A probability (p) is a measure of how likely an event (eg, a malpractice claim ends in an indemnity payment or not) is to occur. It is always between 0 (ie, the event will definitely not occur) and 1 (ie, the event will definitely occur). A p of 0.5 means there is a 50/50 chance that the event will occur (ie, equivalent to a coin flip). Because p is a probability, we need to make sure it is always between 0 and 1. If we were to try to model p with a linear regression, the model would assume that p could extend beyond 0 and 1. What can we do?

Applying a transformation is a commonly used tool in statistics to make data work better within statistical models.⁶ In this case, we will transform the variable p. In logistic regression, we model the probability of experiencing the outcome through a transformation called a logit. The logit represents the natural logarithm (ln) of the ratio of the probability of experiencing the outcome (p) vs the probability of not experiencing the outcome (1 – p), with the ratio being the odds of the event occurring.

This transformation works well for dichotomous outcomes because the logit transformation approximates a straight line as long as p is not too large or too small (between 0.05 and 0.95).

If we are performing a logistic regression with only one independent variable (x) and want to understand the relationship between this variable (x) and the probability of an outcome event (p), then our model is the equation of a line. The equation for the base model of logistic regression with one independent variable (x) is

where β₀ is the y-intercept and β₁ is the slope of the line. Equation (2) is identical to the algebraic equation y = mx + b for a line, just rearranged slightly. In this algebraic equation, m is the slope (the same as β₁) and b is the y-intercept (the same as β₀). We will see that β₀ and β₁ are estimated (ie, assigned numeric values) from the data collected to help us understand how x and

are related and are the basis for estimating odds ratios.

We can build more complex models using multivariable logistic regression by adding more independent variables to the right side of equation (2). Essentially, this is what Schaffer et al⁴ did when, for example, they described clinical factors associated with indemnity payments (Schaffer et al, Table 3).

There are two notable techniques used frequently with multivariable logistic regression models. The first involves choosing which independent variables to include in the model. One way to select variables for multivariable models is defining them a priori, that is deciding which variables are clinically or conceptually associated with the outcome before looking at the data. With this approach, we can test specific hypotheses about the relationships between the independent variables and the outcome. Another common approach is to look at the data and identify the variables that vary significantly between the two outcome groups. Schaffer et al⁴ used an a priori approach to define variables in their multivariable model (ie, “variables for inclusion into the multivariable model were determined a priori”).

A second technique is the evaluation of collinearity, which helps us understand whether the independent variables are related to each other. It is important to consider collinearity between independent variables because the inclusion of two (or more) variables that are highly correlated can cause interference between the two and create misleading results from the model. There are techniques to assess collinear relationships before modeling or as part of the model-building process to determine which variables should be excluded. If there are two (or more) independent variables that are similar, one (or more) must be removed from the model.

Fitting the model is the process by which statistical software (eg, SAS, Stata, R, SPSS) estimates the relationships among independent variables in the model and the outcome within a specific dataset. In equation (2), this essentially means that the software will evaluate the data and provide us with the best estimates for β₀ (the y-intercept) and β₁ (the slope) that describe the relationship between the variable x and

Modeling can be iterative, and part of the process may include removing variables from the model that are not significantly associated with the outcome to create a simpler solution, a process known as model reduction. The results from models describe the independent association between a specific characteristic and the outcome, meaning that the relationship has been adjusted for all the other characteristics in the model.

The relationships among the independent variables and outcome are most often represented as an odds ratio (OR), which quantifies the strength of the association between two variables and is directly calculated from the β values in the model. As the name suggests, an OR is a ratio of odds. But what are odds? Simply, the odds of an outcome (such as mortality) is the probability of experiencing the event divided by the probability of not experiencing that event; in other words, it is the ratio:

The concept of odds is often unfamiliar, so it can be helpful to consider the definition in the context of games of chance. For example, in horse race betting, the outcome of interest is that a horse will lose a race. Imagine that the probability of a horse losing a race is 0.8 and the probability of winning is 0.2. The odds of losing are

These odds usually are listed as 4-to-1, meaning that out of 5 races (ie, 4 + 1) the horse is expected to lose 4 times and win once. When odds are listed this way, we can easily calculate the associated probability by recognizing that the total number of expected races is the sum of two numbers (probability of losing: 4 races out of 5, or 0.80 vs probability of winning: 1 race out of 5, or 0.20).

In medical research, the OR typically represents the odds for one group of patients (A) compared with the odds for another group of patients (B) experiencing an outcome. If the odds of the outcome are the same for group A and group B, then OR = 1.0, meaning that the probability of the outcome is the same between the two groups. If the patients in group A have greater odds of experiencing the outcome compared with group B patients (and a greater probability of the outcome), then the OR will be >1. If the opposite is true, then the OR will be <1.

Schaffer et al⁴ estimated that the OR of an indemnity payment in malpractice cases involving errors in clinical judgment as a contributing factor was 5.01 (95% CI, 3.37-7.45). This means that malpractice cases involving errors in clinical judgement had a 5.01 times greater odds of indemnity payment compared with those without these errors after adjusting for all other variables in the model (eg, age, severity). Note that the 95% CI does not include 1.0. This indicates that the OR is statistically >1, and we can conclude that there is a significant relationship between errors in clinical judgment and payment that is unlikely to be attributed to chance alone.

In logistic regression for categorical independent variables, all categories are compared with a reference group within that variable, with the reference group serving as the denominator of the OR. The authors⁴ did not incorporate continuous independent variables in their multivariable logistic regression model. However, if the authors examined length of hospitalization as a contributing factor in indemnity payments, for example, the OR would represent a 1-unit increase in this variable (eg, 1-day increase in length of stay).

Logistic regression describes the relationships in data and is an important statistical model across many types of research. This Progress Note emphasizes the importance of weighing the advantages and limitations of logistic regression, provides a common approach to data transformation, and guides the correct interpretation of logistic regression model results.

The ability to read and correctly interpret research is an essential skill, but most hospitalists—and physicians in general—do not receive formal training in biostatistics during their medical education.^1-3 In addition to straightforward statistical tests that compare a single exposure and outcome, researchers commonly use statistical models to identify and quantify complex relationships among many exposures (eg, demographics, clinical characteristics, interventions, or other variables) and an outcome. Understanding statistical models can be challenging. Still, it is important to recognize the advantages and limitations of statistical models, how to interpret their results, and the potential implications of findings on current clinical practice.

In the article “Rates and Characteristics of Medical Malpractice Claims Against Hospitalists” published in the July 2021 issue of the Journal of Hospital Medicine, Schaffer et al⁴ used the Comparative Benchmarking System database, which is maintained by a malpractice insurer, to characterize malpractice claims against hospitalists. The authors used multiple logistic regression models to understand the relationship among clinical factors and indemnity payments. In this Progress Note, we describe situations in which logistic regression is the proper statistical method to analyze a data set, explain results from logistic regression analyses, and equip readers with skills to critically appraise conclusions drawn from these models.

Statistical models often are used to describe the relationship among one or more exposure variables (ie, independent variables) and an outcome (ie, dependent variable). These models allow researchers to evaluate the effects of multiple exposure variables simultaneously, which in turn allows them to “isolate” the effect of each variable; in other words, models facilitate an understanding of the relationship between each exposure variable and the outcome, adjusted for (ie, independent of) the other exposure variables in the model.

Several statistical models can be used to quantify relationships within the data, but each type of model has certain assumptions that must be satisfied. Two important assumptions include characteristics of the outcome (eg, the type and distribution) and the nature of the relationships among the outcome and independent variables (eg, linear vs nonlinear). Simple linear regression, one of the most basic statistical models used in research,⁵ assumes that (a) the outcome is continuous (ie, any numeric value is possible) and normally distributed (ie, its histogram is a bell-shaped curve) and (b) the relationship between the independent variable and the outcome is linear (ie, follows a straight line). If an investigator wanted to understand how weight is related to height, a simple linear regression could be used to develop a mathematical equation that tells us how the outcome (weight) generally increases as the independent variable (height) increases.

Often, the outcome in a study is not a continuous variable but a simple success/failure variable (ie, dichotomous variable that can be one of two possible values). Schaffer et al⁴ examined the binary outcome of whether a malpractice claim case would end in an indemnity payment or no payment. Linear regression models are not equipped to handle dichotomous outcomes. Instead, we need to use a different statistical model: logistic regression. In logistic regression, the probability (p) of a defined outcome event is estimated by creating a regression model.

A probability (p) is a measure of how likely an event (eg, a malpractice claim ends in an indemnity payment or not) is to occur. It is always between 0 (ie, the event will definitely not occur) and 1 (ie, the event will definitely occur). A p of 0.5 means there is a 50/50 chance that the event will occur (ie, equivalent to a coin flip). Because p is a probability, we need to make sure it is always between 0 and 1. If we were to try to model p with a linear regression, the model would assume that p could extend beyond 0 and 1. What can we do?

Applying a transformation is a commonly used tool in statistics to make data work better within statistical models.⁶ In this case, we will transform the variable p. In logistic regression, we model the probability of experiencing the outcome through a transformation called a logit. The logit represents the natural logarithm (ln) of the ratio of the probability of experiencing the outcome (p) vs the probability of not experiencing the outcome (1 – p), with the ratio being the odds of the event occurring.

This transformation works well for dichotomous outcomes because the logit transformation approximates a straight line as long as p is not too large or too small (between 0.05 and 0.95).

If we are performing a logistic regression with only one independent variable (x) and want to understand the relationship between this variable (x) and the probability of an outcome event (p), then our model is the equation of a line. The equation for the base model of logistic regression with one independent variable (x) is

where β₀ is the y-intercept and β₁ is the slope of the line. Equation (2) is identical to the algebraic equation y = mx + b for a line, just rearranged slightly. In this algebraic equation, m is the slope (the same as β₁) and b is the y-intercept (the same as β₀). We will see that β₀ and β₁ are estimated (ie, assigned numeric values) from the data collected to help us understand how x and

are related and are the basis for estimating odds ratios.

We can build more complex models using multivariable logistic regression by adding more independent variables to the right side of equation (2). Essentially, this is what Schaffer et al⁴ did when, for example, they described clinical factors associated with indemnity payments (Schaffer et al, Table 3).

There are two notable techniques used frequently with multivariable logistic regression models. The first involves choosing which independent variables to include in the model. One way to select variables for multivariable models is defining them a priori, that is deciding which variables are clinically or conceptually associated with the outcome before looking at the data. With this approach, we can test specific hypotheses about the relationships between the independent variables and the outcome. Another common approach is to look at the data and identify the variables that vary significantly between the two outcome groups. Schaffer et al⁴ used an a priori approach to define variables in their multivariable model (ie, “variables for inclusion into the multivariable model were determined a priori”).

A second technique is the evaluation of collinearity, which helps us understand whether the independent variables are related to each other. It is important to consider collinearity between independent variables because the inclusion of two (or more) variables that are highly correlated can cause interference between the two and create misleading results from the model. There are techniques to assess collinear relationships before modeling or as part of the model-building process to determine which variables should be excluded. If there are two (or more) independent variables that are similar, one (or more) must be removed from the model.

Fitting the model is the process by which statistical software (eg, SAS, Stata, R, SPSS) estimates the relationships among independent variables in the model and the outcome within a specific dataset. In equation (2), this essentially means that the software will evaluate the data and provide us with the best estimates for β₀ (the y-intercept) and β₁ (the slope) that describe the relationship between the variable x and

Modeling can be iterative, and part of the process may include removing variables from the model that are not significantly associated with the outcome to create a simpler solution, a process known as model reduction. The results from models describe the independent association between a specific characteristic and the outcome, meaning that the relationship has been adjusted for all the other characteristics in the model.

The relationships among the independent variables and outcome are most often represented as an odds ratio (OR), which quantifies the strength of the association between two variables and is directly calculated from the β values in the model. As the name suggests, an OR is a ratio of odds. But what are odds? Simply, the odds of an outcome (such as mortality) is the probability of experiencing the event divided by the probability of not experiencing that event; in other words, it is the ratio:

The concept of odds is often unfamiliar, so it can be helpful to consider the definition in the context of games of chance. For example, in horse race betting, the outcome of interest is that a horse will lose a race. Imagine that the probability of a horse losing a race is 0.8 and the probability of winning is 0.2. The odds of losing are

These odds usually are listed as 4-to-1, meaning that out of 5 races (ie, 4 + 1) the horse is expected to lose 4 times and win once. When odds are listed this way, we can easily calculate the associated probability by recognizing that the total number of expected races is the sum of two numbers (probability of losing: 4 races out of 5, or 0.80 vs probability of winning: 1 race out of 5, or 0.20).

In medical research, the OR typically represents the odds for one group of patients (A) compared with the odds for another group of patients (B) experiencing an outcome. If the odds of the outcome are the same for group A and group B, then OR = 1.0, meaning that the probability of the outcome is the same between the two groups. If the patients in group A have greater odds of experiencing the outcome compared with group B patients (and a greater probability of the outcome), then the OR will be >1. If the opposite is true, then the OR will be <1.

Schaffer et al⁴ estimated that the OR of an indemnity payment in malpractice cases involving errors in clinical judgment as a contributing factor was 5.01 (95% CI, 3.37-7.45). This means that malpractice cases involving errors in clinical judgement had a 5.01 times greater odds of indemnity payment compared with those without these errors after adjusting for all other variables in the model (eg, age, severity). Note that the 95% CI does not include 1.0. This indicates that the OR is statistically >1, and we can conclude that there is a significant relationship between errors in clinical judgment and payment that is unlikely to be attributed to chance alone.

In logistic regression for categorical independent variables, all categories are compared with a reference group within that variable, with the reference group serving as the denominator of the OR. The authors⁴ did not incorporate continuous independent variables in their multivariable logistic regression model. However, if the authors examined length of hospitalization as a contributing factor in indemnity payments, for example, the OR would represent a 1-unit increase in this variable (eg, 1-day increase in length of stay).

Logistic regression describes the relationships in data and is an important statistical model across many types of research. This Progress Note emphasizes the importance of weighing the advantages and limitations of logistic regression, provides a common approach to data transformation, and guides the correct interpretation of logistic regression model results.

The burden of chronic liver disease (CLD) in the United States is growing, and it is currently the fourth leading cause of death in adults aged 45 to 64 years.¹ From 2012 to 2016, there were 538,720 hospitalizations in the United States for patients with cirrhosis, with almost a quarter having at least one cirrhosis-related complication. Inpatient hospitalizations for cirrhosis contribute to healthcare resource utilization, with a mean cost per CLD-related hospitalization of $16,271, and the presence of cirrhosis results in higher mortality and cost burden.¹

In hospitalized patients with decompensated cirrhosis with ascites, intravenous human albumin (HA) infusion has been utilized for decades for a variety of indications. Current guidance by the American Association for the Study of Liver Diseases (AASLD) and the European Association for the Study of the Liver (EASL) recommends the use of albumin for the prevention of paracentesis-induced circulatory dysfunction (PICD) for the prevention of kidney injury in spontaneous bacterial peritonitis (SBP) and for the diagnosis and treatment of hepatorenal syndrome (HRS).^2,3 There have been several major trials in recent years studying the use of HA for other indications in patients with cirrhosis, and the Society of Critical Care Medicine (SCCM) updated their guidelines in 2020 to recommend HA administration in resuscitation of critically ill patients with liver failure with hypoalbuminemia.⁴This Clinical Progress Note addresses the use of albumin in hospitalized patients with cirrhosis, focusing on current indications and discussing potential uses published after the 2018 EASL guidelines. We conducted a literature search via the PubMed database. The authors began by using the Medical Subject Heading (MeSH) terms albumins/administration AND dosage; organization AND administration; adverse effects; and therapeutic use combined with liver cirrhosis as a MeSH major topic, which yielded 107 English-language articles published in the previous 10 years, and MeSH major topics of albumins and liver cirrhosis, which yielded 461 English-language articles, with 178 published in the previous 10 years. The search results were reviewed for applicability to albumin strategies for patients with cirrhosis.

There are three widely accepted and evidence-based indications for HA infusion in patients with cirrhosis, considered standard of care (Table).

Prevention of PICD

Therapeutic large-volume paracentesis (LVP) leads to a rise in plasma renin activity (PRA) centrally through several mechanisms and is not impacted by the rate of ascites removal.⁵ LVP relieves abdominal pressure, increasing venous return to the heart and cardiac output, and the corresponding drop in systemic vascular resistance with splanchnic vasodilation decreases effective circulating volume and activates the renin-angiotensin system. This PRA activation and circulatory dysfunction are associated with reaccumulating ascites, renal impairment, hypervolemic hyponatremia, and increased mortality.⁶ A large meta-analysis of 17 trials with 1225 patients found that HA infusion improves outcomes and reduces mortality for patients undergoing LVP (odds ratio [OR], 0.64; 95% CI, 0.41-0.98), reduces the risk of PICD more than other volume expanders tested, and lowers the incidence of hyponatremia.⁶ More recently, in 2017, Kütting et al⁷ analyzed 21 trials with 1277 patients and did not observe a significant mortality benefit for HA after LVP (OR, 0.78; 95% CI, 0.55-1.11). However, negative outcomes such as rise in PRA (OR, 0.53; 95% CI, 0.29-0.97) and hyponatremia (OR, 0.62; 95% CI, 0.42-0.94) were prevented. Guidelines recommend HA after LVP ≥5 L to prevent PICD, with a replacement volume of 6 to 8 g of albumin per liter of ascitic fluid removed.^2,3 Some patients may be at higher risk for PICD with less ascites removed, and the AASLD supports the use of HA to prevent PICD after smaller-volume paracentesis in patients who are already hypotensive (systolic blood pressure <90 mm Hg) or hyponatremic (<130 mmol/L), or have acute kidney injury.³

Spontaneous Bacterial Peritonitis

Spontaneous bacterial peritonitis is diagnosed by paracentesis, defined as ascitic neutrophil count ≥250 cells/µL with or without bacterascites (positive bacteriological culture). Bacterascites may be a precursor to the development of SBP, with the fluid neutrophil count of ≥250 determining the need for SBP treatment.² SBP can lead to circulatory dysfunction, hepatic encephalopathy, and HRS. Treating SBP with HA in addition to antibiotics reduces the risk of kidney injury compared with antibiotics alone (OR for kidney injury with antibiotics alone, 4.6; 95% CI, 1.3-16.1) and also reduces the risk of death (OR for mortality with antibiotics alone, 4.5; 95% CI, 1.0-20.9).⁸ The AASLD recommends albumin in addition to antibiotics in SBP to prevent HRS and acute kidney injury, and high-risk patients who already have kidney dysfunction (creatinine >1 mg/dL) or jaundice (total bilirubin >5 mg/dL) are more likely to benefit from albumin. The treatment schedule is 25% HA at 1.5 g/kg on day 1 and 1 g/kg on day 3.³ The EASL recommends administering HA to all patients with cirrhosis with SBP regardless of renal or liver indices. They acknowledge, however, that the incidence of SBP-associated acute kidney injury will be low in patients without severe hepatic disease or baseline renal impairment.²

Hepatorenal Syndrome

Albumin combined with vasoconstrictors is effective in treating HRS with a response rate of 20% to 80% (average, 50%).³ Vasoactive medications can include combination midodrine and octreotide or norepinephrine (or terlipressin outside of the United States). In patients with suspected HRS, the recommended dosing of 25% HA is 1 g/kg (to a maximum of 100 g of albumin) on day 1 and then 40 to 50 g daily for at least 3 days after the diagnosis is confirmed.³ The optimal duration of therapy beyond 3 days of combined therapy with midodrine, albumin, and octreotide is not established. Terlipressin treatment is recommended for a maximum of 14 days in cases of partial response or nonresponse in renal recovery.²

Hypoalbuminemia

Albumin administration to raise serum albumin levels in hospitalized patients has been a common practice. However, new evidence suggests that treating hypoalbuminemia with infusion of HA in hospitalized patients with decompensated cirrhosis does not protect patients from risk and causes harm. The Albumin To prevenT Infection in chronic liveR (ATTIRE) trial, published in 2021, randomly assigned 777 patients across 35 centers in the United Kingdom to receive daily 20% HA to target a serum albumin level of 3.0 g/dL vs standard care, including HA for established indications.^2,3 The primary end point was a composite of infection, kidney dysfunction, and death within 3 to 15 days of initiating treatment. There were no differences in the primary end point; secondary end points of death at 28 days, 3 months, or 6 months; or duration of hospitalization. The treatment group received 10 times more albumin than the control group and reported more adverse events, including pulmonary edema.⁹

Long-Term Treatment in Patients With Ascites

The human Albumin for the treatmeNt of aScites in patients With hEpatic ciRrhosis (ANSWER) trial, published in 2018, found improved 18-month survival in patients with cirrhosis and ascites treated with diuretics who received long-term albumin. This was an open-label trial of 431 patients at 33 sites in Italy, and the treatment arm received weekly infusions of 40 g of 20% HA. They observed a 38% reduction in mortality hazard ratio and half the number of hospital days annually.¹⁰ Based on these data and those from a 2006 Italian study with similar design and results, the Italian Association for the Study of the Liver (AISF) strongly recommends long-term albumin treatment in patients with cirrhosis with ascites.¹¹ The lead author on the ANSWER trial also authored the AISF statement, although this recommendation has not been adopted by the EASL or the AASLD.

Conversely, the Midodrine and Albumin for CirrHoTic patients (MACHT) trial, also published in 2018, randomly assigned 173 patients with ascites awaiting liver transplant to receive 40 g of HA every 15 days and midodrine in addition to standard care vs placebo. MACHT found no difference in mortality or complications at 1 year.¹²

Long-term albumin therapy as a preventive measure may be a disease modifier, taking advantage of the pleiotropic effects of albumin, though the differing conclusions from ANSWER and MACHT necessitate additional trials. The ongoing PRECIOSA study in Spain is assessing dosage and schedule for this therapy.¹³

Augmenting Diuresis

Loop diuretics are highly protein-bound, and, with hypoalbuminemia, there is less effective drug delivered to the site of action. One clinical approach is to augment diuretics with concomitant HA infusion. This approach is not supported by strong evidence or guidelines.

Hyponatremia

In a retrospective cohort study of 2435 hospitalized patients with cirrhosis, 1126 of whom had hyponatremia, those patients with sodium <130 mmol/L who received HA were more likely to have resolution of hyponatremia to >135 mmol/L. This was associated with improved 30-day survival.¹⁴ From this observational data, the AASLD supports the use of albumin combined with extreme fluid restriction (<1000 mL/d) for patients with severe hyponatremia (<120 mmol/L).³

Non-SBP Infections

A 2019 meta-analysis found no evidence of a benefit of HA for bacterial infections other than SBP. However, only three trials encompassing 407 patients met the inclusion criteria.¹⁵

SCCM Guideline Update: Hypoalbuminemia and Hypotension

The 2020 SCCM Guidelines for the Management of Adult Acute and Acute-on-Chronic Liver Failure in the ICU “suggest using albumin for resuscitation of patients [with liver failure] over other fluids, especially when serum albumin is low (<3 g/dL).” Acute-on-chronic liver failure is decompensation of cirrhosis combined with organ dysfunction (eg, coagulopathy, encephalopathy, kidney injury), a scenario that is frequently encountered by hospitalists outside of intensive care settings. In hypotensive patients with cirrhosis, the SCCM recommends administering albumin to a target mean arterial pressure of 65 mm Hg or otherwise adequate perfusion. This new recommendation is conditional, based on expert consensus, and derives from low-quality evidence, with acknowledgement that “costs may be prohibitive.”⁴

While the ATTIRE study demonstrated no benefit in treating hypoalbuminemia with infusion of HA in hospitalized patients with decompensated cirrhosis, the 2020 SCCM guidelines, released prior to the publication of the ATTIRE study, focused on more acutely ill patients. In the ATTIRE study, only 2% to 3% of the study population was in an intensive care unit.^4,9 The use of albumin infusion in the critically ill, hypoalbuminemic, hypotensive patient is not well studied, and the SCCM acknowledges the lack of supportive evidence for this practice in their guideline statement.

The three cardinal clinical indications for human albumin in patients with cirrhosis—prevention of PICD after LVP, in SBP, and for HRS—remain supported by the literature and guidelines, with the most recent guidance adding more nuance in patient selection based on individual risk (Table). With the publication of several large-scale studies in the past few years and a 2021 update to the AASLD guidance statement, clinicians have more evidence to guide their use of HA in patients with cirrhosis. In particular, the practice of treating isolated hypoalbuminemia with HA is no longer supported by the best evidence and is potentially harmful. A professional society recommendation to preferentially use albumin as a resuscitation fluid in hypoalbuminemia was made without the benefit of the results of the 2021 ATTIRE trial. On the horizon, additional results from ongoing and upcoming studies exploring concepts of effective albumin concentration and the pleiotropic properties of HA will impact the use of this therapy in hospitalized patients with cirrhosis.

The burden of chronic liver disease (CLD) in the United States is growing, and it is currently the fourth leading cause of death in adults aged 45 to 64 years.¹ From 2012 to 2016, there were 538,720 hospitalizations in the United States for patients with cirrhosis, with almost a quarter having at least one cirrhosis-related complication. Inpatient hospitalizations for cirrhosis contribute to healthcare resource utilization, with a mean cost per CLD-related hospitalization of $16,271, and the presence of cirrhosis results in higher mortality and cost burden.¹

In hospitalized patients with decompensated cirrhosis with ascites, intravenous human albumin (HA) infusion has been utilized for decades for a variety of indications. Current guidance by the American Association for the Study of Liver Diseases (AASLD) and the European Association for the Study of the Liver (EASL) recommends the use of albumin for the prevention of paracentesis-induced circulatory dysfunction (PICD) for the prevention of kidney injury in spontaneous bacterial peritonitis (SBP) and for the diagnosis and treatment of hepatorenal syndrome (HRS).^2,3 There have been several major trials in recent years studying the use of HA for other indications in patients with cirrhosis, and the Society of Critical Care Medicine (SCCM) updated their guidelines in 2020 to recommend HA administration in resuscitation of critically ill patients with liver failure with hypoalbuminemia.⁴This Clinical Progress Note addresses the use of albumin in hospitalized patients with cirrhosis, focusing on current indications and discussing potential uses published after the 2018 EASL guidelines. We conducted a literature search via the PubMed database. The authors began by using the Medical Subject Heading (MeSH) terms albumins/administration AND dosage; organization AND administration; adverse effects; and therapeutic use combined with liver cirrhosis as a MeSH major topic, which yielded 107 English-language articles published in the previous 10 years, and MeSH major topics of albumins and liver cirrhosis, which yielded 461 English-language articles, with 178 published in the previous 10 years. The search results were reviewed for applicability to albumin strategies for patients with cirrhosis.

There are three widely accepted and evidence-based indications for HA infusion in patients with cirrhosis, considered standard of care (Table).

Prevention of PICD

Therapeutic large-volume paracentesis (LVP) leads to a rise in plasma renin activity (PRA) centrally through several mechanisms and is not impacted by the rate of ascites removal.⁵ LVP relieves abdominal pressure, increasing venous return to the heart and cardiac output, and the corresponding drop in systemic vascular resistance with splanchnic vasodilation decreases effective circulating volume and activates the renin-angiotensin system. This PRA activation and circulatory dysfunction are associated with reaccumulating ascites, renal impairment, hypervolemic hyponatremia, and increased mortality.⁶ A large meta-analysis of 17 trials with 1225 patients found that HA infusion improves outcomes and reduces mortality for patients undergoing LVP (odds ratio [OR], 0.64; 95% CI, 0.41-0.98), reduces the risk of PICD more than other volume expanders tested, and lowers the incidence of hyponatremia.⁶ More recently, in 2017, Kütting et al⁷ analyzed 21 trials with 1277 patients and did not observe a significant mortality benefit for HA after LVP (OR, 0.78; 95% CI, 0.55-1.11). However, negative outcomes such as rise in PRA (OR, 0.53; 95% CI, 0.29-0.97) and hyponatremia (OR, 0.62; 95% CI, 0.42-0.94) were prevented. Guidelines recommend HA after LVP ≥5 L to prevent PICD, with a replacement volume of 6 to 8 g of albumin per liter of ascitic fluid removed.^2,3 Some patients may be at higher risk for PICD with less ascites removed, and the AASLD supports the use of HA to prevent PICD after smaller-volume paracentesis in patients who are already hypotensive (systolic blood pressure <90 mm Hg) or hyponatremic (<130 mmol/L), or have acute kidney injury.³

Spontaneous Bacterial Peritonitis

Spontaneous bacterial peritonitis is diagnosed by paracentesis, defined as ascitic neutrophil count ≥250 cells/µL with or without bacterascites (positive bacteriological culture). Bacterascites may be a precursor to the development of SBP, with the fluid neutrophil count of ≥250 determining the need for SBP treatment.² SBP can lead to circulatory dysfunction, hepatic encephalopathy, and HRS. Treating SBP with HA in addition to antibiotics reduces the risk of kidney injury compared with antibiotics alone (OR for kidney injury with antibiotics alone, 4.6; 95% CI, 1.3-16.1) and also reduces the risk of death (OR for mortality with antibiotics alone, 4.5; 95% CI, 1.0-20.9).⁸ The AASLD recommends albumin in addition to antibiotics in SBP to prevent HRS and acute kidney injury, and high-risk patients who already have kidney dysfunction (creatinine >1 mg/dL) or jaundice (total bilirubin >5 mg/dL) are more likely to benefit from albumin. The treatment schedule is 25% HA at 1.5 g/kg on day 1 and 1 g/kg on day 3.³ The EASL recommends administering HA to all patients with cirrhosis with SBP regardless of renal or liver indices. They acknowledge, however, that the incidence of SBP-associated acute kidney injury will be low in patients without severe hepatic disease or baseline renal impairment.²

Hepatorenal Syndrome

Albumin combined with vasoconstrictors is effective in treating HRS with a response rate of 20% to 80% (average, 50%).³ Vasoactive medications can include combination midodrine and octreotide or norepinephrine (or terlipressin outside of the United States). In patients with suspected HRS, the recommended dosing of 25% HA is 1 g/kg (to a maximum of 100 g of albumin) on day 1 and then 40 to 50 g daily for at least 3 days after the diagnosis is confirmed.³ The optimal duration of therapy beyond 3 days of combined therapy with midodrine, albumin, and octreotide is not established. Terlipressin treatment is recommended for a maximum of 14 days in cases of partial response or nonresponse in renal recovery.²

Hypoalbuminemia

Albumin administration to raise serum albumin levels in hospitalized patients has been a common practice. However, new evidence suggests that treating hypoalbuminemia with infusion of HA in hospitalized patients with decompensated cirrhosis does not protect patients from risk and causes harm. The Albumin To prevenT Infection in chronic liveR (ATTIRE) trial, published in 2021, randomly assigned 777 patients across 35 centers in the United Kingdom to receive daily 20% HA to target a serum albumin level of 3.0 g/dL vs standard care, including HA for established indications.^2,3 The primary end point was a composite of infection, kidney dysfunction, and death within 3 to 15 days of initiating treatment. There were no differences in the primary end point; secondary end points of death at 28 days, 3 months, or 6 months; or duration of hospitalization. The treatment group received 10 times more albumin than the control group and reported more adverse events, including pulmonary edema.⁹

Long-Term Treatment in Patients With Ascites

The human Albumin for the treatmeNt of aScites in patients With hEpatic ciRrhosis (ANSWER) trial, published in 2018, found improved 18-month survival in patients with cirrhosis and ascites treated with diuretics who received long-term albumin. This was an open-label trial of 431 patients at 33 sites in Italy, and the treatment arm received weekly infusions of 40 g of 20% HA. They observed a 38% reduction in mortality hazard ratio and half the number of hospital days annually.¹⁰ Based on these data and those from a 2006 Italian study with similar design and results, the Italian Association for the Study of the Liver (AISF) strongly recommends long-term albumin treatment in patients with cirrhosis with ascites.¹¹ The lead author on the ANSWER trial also authored the AISF statement, although this recommendation has not been adopted by the EASL or the AASLD.

Conversely, the Midodrine and Albumin for CirrHoTic patients (MACHT) trial, also published in 2018, randomly assigned 173 patients with ascites awaiting liver transplant to receive 40 g of HA every 15 days and midodrine in addition to standard care vs placebo. MACHT found no difference in mortality or complications at 1 year.¹²

Long-term albumin therapy as a preventive measure may be a disease modifier, taking advantage of the pleiotropic effects of albumin, though the differing conclusions from ANSWER and MACHT necessitate additional trials. The ongoing PRECIOSA study in Spain is assessing dosage and schedule for this therapy.¹³

Augmenting Diuresis

Loop diuretics are highly protein-bound, and, with hypoalbuminemia, there is less effective drug delivered to the site of action. One clinical approach is to augment diuretics with concomitant HA infusion. This approach is not supported by strong evidence or guidelines.

Hyponatremia

In a retrospective cohort study of 2435 hospitalized patients with cirrhosis, 1126 of whom had hyponatremia, those patients with sodium <130 mmol/L who received HA were more likely to have resolution of hyponatremia to >135 mmol/L. This was associated with improved 30-day survival.¹⁴ From this observational data, the AASLD supports the use of albumin combined with extreme fluid restriction (<1000 mL/d) for patients with severe hyponatremia (<120 mmol/L).³

Non-SBP Infections

A 2019 meta-analysis found no evidence of a benefit of HA for bacterial infections other than SBP. However, only three trials encompassing 407 patients met the inclusion criteria.¹⁵

SCCM Guideline Update: Hypoalbuminemia and Hypotension

The 2020 SCCM Guidelines for the Management of Adult Acute and Acute-on-Chronic Liver Failure in the ICU “suggest using albumin for resuscitation of patients [with liver failure] over other fluids, especially when serum albumin is low (<3 g/dL).” Acute-on-chronic liver failure is decompensation of cirrhosis combined with organ dysfunction (eg, coagulopathy, encephalopathy, kidney injury), a scenario that is frequently encountered by hospitalists outside of intensive care settings. In hypotensive patients with cirrhosis, the SCCM recommends administering albumin to a target mean arterial pressure of 65 mm Hg or otherwise adequate perfusion. This new recommendation is conditional, based on expert consensus, and derives from low-quality evidence, with acknowledgement that “costs may be prohibitive.”⁴

While the ATTIRE study demonstrated no benefit in treating hypoalbuminemia with infusion of HA in hospitalized patients with decompensated cirrhosis, the 2020 SCCM guidelines, released prior to the publication of the ATTIRE study, focused on more acutely ill patients. In the ATTIRE study, only 2% to 3% of the study population was in an intensive care unit.^4,9 The use of albumin infusion in the critically ill, hypoalbuminemic, hypotensive patient is not well studied, and the SCCM acknowledges the lack of supportive evidence for this practice in their guideline statement.

The three cardinal clinical indications for human albumin in patients with cirrhosis—prevention of PICD after LVP, in SBP, and for HRS—remain supported by the literature and guidelines, with the most recent guidance adding more nuance in patient selection based on individual risk (Table). With the publication of several large-scale studies in the past few years and a 2021 update to the AASLD guidance statement, clinicians have more evidence to guide their use of HA in patients with cirrhosis. In particular, the practice of treating isolated hypoalbuminemia with HA is no longer supported by the best evidence and is potentially harmful. A professional society recommendation to preferentially use albumin as a resuscitation fluid in hypoalbuminemia was made without the benefit of the results of the 2021 ATTIRE trial. On the horizon, additional results from ongoing and upcoming studies exploring concepts of effective albumin concentration and the pleiotropic properties of HA will impact the use of this therapy in hospitalized patients with cirrhosis.

The burden of chronic liver disease (CLD) in the United States is growing, and it is currently the fourth leading cause of death in adults aged 45 to 64 years.¹ From 2012 to 2016, there were 538,720 hospitalizations in the United States for patients with cirrhosis, with almost a quarter having at least one cirrhosis-related complication. Inpatient hospitalizations for cirrhosis contribute to healthcare resource utilization, with a mean cost per CLD-related hospitalization of $16,271, and the presence of cirrhosis results in higher mortality and cost burden.¹

In hospitalized patients with decompensated cirrhosis with ascites, intravenous human albumin (HA) infusion has been utilized for decades for a variety of indications. Current guidance by the American Association for the Study of Liver Diseases (AASLD) and the European Association for the Study of the Liver (EASL) recommends the use of albumin for the prevention of paracentesis-induced circulatory dysfunction (PICD) for the prevention of kidney injury in spontaneous bacterial peritonitis (SBP) and for the diagnosis and treatment of hepatorenal syndrome (HRS).^2,3 There have been several major trials in recent years studying the use of HA for other indications in patients with cirrhosis, and the Society of Critical Care Medicine (SCCM) updated their guidelines in 2020 to recommend HA administration in resuscitation of critically ill patients with liver failure with hypoalbuminemia.⁴This Clinical Progress Note addresses the use of albumin in hospitalized patients with cirrhosis, focusing on current indications and discussing potential uses published after the 2018 EASL guidelines. We conducted a literature search via the PubMed database. The authors began by using the Medical Subject Heading (MeSH) terms albumins/administration AND dosage; organization AND administration; adverse effects; and therapeutic use combined with liver cirrhosis as a MeSH major topic, which yielded 107 English-language articles published in the previous 10 years, and MeSH major topics of albumins and liver cirrhosis, which yielded 461 English-language articles, with 178 published in the previous 10 years. The search results were reviewed for applicability to albumin strategies for patients with cirrhosis.

There are three widely accepted and evidence-based indications for HA infusion in patients with cirrhosis, considered standard of care (Table).

Prevention of PICD

Therapeutic large-volume paracentesis (LVP) leads to a rise in plasma renin activity (PRA) centrally through several mechanisms and is not impacted by the rate of ascites removal.⁵ LVP relieves abdominal pressure, increasing venous return to the heart and cardiac output, and the corresponding drop in systemic vascular resistance with splanchnic vasodilation decreases effective circulating volume and activates the renin-angiotensin system. This PRA activation and circulatory dysfunction are associated with reaccumulating ascites, renal impairment, hypervolemic hyponatremia, and increased mortality.⁶ A large meta-analysis of 17 trials with 1225 patients found that HA infusion improves outcomes and reduces mortality for patients undergoing LVP (odds ratio [OR], 0.64; 95% CI, 0.41-0.98), reduces the risk of PICD more than other volume expanders tested, and lowers the incidence of hyponatremia.⁶ More recently, in 2017, Kütting et al⁷ analyzed 21 trials with 1277 patients and did not observe a significant mortality benefit for HA after LVP (OR, 0.78; 95% CI, 0.55-1.11). However, negative outcomes such as rise in PRA (OR, 0.53; 95% CI, 0.29-0.97) and hyponatremia (OR, 0.62; 95% CI, 0.42-0.94) were prevented. Guidelines recommend HA after LVP ≥5 L to prevent PICD, with a replacement volume of 6 to 8 g of albumin per liter of ascitic fluid removed.^2,3 Some patients may be at higher risk for PICD with less ascites removed, and the AASLD supports the use of HA to prevent PICD after smaller-volume paracentesis in patients who are already hypotensive (systolic blood pressure <90 mm Hg) or hyponatremic (<130 mmol/L), or have acute kidney injury.³

Spontaneous Bacterial Peritonitis

Spontaneous bacterial peritonitis is diagnosed by paracentesis, defined as ascitic neutrophil count ≥250 cells/µL with or without bacterascites (positive bacteriological culture). Bacterascites may be a precursor to the development of SBP, with the fluid neutrophil count of ≥250 determining the need for SBP treatment.² SBP can lead to circulatory dysfunction, hepatic encephalopathy, and HRS. Treating SBP with HA in addition to antibiotics reduces the risk of kidney injury compared with antibiotics alone (OR for kidney injury with antibiotics alone, 4.6; 95% CI, 1.3-16.1) and also reduces the risk of death (OR for mortality with antibiotics alone, 4.5; 95% CI, 1.0-20.9).⁸ The AASLD recommends albumin in addition to antibiotics in SBP to prevent HRS and acute kidney injury, and high-risk patients who already have kidney dysfunction (creatinine >1 mg/dL) or jaundice (total bilirubin >5 mg/dL) are more likely to benefit from albumin. The treatment schedule is 25% HA at 1.5 g/kg on day 1 and 1 g/kg on day 3.³ The EASL recommends administering HA to all patients with cirrhosis with SBP regardless of renal or liver indices. They acknowledge, however, that the incidence of SBP-associated acute kidney injury will be low in patients without severe hepatic disease or baseline renal impairment.²

Hepatorenal Syndrome

Albumin combined with vasoconstrictors is effective in treating HRS with a response rate of 20% to 80% (average, 50%).³ Vasoactive medications can include combination midodrine and octreotide or norepinephrine (or terlipressin outside of the United States). In patients with suspected HRS, the recommended dosing of 25% HA is 1 g/kg (to a maximum of 100 g of albumin) on day 1 and then 40 to 50 g daily for at least 3 days after the diagnosis is confirmed.³ The optimal duration of therapy beyond 3 days of combined therapy with midodrine, albumin, and octreotide is not established. Terlipressin treatment is recommended for a maximum of 14 days in cases of partial response or nonresponse in renal recovery.²

Hypoalbuminemia

Albumin administration to raise serum albumin levels in hospitalized patients has been a common practice. However, new evidence suggests that treating hypoalbuminemia with infusion of HA in hospitalized patients with decompensated cirrhosis does not protect patients from risk and causes harm. The Albumin To prevenT Infection in chronic liveR (ATTIRE) trial, published in 2021, randomly assigned 777 patients across 35 centers in the United Kingdom to receive daily 20% HA to target a serum albumin level of 3.0 g/dL vs standard care, including HA for established indications.^2,3 The primary end point was a composite of infection, kidney dysfunction, and death within 3 to 15 days of initiating treatment. There were no differences in the primary end point; secondary end points of death at 28 days, 3 months, or 6 months; or duration of hospitalization. The treatment group received 10 times more albumin than the control group and reported more adverse events, including pulmonary edema.⁹

Long-Term Treatment in Patients With Ascites

The human Albumin for the treatmeNt of aScites in patients With hEpatic ciRrhosis (ANSWER) trial, published in 2018, found improved 18-month survival in patients with cirrhosis and ascites treated with diuretics who received long-term albumin. This was an open-label trial of 431 patients at 33 sites in Italy, and the treatment arm received weekly infusions of 40 g of 20% HA. They observed a 38% reduction in mortality hazard ratio and half the number of hospital days annually.¹⁰ Based on these data and those from a 2006 Italian study with similar design and results, the Italian Association for the Study of the Liver (AISF) strongly recommends long-term albumin treatment in patients with cirrhosis with ascites.¹¹ The lead author on the ANSWER trial also authored the AISF statement, although this recommendation has not been adopted by the EASL or the AASLD.

Conversely, the Midodrine and Albumin for CirrHoTic patients (MACHT) trial, also published in 2018, randomly assigned 173 patients with ascites awaiting liver transplant to receive 40 g of HA every 15 days and midodrine in addition to standard care vs placebo. MACHT found no difference in mortality or complications at 1 year.¹²

Long-term albumin therapy as a preventive measure may be a disease modifier, taking advantage of the pleiotropic effects of albumin, though the differing conclusions from ANSWER and MACHT necessitate additional trials. The ongoing PRECIOSA study in Spain is assessing dosage and schedule for this therapy.¹³

Augmenting Diuresis

Loop diuretics are highly protein-bound, and, with hypoalbuminemia, there is less effective drug delivered to the site of action. One clinical approach is to augment diuretics with concomitant HA infusion. This approach is not supported by strong evidence or guidelines.

Hyponatremia

In a retrospective cohort study of 2435 hospitalized patients with cirrhosis, 1126 of whom had hyponatremia, those patients with sodium <130 mmol/L who received HA were more likely to have resolution of hyponatremia to >135 mmol/L. This was associated with improved 30-day survival.¹⁴ From this observational data, the AASLD supports the use of albumin combined with extreme fluid restriction (<1000 mL/d) for patients with severe hyponatremia (<120 mmol/L).³

Non-SBP Infections

A 2019 meta-analysis found no evidence of a benefit of HA for bacterial infections other than SBP. However, only three trials encompassing 407 patients met the inclusion criteria.¹⁵

SCCM Guideline Update: Hypoalbuminemia and Hypotension

The 2020 SCCM Guidelines for the Management of Adult Acute and Acute-on-Chronic Liver Failure in the ICU “suggest using albumin for resuscitation of patients [with liver failure] over other fluids, especially when serum albumin is low (<3 g/dL).” Acute-on-chronic liver failure is decompensation of cirrhosis combined with organ dysfunction (eg, coagulopathy, encephalopathy, kidney injury), a scenario that is frequently encountered by hospitalists outside of intensive care settings. In hypotensive patients with cirrhosis, the SCCM recommends administering albumin to a target mean arterial pressure of 65 mm Hg or otherwise adequate perfusion. This new recommendation is conditional, based on expert consensus, and derives from low-quality evidence, with acknowledgement that “costs may be prohibitive.”⁴

While the ATTIRE study demonstrated no benefit in treating hypoalbuminemia with infusion of HA in hospitalized patients with decompensated cirrhosis, the 2020 SCCM guidelines, released prior to the publication of the ATTIRE study, focused on more acutely ill patients. In the ATTIRE study, only 2% to 3% of the study population was in an intensive care unit.^4,9 The use of albumin infusion in the critically ill, hypoalbuminemic, hypotensive patient is not well studied, and the SCCM acknowledges the lack of supportive evidence for this practice in their guideline statement.

The three cardinal clinical indications for human albumin in patients with cirrhosis—prevention of PICD after LVP, in SBP, and for HRS—remain supported by the literature and guidelines, with the most recent guidance adding more nuance in patient selection based on individual risk (Table). With the publication of several large-scale studies in the past few years and a 2021 update to the AASLD guidance statement, clinicians have more evidence to guide their use of HA in patients with cirrhosis. In particular, the practice of treating isolated hypoalbuminemia with HA is no longer supported by the best evidence and is potentially harmful. A professional society recommendation to preferentially use albumin as a resuscitation fluid in hypoalbuminemia was made without the benefit of the results of the 2021 ATTIRE trial. On the horizon, additional results from ongoing and upcoming studies exploring concepts of effective albumin concentration and the pleiotropic properties of HA will impact the use of this therapy in hospitalized patients with cirrhosis.

Identifying and supporting patients’ care goals through shared decision-making was named the highest priority in the Improving Hospital Outcomes through Patient Engagement (i-HOPE) study.¹ Ensuring that seriously ill patients’ goals for their future care are understood and honored is particularly important for patients hospitalized with conditions known to be associated with high near-term mortality or functional disability, such as sepsis. It is increasingly recognized that a hospital admission for sepsis is associated with poor outcomes, including high rates of readmission and postdischarge mortality,^2-5 yet little is known about the assessment, status, and stability of patient care goals after discharge for sepsis. Using a cohort of high-risk sepsis survivors enrolled in a clinical trial, we aimed to determine how frequently care goals were documented, describe patterns in care goals, and evaluate how frequently care goals changed over 90 days after sepsis discharge. We also used expert reviewers to assess care delivered in the 90 days after hospitalization and determine the proportion of patients who received goal-concordant care.^6,7

Design, Setting, Participants

We conducted a secondary analysis using data from the Improving Morbidity During Post-Acute Care Transitions for Sepsis (IMPACTS) study,⁸ a pragmatic randomized trial evaluating the effectiveness of a multicomponent transition program to reduce mortality and rehospitalization after sepsis among patients enrolled from three hospitals between January 2019 and March 2020 (NCT03865602). The study intervention emphasized preference-sensitive care for patients but did not specifically require documentation of care goals in the electronic health record (EHR).

Data Collection

Clinical and outcomes data were collected from the EHR and enterprise data warehouse. We included data collected as part of routine care at IMPACTS trial enrollment (ie, age at admission, gender, race, marital status, coexisting conditions) and during index hospitalization (ie, organ failures, hospital length of stay, discharge disposition). The Charlson Comorbidity Index score was calculated from diagnosis codes captured during both inpatient and outpatient healthcare encounters in the 12 months prior to trial enrollment. The Centers for Disease Control and Prevention Adult Sepsis Event definitions⁹ were applied to measure organ failures.

Two palliative care physicians, three internal medicine physicians, and one critical care clinician retrospectively reviewed the EHR of study patients to: (1) identify whether patient care goals were documented in a standardized care alignment tool at discharge or in the subsequent 90 days; (2) categorize each patient’s care goals as focused on longevity, function, or comfort⁶ using either standardized documentation or unstructured information from the EHR; and (3) determine whether care goals changed over the first 90 days after discharge. Reviewers also classified care received over the 90-day postdischarge period as focused on longevity, function, or comfort. A random sample of 75 cases was selected for double review by a palliative care reviewer to assess interrater agreement in these assessments. Reviewers indicated whether the goal changed and, if so, what the new goal was. The data collection form is provided in the Appendix. The study was approved by the Atrium Health Institutional Review Board.

Outcomes

The primary outcome was the proportion of cases with care goals documented in the standardized care alignment tool, an EHR-embedded tool prompting questions about goals for future health states, including choices among longevity-, function-, and comfort-focused goals. A secondary outcome was the proportion of cases for which a goal could be determined using all information available in the EHR, such as family meeting notes, discharge summaries, and inpatient or outpatient visit notes. We also measured the proportion of patients who received goal-concordant care, defined as agreement between reviewers’ categorizations of patients’ goals and the primary focus of the care delivered, using a well-defined approach.⁶ In this approach, reviewers first categorized the care delivered during the 90 days after hospital discharge as focused on longevity, function, or comfort using clinical documentation in each patient’s medical record. To enhance transparency of this decision process, reviewers indicated which specific treatments (eg, new medications, hospital admission, hospice enrollment) supported their categorization. Reviewers then separately categorized the patient’s primary goal over the same period. Reviewer training emphasized that classifications of goals and care delivered should be independent. Patients were considered to have received goal-concordant care if the category of care delivered matched the category of the primary care goal. For patients with changing goals, care delivered was compared with the most recent documented goal.

Analyses

We characterized distributions of care goals and care delivered and reported rates of goal-concordant care overall and by care goals. We calculated weighted kappa statistics to assess interrater reliability. We conducted a multivariable logistic regression analysis in the full cohort to evaluate the association of standardized care goal documentation in the EHR with the dependent outcome of goal-concordant care, adjusting for other risk factors (ie, gender, race, marital status, coexisting chronic conditions, organ failures, and hospital length of stay).

Six hundred seventy-nine patients who survived to hospital discharge were included in the study; 52% of these patients were female, and 29% were Black. Median age of the cohort was 65 years (interquartile range [IQR], 55-74) and median Charlson Comorbidity Index score was 4 (IQR, 2-7). Study patients had a median two (IQR, 1-3) organ failures at admission and median hospital length of stay of 6 days (IQR, 4-10). One-hundred-sixty (24%) were discharged to home with health services, 145 (21%) to a skilled nursing facility, 44 (6%) to long-term acute care or other acute care hospital, and 21 (3%) to hospice. The Table shows characteristics between patients with goals documented and those without. Sepsis survivors with goals documented were older; had higher comorbidity scores, greater number of failed organs, and longer hospital length of stay; and were more frequently discharged to hospice or facility-based care settings compared with individuals without goals documented (all P < .05).

Characterization of Sepsis Survivors’ Goals

The Figure shows patterns of goal documentation and goal-concordant care in the study cohort. Care goals for sepsis survivors were documented in the standardized EHR care alignment tool at discharge for 130 (19%) patients. When reviewers used all information available in the EHR to categorize goals (73% interrater agreement; interrater reliability by weighted κ, 0.71; 95% CI, 0.58-0.83), reviewers were able to categorize patients’ goals in 269 (40%) cases. Among those categorized, goals were classified as prioritizing longevity in 95 (35%), function in 141 (52%), and comfort in 33 (12%) cases.

Goals changed over the 90-day observation period for 41 (6%) patients. Of patients whose goals changed, 15 (37%) initially had a goal focused on longevity, 24 (59%) had a goal focused on function, and 2 (5%) had a goal focused on comfort. Of goals that changed, the most frequent new goal was comfort, which was documented in 33 (80%) patients.

Characterization of Goal-Concordant Care

Interrater reliability was moderate for reviewer-based determination of care delivered (73% interrater agreement; weighted κ, 0.60; 95% CI, 0.43-0.78). Reviewers categorized care delivered as focused on longevity in 374 (55%), function in 290 (43%), and comfort in 13 (2%) patients, with <1% unable to be determined. Care elements most frequently cited for longevity-focused classification included intensive care unit (ICU) stay (39%) and new medications for nonsymptom benefit (29%). Care elements most frequently cited for function-focused classification included new medications for nonsymptom benefit (50%) and new medication for symptom benefit (41%). Care elements most frequently cited for comfort-focused classification included hospice enrollment (50%) and new medications for symptom benefit (48%). The rate of goal-concordant care was 68% among those with care goals determined and 27% when cases with unknown goals were classified as not concordant. Concordance was highest among those with longevity-focused (72%) and function-focused (73%) care goals compared with comfort-focused (39%) care goals (P < .01). Adjusting for other potential risk factors, completion of the standardized EHR care alignment tool was associated with higher odds of receiving goal-concordant care (OR, 3.6; 95% CI, 2.4-5.5).

Our study identified deficits in the current delivery of goal-concordant care in the first 90 days after sepsis hospitalization. First, goals were only documented in the standardized EHR care alignment tool in one-fifth of cases. Otherwise, information about goals, values, and treatment preferences of sepsis patients was documented idiosyncratically in progress notes, which may not be apparent to clinicians involved in patients’ future care. Lack of clinician attention to documenting the goals of sepsis patients post discharge may reflect suboptimal awareness of the lasting health consequences of sepsis, including persistently elevated risk of mortality up to 2 years following the index hospitalization.^2-5 Second, even when goals could be classified by reviewers, the focus of care delivered did not match patients’ goals in nearly one-third of cases.

Our findings inspire several considerations for postsepsis care during hospitalization or in the peridischarge period. First, efforts should focus on increasing assessment and documentation of sepsis survivors’ goals—this might begin with enhanced education about the lasting health consequences after sepsis and communication skills training. Importantly, sepsis survivors’ goals were relatively stable over 90 days after discharge, suggesting that hospitalization for sepsis represents an important opportunity to assess and document patients’ goals. Improving documentation of care goals explicitly in a standardized EHR tool may be an important target for quality-improvement initiatives, as this practice was associated with higher odds of receiving goal-concordant care in our cohort. Second, our findings that one-third of patients received care that was not consistent with their goals is worrisome. Concordance was lowest among comfort-focused care goals, suggesting that some of the high rates of healthcare utilization after sepsis may be unwanted.^10-12 For example, ICU stay and new medication for nonsymptom benefit were commonly cited as indications of longevity-focused care among patients with comfort-focused goals. Thus, improving the alignment between sepsis survivors’ goals and subsequent care received is an important target from both a patient-centered and value perspective. Consistent with the recommendations of the i-HOPE study,¹ future interventions designed to improve posthospitalization care of sepsis patients should aim to capture goal-concordant care as a patient-centered outcome, if possible.

Our examination of goals and goal-concordant care after sepsis hospitalization advances the goal of enhancing understanding of survivorship in this population.⁴ Strengths of this study include the large, real-world sample and use of expert palliative care physicians conducting granular EHR review to assess goal-concordant care. Our utilization of this methodology to evaluate goal-concordant care provides information to refine efforts toward developing reliable measures of this important outcome—for example, interrater reliability was similar among reviewers in our study compared with studies assessing goal-concordant care using similar methodology.¹³

Limitations include potential generalizability challenges for goal and goal-concordant care assessments in other health systems with different EHR platforms or local documentation practices, although deficits in EHR documentation of care goals have been reported in other settings.^14,15 We double-reviewed a sample of cases to evaluate interrater reliability, but double-review of all cases with a discussion and adjudication approach may have increased the number of goals that could ultimately be classified. However, this might overestimate the number of goals that are identifiable in real-world practice by a treating clinician. Finally, reviewers may have been challenged to select one goal when two or more competing goals existed. Future refinements of goal-concordant care measurement will need to define methods for handling tradeoffs and prioritization associated with competing goals.

The hospitalization and peridischarge periods represent an important opportunity to address deficits in the documentation of goals and provision of goal-concordant care for sepsis survivors. Doing so may improve patient-centered care and reduce the high rates of healthcare utilization after sepsis.

Identifying and supporting patients’ care goals through shared decision-making was named the highest priority in the Improving Hospital Outcomes through Patient Engagement (i-HOPE) study.¹ Ensuring that seriously ill patients’ goals for their future care are understood and honored is particularly important for patients hospitalized with conditions known to be associated with high near-term mortality or functional disability, such as sepsis. It is increasingly recognized that a hospital admission for sepsis is associated with poor outcomes, including high rates of readmission and postdischarge mortality,^2-5 yet little is known about the assessment, status, and stability of patient care goals after discharge for sepsis. Using a cohort of high-risk sepsis survivors enrolled in a clinical trial, we aimed to determine how frequently care goals were documented, describe patterns in care goals, and evaluate how frequently care goals changed over 90 days after sepsis discharge. We also used expert reviewers to assess care delivered in the 90 days after hospitalization and determine the proportion of patients who received goal-concordant care.^6,7

Design, Setting, Participants

We conducted a secondary analysis using data from the Improving Morbidity During Post-Acute Care Transitions for Sepsis (IMPACTS) study,⁸ a pragmatic randomized trial evaluating the effectiveness of a multicomponent transition program to reduce mortality and rehospitalization after sepsis among patients enrolled from three hospitals between January 2019 and March 2020 (NCT03865602). The study intervention emphasized preference-sensitive care for patients but did not specifically require documentation of care goals in the electronic health record (EHR).

Data Collection

Clinical and outcomes data were collected from the EHR and enterprise data warehouse. We included data collected as part of routine care at IMPACTS trial enrollment (ie, age at admission, gender, race, marital status, coexisting conditions) and during index hospitalization (ie, organ failures, hospital length of stay, discharge disposition). The Charlson Comorbidity Index score was calculated from diagnosis codes captured during both inpatient and outpatient healthcare encounters in the 12 months prior to trial enrollment. The Centers for Disease Control and Prevention Adult Sepsis Event definitions⁹ were applied to measure organ failures.

Two palliative care physicians, three internal medicine physicians, and one critical care clinician retrospectively reviewed the EHR of study patients to: (1) identify whether patient care goals were documented in a standardized care alignment tool at discharge or in the subsequent 90 days; (2) categorize each patient’s care goals as focused on longevity, function, or comfort⁶ using either standardized documentation or unstructured information from the EHR; and (3) determine whether care goals changed over the first 90 days after discharge. Reviewers also classified care received over the 90-day postdischarge period as focused on longevity, function, or comfort. A random sample of 75 cases was selected for double review by a palliative care reviewer to assess interrater agreement in these assessments. Reviewers indicated whether the goal changed and, if so, what the new goal was. The data collection form is provided in the Appendix. The study was approved by the Atrium Health Institutional Review Board.

Outcomes

The primary outcome was the proportion of cases with care goals documented in the standardized care alignment tool, an EHR-embedded tool prompting questions about goals for future health states, including choices among longevity-, function-, and comfort-focused goals. A secondary outcome was the proportion of cases for which a goal could be determined using all information available in the EHR, such as family meeting notes, discharge summaries, and inpatient or outpatient visit notes. We also measured the proportion of patients who received goal-concordant care, defined as agreement between reviewers’ categorizations of patients’ goals and the primary focus of the care delivered, using a well-defined approach.⁶ In this approach, reviewers first categorized the care delivered during the 90 days after hospital discharge as focused on longevity, function, or comfort using clinical documentation in each patient’s medical record. To enhance transparency of this decision process, reviewers indicated which specific treatments (eg, new medications, hospital admission, hospice enrollment) supported their categorization. Reviewers then separately categorized the patient’s primary goal over the same period. Reviewer training emphasized that classifications of goals and care delivered should be independent. Patients were considered to have received goal-concordant care if the category of care delivered matched the category of the primary care goal. For patients with changing goals, care delivered was compared with the most recent documented goal.

Analyses

We characterized distributions of care goals and care delivered and reported rates of goal-concordant care overall and by care goals. We calculated weighted kappa statistics to assess interrater reliability. We conducted a multivariable logistic regression analysis in the full cohort to evaluate the association of standardized care goal documentation in the EHR with the dependent outcome of goal-concordant care, adjusting for other risk factors (ie, gender, race, marital status, coexisting chronic conditions, organ failures, and hospital length of stay).

Six hundred seventy-nine patients who survived to hospital discharge were included in the study; 52% of these patients were female, and 29% were Black. Median age of the cohort was 65 years (interquartile range [IQR], 55-74) and median Charlson Comorbidity Index score was 4 (IQR, 2-7). Study patients had a median two (IQR, 1-3) organ failures at admission and median hospital length of stay of 6 days (IQR, 4-10). One-hundred-sixty (24%) were discharged to home with health services, 145 (21%) to a skilled nursing facility, 44 (6%) to long-term acute care or other acute care hospital, and 21 (3%) to hospice. The Table shows characteristics between patients with goals documented and those without. Sepsis survivors with goals documented were older; had higher comorbidity scores, greater number of failed organs, and longer hospital length of stay; and were more frequently discharged to hospice or facility-based care settings compared with individuals without goals documented (all P < .05).

Characterization of Sepsis Survivors’ Goals

The Figure shows patterns of goal documentation and goal-concordant care in the study cohort. Care goals for sepsis survivors were documented in the standardized EHR care alignment tool at discharge for 130 (19%) patients. When reviewers used all information available in the EHR to categorize goals (73% interrater agreement; interrater reliability by weighted κ, 0.71; 95% CI, 0.58-0.83), reviewers were able to categorize patients’ goals in 269 (40%) cases. Among those categorized, goals were classified as prioritizing longevity in 95 (35%), function in 141 (52%), and comfort in 33 (12%) cases.

Goals changed over the 90-day observation period for 41 (6%) patients. Of patients whose goals changed, 15 (37%) initially had a goal focused on longevity, 24 (59%) had a goal focused on function, and 2 (5%) had a goal focused on comfort. Of goals that changed, the most frequent new goal was comfort, which was documented in 33 (80%) patients.

Characterization of Goal-Concordant Care

Interrater reliability was moderate for reviewer-based determination of care delivered (73% interrater agreement; weighted κ, 0.60; 95% CI, 0.43-0.78). Reviewers categorized care delivered as focused on longevity in 374 (55%), function in 290 (43%), and comfort in 13 (2%) patients, with <1% unable to be determined. Care elements most frequently cited for longevity-focused classification included intensive care unit (ICU) stay (39%) and new medications for nonsymptom benefit (29%). Care elements most frequently cited for function-focused classification included new medications for nonsymptom benefit (50%) and new medication for symptom benefit (41%). Care elements most frequently cited for comfort-focused classification included hospice enrollment (50%) and new medications for symptom benefit (48%). The rate of goal-concordant care was 68% among those with care goals determined and 27% when cases with unknown goals were classified as not concordant. Concordance was highest among those with longevity-focused (72%) and function-focused (73%) care goals compared with comfort-focused (39%) care goals (P < .01). Adjusting for other potential risk factors, completion of the standardized EHR care alignment tool was associated with higher odds of receiving goal-concordant care (OR, 3.6; 95% CI, 2.4-5.5).

Our study identified deficits in the current delivery of goal-concordant care in the first 90 days after sepsis hospitalization. First, goals were only documented in the standardized EHR care alignment tool in one-fifth of cases. Otherwise, information about goals, values, and treatment preferences of sepsis patients was documented idiosyncratically in progress notes, which may not be apparent to clinicians involved in patients’ future care. Lack of clinician attention to documenting the goals of sepsis patients post discharge may reflect suboptimal awareness of the lasting health consequences of sepsis, including persistently elevated risk of mortality up to 2 years following the index hospitalization.^2-5 Second, even when goals could be classified by reviewers, the focus of care delivered did not match patients’ goals in nearly one-third of cases.

Our findings inspire several considerations for postsepsis care during hospitalization or in the peridischarge period. First, efforts should focus on increasing assessment and documentation of sepsis survivors’ goals—this might begin with enhanced education about the lasting health consequences after sepsis and communication skills training. Importantly, sepsis survivors’ goals were relatively stable over 90 days after discharge, suggesting that hospitalization for sepsis represents an important opportunity to assess and document patients’ goals. Improving documentation of care goals explicitly in a standardized EHR tool may be an important target for quality-improvement initiatives, as this practice was associated with higher odds of receiving goal-concordant care in our cohort. Second, our findings that one-third of patients received care that was not consistent with their goals is worrisome. Concordance was lowest among comfort-focused care goals, suggesting that some of the high rates of healthcare utilization after sepsis may be unwanted.^10-12 For example, ICU stay and new medication for nonsymptom benefit were commonly cited as indications of longevity-focused care among patients with comfort-focused goals. Thus, improving the alignment between sepsis survivors’ goals and subsequent care received is an important target from both a patient-centered and value perspective. Consistent with the recommendations of the i-HOPE study,¹ future interventions designed to improve posthospitalization care of sepsis patients should aim to capture goal-concordant care as a patient-centered outcome, if possible.

Our examination of goals and goal-concordant care after sepsis hospitalization advances the goal of enhancing understanding of survivorship in this population.⁴ Strengths of this study include the large, real-world sample and use of expert palliative care physicians conducting granular EHR review to assess goal-concordant care. Our utilization of this methodology to evaluate goal-concordant care provides information to refine efforts toward developing reliable measures of this important outcome—for example, interrater reliability was similar among reviewers in our study compared with studies assessing goal-concordant care using similar methodology.¹³

Limitations include potential generalizability challenges for goal and goal-concordant care assessments in other health systems with different EHR platforms or local documentation practices, although deficits in EHR documentation of care goals have been reported in other settings.^14,15 We double-reviewed a sample of cases to evaluate interrater reliability, but double-review of all cases with a discussion and adjudication approach may have increased the number of goals that could ultimately be classified. However, this might overestimate the number of goals that are identifiable in real-world practice by a treating clinician. Finally, reviewers may have been challenged to select one goal when two or more competing goals existed. Future refinements of goal-concordant care measurement will need to define methods for handling tradeoffs and prioritization associated with competing goals.

The hospitalization and peridischarge periods represent an important opportunity to address deficits in the documentation of goals and provision of goal-concordant care for sepsis survivors. Doing so may improve patient-centered care and reduce the high rates of healthcare utilization after sepsis.

Identifying and supporting patients’ care goals through shared decision-making was named the highest priority in the Improving Hospital Outcomes through Patient Engagement (i-HOPE) study.¹ Ensuring that seriously ill patients’ goals for their future care are understood and honored is particularly important for patients hospitalized with conditions known to be associated with high near-term mortality or functional disability, such as sepsis. It is increasingly recognized that a hospital admission for sepsis is associated with poor outcomes, including high rates of readmission and postdischarge mortality,^2-5 yet little is known about the assessment, status, and stability of patient care goals after discharge for sepsis. Using a cohort of high-risk sepsis survivors enrolled in a clinical trial, we aimed to determine how frequently care goals were documented, describe patterns in care goals, and evaluate how frequently care goals changed over 90 days after sepsis discharge. We also used expert reviewers to assess care delivered in the 90 days after hospitalization and determine the proportion of patients who received goal-concordant care.^6,7

Design, Setting, Participants

We conducted a secondary analysis using data from the Improving Morbidity During Post-Acute Care Transitions for Sepsis (IMPACTS) study,⁸ a pragmatic randomized trial evaluating the effectiveness of a multicomponent transition program to reduce mortality and rehospitalization after sepsis among patients enrolled from three hospitals between January 2019 and March 2020 (NCT03865602). The study intervention emphasized preference-sensitive care for patients but did not specifically require documentation of care goals in the electronic health record (EHR).

Data Collection

Clinical and outcomes data were collected from the EHR and enterprise data warehouse. We included data collected as part of routine care at IMPACTS trial enrollment (ie, age at admission, gender, race, marital status, coexisting conditions) and during index hospitalization (ie, organ failures, hospital length of stay, discharge disposition). The Charlson Comorbidity Index score was calculated from diagnosis codes captured during both inpatient and outpatient healthcare encounters in the 12 months prior to trial enrollment. The Centers for Disease Control and Prevention Adult Sepsis Event definitions⁹ were applied to measure organ failures.

Two palliative care physicians, three internal medicine physicians, and one critical care clinician retrospectively reviewed the EHR of study patients to: (1) identify whether patient care goals were documented in a standardized care alignment tool at discharge or in the subsequent 90 days; (2) categorize each patient’s care goals as focused on longevity, function, or comfort⁶ using either standardized documentation or unstructured information from the EHR; and (3) determine whether care goals changed over the first 90 days after discharge. Reviewers also classified care received over the 90-day postdischarge period as focused on longevity, function, or comfort. A random sample of 75 cases was selected for double review by a palliative care reviewer to assess interrater agreement in these assessments. Reviewers indicated whether the goal changed and, if so, what the new goal was. The data collection form is provided in the Appendix. The study was approved by the Atrium Health Institutional Review Board.

Outcomes

The primary outcome was the proportion of cases with care goals documented in the standardized care alignment tool, an EHR-embedded tool prompting questions about goals for future health states, including choices among longevity-, function-, and comfort-focused goals. A secondary outcome was the proportion of cases for which a goal could be determined using all information available in the EHR, such as family meeting notes, discharge summaries, and inpatient or outpatient visit notes. We also measured the proportion of patients who received goal-concordant care, defined as agreement between reviewers’ categorizations of patients’ goals and the primary focus of the care delivered, using a well-defined approach.⁶ In this approach, reviewers first categorized the care delivered during the 90 days after hospital discharge as focused on longevity, function, or comfort using clinical documentation in each patient’s medical record. To enhance transparency of this decision process, reviewers indicated which specific treatments (eg, new medications, hospital admission, hospice enrollment) supported their categorization. Reviewers then separately categorized the patient’s primary goal over the same period. Reviewer training emphasized that classifications of goals and care delivered should be independent. Patients were considered to have received goal-concordant care if the category of care delivered matched the category of the primary care goal. For patients with changing goals, care delivered was compared with the most recent documented goal.

Analyses

We characterized distributions of care goals and care delivered and reported rates of goal-concordant care overall and by care goals. We calculated weighted kappa statistics to assess interrater reliability. We conducted a multivariable logistic regression analysis in the full cohort to evaluate the association of standardized care goal documentation in the EHR with the dependent outcome of goal-concordant care, adjusting for other risk factors (ie, gender, race, marital status, coexisting chronic conditions, organ failures, and hospital length of stay).

Six hundred seventy-nine patients who survived to hospital discharge were included in the study; 52% of these patients were female, and 29% were Black. Median age of the cohort was 65 years (interquartile range [IQR], 55-74) and median Charlson Comorbidity Index score was 4 (IQR, 2-7). Study patients had a median two (IQR, 1-3) organ failures at admission and median hospital length of stay of 6 days (IQR, 4-10). One-hundred-sixty (24%) were discharged to home with health services, 145 (21%) to a skilled nursing facility, 44 (6%) to long-term acute care or other acute care hospital, and 21 (3%) to hospice. The Table shows characteristics between patients with goals documented and those without. Sepsis survivors with goals documented were older; had higher comorbidity scores, greater number of failed organs, and longer hospital length of stay; and were more frequently discharged to hospice or facility-based care settings compared with individuals without goals documented (all P < .05).

Characterization of Sepsis Survivors’ Goals

The Figure shows patterns of goal documentation and goal-concordant care in the study cohort. Care goals for sepsis survivors were documented in the standardized EHR care alignment tool at discharge for 130 (19%) patients. When reviewers used all information available in the EHR to categorize goals (73% interrater agreement; interrater reliability by weighted κ, 0.71; 95% CI, 0.58-0.83), reviewers were able to categorize patients’ goals in 269 (40%) cases. Among those categorized, goals were classified as prioritizing longevity in 95 (35%), function in 141 (52%), and comfort in 33 (12%) cases.

Goals changed over the 90-day observation period for 41 (6%) patients. Of patients whose goals changed, 15 (37%) initially had a goal focused on longevity, 24 (59%) had a goal focused on function, and 2 (5%) had a goal focused on comfort. Of goals that changed, the most frequent new goal was comfort, which was documented in 33 (80%) patients.

Characterization of Goal-Concordant Care

Interrater reliability was moderate for reviewer-based determination of care delivered (73% interrater agreement; weighted κ, 0.60; 95% CI, 0.43-0.78). Reviewers categorized care delivered as focused on longevity in 374 (55%), function in 290 (43%), and comfort in 13 (2%) patients, with <1% unable to be determined. Care elements most frequently cited for longevity-focused classification included intensive care unit (ICU) stay (39%) and new medications for nonsymptom benefit (29%). Care elements most frequently cited for function-focused classification included new medications for nonsymptom benefit (50%) and new medication for symptom benefit (41%). Care elements most frequently cited for comfort-focused classification included hospice enrollment (50%) and new medications for symptom benefit (48%). The rate of goal-concordant care was 68% among those with care goals determined and 27% when cases with unknown goals were classified as not concordant. Concordance was highest among those with longevity-focused (72%) and function-focused (73%) care goals compared with comfort-focused (39%) care goals (P < .01). Adjusting for other potential risk factors, completion of the standardized EHR care alignment tool was associated with higher odds of receiving goal-concordant care (OR, 3.6; 95% CI, 2.4-5.5).

Our study identified deficits in the current delivery of goal-concordant care in the first 90 days after sepsis hospitalization. First, goals were only documented in the standardized EHR care alignment tool in one-fifth of cases. Otherwise, information about goals, values, and treatment preferences of sepsis patients was documented idiosyncratically in progress notes, which may not be apparent to clinicians involved in patients’ future care. Lack of clinician attention to documenting the goals of sepsis patients post discharge may reflect suboptimal awareness of the lasting health consequences of sepsis, including persistently elevated risk of mortality up to 2 years following the index hospitalization.^2-5 Second, even when goals could be classified by reviewers, the focus of care delivered did not match patients’ goals in nearly one-third of cases.

Our findings inspire several considerations for postsepsis care during hospitalization or in the peridischarge period. First, efforts should focus on increasing assessment and documentation of sepsis survivors’ goals—this might begin with enhanced education about the lasting health consequences after sepsis and communication skills training. Importantly, sepsis survivors’ goals were relatively stable over 90 days after discharge, suggesting that hospitalization for sepsis represents an important opportunity to assess and document patients’ goals. Improving documentation of care goals explicitly in a standardized EHR tool may be an important target for quality-improvement initiatives, as this practice was associated with higher odds of receiving goal-concordant care in our cohort. Second, our findings that one-third of patients received care that was not consistent with their goals is worrisome. Concordance was lowest among comfort-focused care goals, suggesting that some of the high rates of healthcare utilization after sepsis may be unwanted.^10-12 For example, ICU stay and new medication for nonsymptom benefit were commonly cited as indications of longevity-focused care among patients with comfort-focused goals. Thus, improving the alignment between sepsis survivors’ goals and subsequent care received is an important target from both a patient-centered and value perspective. Consistent with the recommendations of the i-HOPE study,¹ future interventions designed to improve posthospitalization care of sepsis patients should aim to capture goal-concordant care as a patient-centered outcome, if possible.

Our examination of goals and goal-concordant care after sepsis hospitalization advances the goal of enhancing understanding of survivorship in this population.⁴ Strengths of this study include the large, real-world sample and use of expert palliative care physicians conducting granular EHR review to assess goal-concordant care. Our utilization of this methodology to evaluate goal-concordant care provides information to refine efforts toward developing reliable measures of this important outcome—for example, interrater reliability was similar among reviewers in our study compared with studies assessing goal-concordant care using similar methodology.¹³

Limitations include potential generalizability challenges for goal and goal-concordant care assessments in other health systems with different EHR platforms or local documentation practices, although deficits in EHR documentation of care goals have been reported in other settings.^14,15 We double-reviewed a sample of cases to evaluate interrater reliability, but double-review of all cases with a discussion and adjudication approach may have increased the number of goals that could ultimately be classified. However, this might overestimate the number of goals that are identifiable in real-world practice by a treating clinician. Finally, reviewers may have been challenged to select one goal when two or more competing goals existed. Future refinements of goal-concordant care measurement will need to define methods for handling tradeoffs and prioritization associated with competing goals.

The hospitalization and peridischarge periods represent an important opportunity to address deficits in the documentation of goals and provision of goal-concordant care for sepsis survivors. Doing so may improve patient-centered care and reduce the high rates of healthcare utilization after sepsis.

The rate of inpatient admissions among adults aged 65 years and older has decreased by approximately 25% since 2000.^1,2 This long-term trend raises important questions about inpatient-related spending in the traditional Medicare program for hospitals and providers who treat beneficiaries after a hospitalization. As traditional Medicare’s most expensive sector (accounting for 21% of all Medicare spending³), reducing hospitalizations is often championed as an opportunity to moderate Medicare spending growth.

Medicare’s ability to achieve significant savings from declining inpatient use may be tempered by a shift toward more expensive hospitalizations. If marginal hospitalizations among healthier beneficiaries are avoided, then the remaining inpatient users may be sicker and have greater spending per hospitalization and greater need for follow-up services. This study examines trends in Medicare spending related to episodes initiated by an inpatient stay because of its importance to overall Medicare spending and the implications for several Medicare value-based payment initiatives. In care models seeking to contain spending at a population level, such as accountable care organizations and managed care plans, reducing inpatient use and associated services may have the largest impact in curbing overall spending growth per beneficiary. Other models focused on spending at an episode level, including bundled payment initiatives, may face challenges if inpatient episodes become more expensive over time.

As Medicare shifts toward value-based payments, hospitalists and other hospital leaders are often involved in redesigning care delivery models for the hospital or accountable care organization (eg, through readmission reduction initiatives, post–acute care coordination, and bundled-care delivery programs). Not all savings strategies rely on providers to change how services are delivered; Medicare can modify payment rates, such as Affordable Care Act provisions that slowed how quickly Medicare payment rates increased.⁴ For clinicians to navigate the shift toward new payment models, it is important to recognize how each of these elements—declining hospital admissions, spending per inpatient episode, and payment rates—affect spending trends for inpatient services and associated care. Previous articles on overall Medicare inpatient spending have examined inpatient stays alone⁵ or focused mainly on spending per episode^6,7 without quantifying how these elements contributed to overall episode-related Medicare spending per beneficiary. This article addresses this gap by demonstrating how inpatient-related spending trends reflect each component.

This study examined trends in Medicare’s spending on inpatient episodes during the years 2009 to 2017. We described changes in the volume and spending on inpatient-initiated episodes across several dimensions, including beneficiary-level and hospitalization-level factors. We examined whether declines in spending associated with fewer inpatient-initiated episodes have been offset by increased spending per episode and how spending would have differed without changes in Medicare payment rates.

Episode Definition

We constructed an episode measure that captured traditional Medicare spending for 30 days prior to hospital admission, hospitalization duration, and 90 days following hospital discharge (additional details in the Appendix). As with bundled payments that include pre- and postdischarge services, this window allowed us to observe any services that preceded or followed a hospitalization. Using Medicare Part A, B, and D claims data for the years 2008 to 2018, we captured spending across all sectors for episodes with an index hospital admission in the years 2009 to 2017. If a beneficiary died during an episode, we measured healthcare spending until death. We excluded episodes where beneficiaries did not have traditional Medicare Part A and B for the entire episode or lived outside the 50 states or Washington, DC.

Any acute hospitalization triggered a new episode, with one exception: if a beneficiary was discharged and readmitted within 90 days for the same diagnosis-related group (DRG), then the readmission did not trigger a new episode. The spending for that readmission was attributed to the prior hospital stay. In effect, the annual number of episodes is equivalent to the annual number of hospital admissions minus subsequent rehospitalizations for the same DRG. Neither observation stays nor hospitalizations in inpatient rehabilitation, psychiatric, or long-term facilities were considered acute hospital admissions.

We assigned claims from noninpatient sectors to an episode based on whether the claim start date fell within the episode window. All traditional Medicare sectors were measured, including outpatient services, physician claims, post–acute care services, and Medicare Part D prescription drug events.

Our analysis aimed to measure all spending related to inpatient episodes without double-counting spending for overlapping episodes. If episodes overlapped, then spending for overlapping days was weighted to be evenly divided across episodes.

Outcome Measures

The study’s main outcomes summarized episode trends across the entire traditional Medicare population, including beneficiaries without an episode, in annual mean number of episodes per beneficiary and annual mean episode-related spending per beneficiary. The denominator of these measures is person-years, or total number of beneficiary months with Medicare Part A and B coverage divided by 12. The annual mean number of episodes per beneficiary is the total number of episodes initiated in a calendar year divided by person-years. The annual mean episode-related spending per beneficiary is the total amount of spending attributed to episodes divided by person-years. We also measured annual mean spending per episode, or total amount of spending attributed to episodes divided by the total number of episodes.

Medicare annually updates each sector’s payment rates for several factors, including inflation. We constructed an index for each sector to adjust for these annual payment rate changes. We also accounted for sequestration measures in effect since April 2013 that reduced Medicare payments to all sectors by 2%. We report our spending measures twice, with and without adjusting for changes in payment rates. Adjusted numbers reflect payment rates in effect in 2015.

Analysis Approach

We present annual trends on changes in the number of inpatient episodes per beneficiary, mean episode-related spending per beneficiary, and mean spending per episode. To quantify how changes in episode-related spending per beneficiary reflect changes in the number of episodes per beneficiary vs changes in spending per episode, we modified an approach implemented by Rosen and colleagues.⁸We calculated how much episode-related spending per beneficiary would have changed between 2009 and 2017 if spending per episode remained at 2009 levels but the number of episodes per beneficiary declined, as observed between 2009 and 2017 (see example calculation in Appendix). Conversely, we estimated how much episode-related spending per beneficiary would have changed if the number of episodes per beneficiary remained at 2009 levels but spending per episode increased, as observed between 2009 and 2017. In reality, the number of episodes per beneficiary and spending per episode concurrently changed, so the decomposition also includes an interaction term that quantifies how much of the change in spending reflects changes in both factors. We present these estimates for all sectors and separately for inpatient and all other sectors.

To better understand which beneficiaries have declining inpatient use, we performed stratified analyses describing changes in the number of episodes per beneficiary between 2009 and 2017, spending per episode, and total episode-related spending per beneficiary. We report these measures for several subpopulations defined by age, sex, race, dual-eligible status, and whether the beneficiary used long-term nursing home services during the episode’s calendar year. Descriptive statistics also detail how these measures changed between 2009 and 2017 for episodes stratified by characteristics of the index hospital stay: planned vs unplanned, medical vs surgical, and any use of intensive care unit (ICU) or coronary care unit services. We also stratify study measures by whether an episode included any use of post–acute care services (skilled nursing facility, home health, or inpatient rehabilitation facility use). Finally, we aggregate the episodes into major diagnostic categories (MDCs) based on the index hospital stay’s DRG to report study outcomes by condition. Because of a shift in coding hospitalizations for pneumonia as sepsis,^9,10 we exclude these two diseases from their respective MDCs and analyze them jointly as a unique category.

Changes in Number of Inpatient Episodes and Related Spending

From 2009 to 2017, the number of inpatient episodes per 1000 traditional Medicare beneficiaries declined from 326 to 267 (Table 1), or a relative decline of 18.2% (Figure 1). The total volume of inpatient episodes declined by only 13.4%, from 10.2 million to 8.8 million, reflecting that the size of the traditional Medicare population grew during these years. Over the same years, mean payment-rate–adjusted spending per episode increased 11.4% from $20,891 to $23,273.

When considering overall episode-related spending, the large decline in the volume of episodes outweighed increased spending per episode: the mean amount of episode-related Medicare spending per beneficiary decreased 8.9% from $6810 to $6206 (Table 1), or a net change of $604 (Figure 2). This net change reflects decreased spending due to fewer episodes per beneficiary ($1239 reduction in episode-related spending) offset by increased spending per episode (translating to a $776 increase in episode-related spending per beneficiary). These two factors, plus their interaction reflecting the combined influence of these factors ($141), comprise the overall change in episode-related spending per beneficiary over this period.

When these estimates are calculated separately for the inpatient sector and all other sectors, the inpatient sector experienced small increases in spending associated with greater spending per episode ($304) compared with noninpatient sectors ($472). Accordingly, the inpatient sector had a larger net decline in episode-related spending per beneficiary ($420) than noninpatient sectors ($184) after taking into account declining episode volume.

As expected, episode-related spending increased more when measures were not adjusted for annual payment rate increases. Without such adjustment, mean spending per episode increased 25.5%, and episode-related spending per beneficiary was nearly flat (2.6% between 2009 and 2017 [Figure 1]). The decline in unadjusted spending associated with fewer episodes ($1138) was offset by the spending increase associated with higher spending per episode ($1592) (Figure 2).

Analyses Stratified by Beneficiary Characteristics

Every population examined had declines in the number of inpatient episodes, even beneficiaries with more frequent inpatient use (Table 2). Among Medicare beneficiaries aged 85 years and older, the mean number of episodes per 1000 beneficiaries declined by 12.7%, from 524 to 457. Populations with less frequent inpatient use often experienced larger relative declines in number of episodes than populations with more frequent inpatient use. For example, the mean number of episodes per 1000 beneficiaries decreased by 17.7% for beneficiaries without nursing home use (306 to 252), as compared with an 8.1% decline for beneficiaries with nursing home use (from 888 to 816). In contrast, populations with less frequent inpatient use had larger relative increases in spending per episode with adjustment for payment rate changes. For example, spending per episode increased by 13.1% for beneficiaries aged 65 to 74 years ($20,904 to $23,644), but only by 8.6% for beneficiaries 85 years and older ($20,384 to $22,138).

Analyses Stratified by Service Use Characteristics

Some types of inpatient episodes had larger declines in the number of episodes, including episodes with planned admissions for the index hospital stay (28.8% decline from 68 to 48 episodes per 1000 beneficiaries) and episodes without post–acute care use (23.9% decline from 169 to 129 episodes per 1000 beneficiaries) (Appendix Table). In contrast, declines in the number of episodes were similar for index hospital admissions that did or did not involve ICU use (17.8% and 18.3% reduction in mean number of episodes per 1000 beneficiaries, respectively) or that included a surgical procedure or not (17.1% versus 18.6%, respectively). Several types of inpatient episodes had larger increases in spending per episode, such as a 15.1% increase for planned admissions and a 13.2% increase for hospitalizations without ICU use.

According to diagnosis information for an episode’s index hospital stay, inpatient episodes related to conditions affecting the circulatory system had the largest decline in mean number of episodes, decreasing by 31.8% from 78 to 53 episodes per 1000 beneficiaries (Appendix Table). Episodes for other diseases had much smaller declines in volume. Admissions for diagnoses of pneumonia or sepsis had notable increases in the volume of episodes, increasing by 20.7% from 25 to 30 admissions per 1000 beneficiaries.

Medicare spending per beneficiary on inpatient episodes, including services provided pre- and post hospitalization, declined by 8.9% from 2009 to 2017 after adjusting for payment rate changes. This decline reflects two components. First, the number of episodes per 1000 beneficiaries declined by 18.2%. Although the extent of this decrease varied across populations, every group examined had declines in inpatient use. In particular, hospitalizations for conditions affecting the circulatory system, such as heart attacks and cardiac procedures, decreased. Second, as inpatient volume declined, spending per episode increased by 11.4% to an average of $23,273 in 2017. This increase in spending per episode offset how much overall Medicare spending on episode-related care declined.

Medicare is increasingly challenging hospitals to demonstrate the value of inpatient services and associated treatment, which requires hospital leaders to recognize how their facilities’ spending trends relate to these national patterns. Understanding how much national episode-related spending has decreased over time with declining inpatient volume can help an accountable care organization evaluate whether it is feasible to achieve significant savings by reducing hospitalizations. Bundled payment providers focused on managing spending per episode can benefit from identifying which types of hospitalizations have increased spending per episode, especially for certain diagnoses.

These results also highlight the continued importance of a perennial factor in Medicare spending: payment rates. If Medicare payment rates had not increased over our study period, Medicare spending per inpatient episode would have increased by only 11%. Actual Medicare spending per episode increased by 25%, demonstrating that over half of the relative increase in spending per episode reflected increases in Medicare’s payment rates.

Increased spending per episode, even after adjustment for payment rate changes, suggests that services provided during an episode have increased in intensity or shifted toward higher-cost treatments. In order to understand how Medicare episode-related spending changed without introducing assumptions about factors underlying that change, our analysis did not adjust for inpatient acuity. We observed a smaller decline in the volume of hospitalizations with post–acute care use but similar decreases in the volume of hospitalizations with and without ICU use. This finding is consistent with previous evidence suggesting inpatient acuity has increased, with some caveats. The case-mix index for inpatients increased in Medicare claims,¹¹ but some of this increase may reflect expanded opportunities for hospitals to document comorbidities.¹² Geographic areas with larger decreases in inpatient admissions between 2010 and 2013 experienced greater risk-adjusted mortality among inpatients, consistent with a higher level of acuity among inpatients.¹³ The volume of ICU admissions declined, but ICU patients were more likely to have organ failure and to use complex services, such as mechanical ventilation, than patients admitted in earlier years.¹⁴

When interpreting these trends, several points are notable. The underlying health of the Medicare population may contribute to declining inpatient use but is difficult to quantify. The observed decline in cardiac-related hospitalizations is consistent with evidence that the impact of ischemic heart disease, the leading source of disease or injury in the US population, has dramatically declined over recent decades¹⁵ and that the Medicare program has experienced large declines in overall spending and use related to cardiac conditions.^16-18

Other potential factors include a shift toward hospitals treating Medicare beneficiaries as outpatients during an observation stay instead of admitting them as inpatients. Observation stays have increased as traditional Medicare implemented measures to penalize readmissions and limit payments for short inpatient stays.^19-21 Even so, the increase in observation stays would have to be at least three times as large as described in other work to fully substitute for the decrease in inpatient stays: between the years 2007 and 2018, the number of observation stays per 1000 beneficiaries increased by only 26 stays, whereas the number of hospitalizations per 1000 beneficiaries decreased by 83 hospitalizations.²⁰

Outpatient services may also broaden treatment availability in alternative settings or enable beneficiaries to avoid inpatient treatment with appropriate preventative care.^22-27 These considerations are even more relevant as the COVID-19 pandemic spurred reduced admissions and shifted acute services outside of hospitals.^28,29 Some services, such as elective surgeries, have probably shifted from an inpatient to an outpatient setting, which would be consistent with our finding that there are larger relative declines in planned hospitalizations. Although this analysis does not capture spending for outpatient services that are not linked to an inpatient admission, prior work demonstrates that annual growth in total Medicare spending per beneficiary (episode related or not) has recently declined for the inpatient sector but increased for outpatient and physician sectors.³⁰ By offering other outpatient services, hospitals may be able to recoup some declining inpatient revenues. However, outpatient services are reimbursed at a lower rate than inpatient services, suggesting these trends may create financial pressure for hospitals.

There are several limitations to our analysis. First, our analysis is not designed to uncover the reason for the shift away from inpatient services nor to analyze how it has affected beneficiaries’ overall quality of care. Second, in accounting for payment rate changes, we do not consider that facilities may have changed their behavior in response to payment rate changes. If the profitability of treating Medicare patients declined, then facilities may no longer have as much financial incentive to offer services that attract Medicare beneficiaries as inpatients. Third, our analysis excludes the Medicare Advantage population, which more than doubled over this time period and experienced smaller declines in inpatient use over our study years.^31,32 Fourth, our analysis does not include spending on services provided outside of inpatient episodes, so we do not estimate how much declines in episode-related spending contributed to overall Medicare spending. Finally, as with the trends noted for sepsis and pneumonia,⁹ some of the changes in diagnosis categories might reflect changes in coding practices to ensure that conditions with higher DRG payment rates are listed as the primary diagnosis, even if the actual services rendered or conditions treated did not change.

Over an 8-year period, Medicare spending per beneficiary on inpatient episodes, including all services immediately preceding and following hospitalizations, declined by 8.9% after taking into account payment rate increases. This broad shift away from inpatient services among all Medicare beneficiaries suggests policymakers should aim for payment policies that balance financial sustainability for hospitals and associated facilities with more efficient use of inpatient and related services.

The authors thank Sunita Thapa, Lucas Stewart, Christine Lai, and Liliana Podczerwinski for contributions in data analysis and manuscript preparation.

The rate of inpatient admissions among adults aged 65 years and older has decreased by approximately 25% since 2000.^1,2 This long-term trend raises important questions about inpatient-related spending in the traditional Medicare program for hospitals and providers who treat beneficiaries after a hospitalization. As traditional Medicare’s most expensive sector (accounting for 21% of all Medicare spending³), reducing hospitalizations is often championed as an opportunity to moderate Medicare spending growth.

Medicare’s ability to achieve significant savings from declining inpatient use may be tempered by a shift toward more expensive hospitalizations. If marginal hospitalizations among healthier beneficiaries are avoided, then the remaining inpatient users may be sicker and have greater spending per hospitalization and greater need for follow-up services. This study examines trends in Medicare spending related to episodes initiated by an inpatient stay because of its importance to overall Medicare spending and the implications for several Medicare value-based payment initiatives. In care models seeking to contain spending at a population level, such as accountable care organizations and managed care plans, reducing inpatient use and associated services may have the largest impact in curbing overall spending growth per beneficiary. Other models focused on spending at an episode level, including bundled payment initiatives, may face challenges if inpatient episodes become more expensive over time.

As Medicare shifts toward value-based payments, hospitalists and other hospital leaders are often involved in redesigning care delivery models for the hospital or accountable care organization (eg, through readmission reduction initiatives, post–acute care coordination, and bundled-care delivery programs). Not all savings strategies rely on providers to change how services are delivered; Medicare can modify payment rates, such as Affordable Care Act provisions that slowed how quickly Medicare payment rates increased.⁴ For clinicians to navigate the shift toward new payment models, it is important to recognize how each of these elements—declining hospital admissions, spending per inpatient episode, and payment rates—affect spending trends for inpatient services and associated care. Previous articles on overall Medicare inpatient spending have examined inpatient stays alone⁵ or focused mainly on spending per episode^6,7 without quantifying how these elements contributed to overall episode-related Medicare spending per beneficiary. This article addresses this gap by demonstrating how inpatient-related spending trends reflect each component.

This study examined trends in Medicare’s spending on inpatient episodes during the years 2009 to 2017. We described changes in the volume and spending on inpatient-initiated episodes across several dimensions, including beneficiary-level and hospitalization-level factors. We examined whether declines in spending associated with fewer inpatient-initiated episodes have been offset by increased spending per episode and how spending would have differed without changes in Medicare payment rates.

Episode Definition

We constructed an episode measure that captured traditional Medicare spending for 30 days prior to hospital admission, hospitalization duration, and 90 days following hospital discharge (additional details in the Appendix). As with bundled payments that include pre- and postdischarge services, this window allowed us to observe any services that preceded or followed a hospitalization. Using Medicare Part A, B, and D claims data for the years 2008 to 2018, we captured spending across all sectors for episodes with an index hospital admission in the years 2009 to 2017. If a beneficiary died during an episode, we measured healthcare spending until death. We excluded episodes where beneficiaries did not have traditional Medicare Part A and B for the entire episode or lived outside the 50 states or Washington, DC.

Any acute hospitalization triggered a new episode, with one exception: if a beneficiary was discharged and readmitted within 90 days for the same diagnosis-related group (DRG), then the readmission did not trigger a new episode. The spending for that readmission was attributed to the prior hospital stay. In effect, the annual number of episodes is equivalent to the annual number of hospital admissions minus subsequent rehospitalizations for the same DRG. Neither observation stays nor hospitalizations in inpatient rehabilitation, psychiatric, or long-term facilities were considered acute hospital admissions.

We assigned claims from noninpatient sectors to an episode based on whether the claim start date fell within the episode window. All traditional Medicare sectors were measured, including outpatient services, physician claims, post–acute care services, and Medicare Part D prescription drug events.

Our analysis aimed to measure all spending related to inpatient episodes without double-counting spending for overlapping episodes. If episodes overlapped, then spending for overlapping days was weighted to be evenly divided across episodes.

Outcome Measures

The study’s main outcomes summarized episode trends across the entire traditional Medicare population, including beneficiaries without an episode, in annual mean number of episodes per beneficiary and annual mean episode-related spending per beneficiary. The denominator of these measures is person-years, or total number of beneficiary months with Medicare Part A and B coverage divided by 12. The annual mean number of episodes per beneficiary is the total number of episodes initiated in a calendar year divided by person-years. The annual mean episode-related spending per beneficiary is the total amount of spending attributed to episodes divided by person-years. We also measured annual mean spending per episode, or total amount of spending attributed to episodes divided by the total number of episodes.

Medicare annually updates each sector’s payment rates for several factors, including inflation. We constructed an index for each sector to adjust for these annual payment rate changes. We also accounted for sequestration measures in effect since April 2013 that reduced Medicare payments to all sectors by 2%. We report our spending measures twice, with and without adjusting for changes in payment rates. Adjusted numbers reflect payment rates in effect in 2015.

Analysis Approach

We present annual trends on changes in the number of inpatient episodes per beneficiary, mean episode-related spending per beneficiary, and mean spending per episode. To quantify how changes in episode-related spending per beneficiary reflect changes in the number of episodes per beneficiary vs changes in spending per episode, we modified an approach implemented by Rosen and colleagues.⁸We calculated how much episode-related spending per beneficiary would have changed between 2009 and 2017 if spending per episode remained at 2009 levels but the number of episodes per beneficiary declined, as observed between 2009 and 2017 (see example calculation in Appendix). Conversely, we estimated how much episode-related spending per beneficiary would have changed if the number of episodes per beneficiary remained at 2009 levels but spending per episode increased, as observed between 2009 and 2017. In reality, the number of episodes per beneficiary and spending per episode concurrently changed, so the decomposition also includes an interaction term that quantifies how much of the change in spending reflects changes in both factors. We present these estimates for all sectors and separately for inpatient and all other sectors.

To better understand which beneficiaries have declining inpatient use, we performed stratified analyses describing changes in the number of episodes per beneficiary between 2009 and 2017, spending per episode, and total episode-related spending per beneficiary. We report these measures for several subpopulations defined by age, sex, race, dual-eligible status, and whether the beneficiary used long-term nursing home services during the episode’s calendar year. Descriptive statistics also detail how these measures changed between 2009 and 2017 for episodes stratified by characteristics of the index hospital stay: planned vs unplanned, medical vs surgical, and any use of intensive care unit (ICU) or coronary care unit services. We also stratify study measures by whether an episode included any use of post–acute care services (skilled nursing facility, home health, or inpatient rehabilitation facility use). Finally, we aggregate the episodes into major diagnostic categories (MDCs) based on the index hospital stay’s DRG to report study outcomes by condition. Because of a shift in coding hospitalizations for pneumonia as sepsis,^9,10 we exclude these two diseases from their respective MDCs and analyze them jointly as a unique category.

Changes in Number of Inpatient Episodes and Related Spending

From 2009 to 2017, the number of inpatient episodes per 1000 traditional Medicare beneficiaries declined from 326 to 267 (Table 1), or a relative decline of 18.2% (Figure 1). The total volume of inpatient episodes declined by only 13.4%, from 10.2 million to 8.8 million, reflecting that the size of the traditional Medicare population grew during these years. Over the same years, mean payment-rate–adjusted spending per episode increased 11.4% from $20,891 to $23,273.

When considering overall episode-related spending, the large decline in the volume of episodes outweighed increased spending per episode: the mean amount of episode-related Medicare spending per beneficiary decreased 8.9% from $6810 to $6206 (Table 1), or a net change of $604 (Figure 2). This net change reflects decreased spending due to fewer episodes per beneficiary ($1239 reduction in episode-related spending) offset by increased spending per episode (translating to a $776 increase in episode-related spending per beneficiary). These two factors, plus their interaction reflecting the combined influence of these factors ($141), comprise the overall change in episode-related spending per beneficiary over this period.

When these estimates are calculated separately for the inpatient sector and all other sectors, the inpatient sector experienced small increases in spending associated with greater spending per episode ($304) compared with noninpatient sectors ($472). Accordingly, the inpatient sector had a larger net decline in episode-related spending per beneficiary ($420) than noninpatient sectors ($184) after taking into account declining episode volume.

As expected, episode-related spending increased more when measures were not adjusted for annual payment rate increases. Without such adjustment, mean spending per episode increased 25.5%, and episode-related spending per beneficiary was nearly flat (2.6% between 2009 and 2017 [Figure 1]). The decline in unadjusted spending associated with fewer episodes ($1138) was offset by the spending increase associated with higher spending per episode ($1592) (Figure 2).

Analyses Stratified by Beneficiary Characteristics

Every population examined had declines in the number of inpatient episodes, even beneficiaries with more frequent inpatient use (Table 2). Among Medicare beneficiaries aged 85 years and older, the mean number of episodes per 1000 beneficiaries declined by 12.7%, from 524 to 457. Populations with less frequent inpatient use often experienced larger relative declines in number of episodes than populations with more frequent inpatient use. For example, the mean number of episodes per 1000 beneficiaries decreased by 17.7% for beneficiaries without nursing home use (306 to 252), as compared with an 8.1% decline for beneficiaries with nursing home use (from 888 to 816). In contrast, populations with less frequent inpatient use had larger relative increases in spending per episode with adjustment for payment rate changes. For example, spending per episode increased by 13.1% for beneficiaries aged 65 to 74 years ($20,904 to $23,644), but only by 8.6% for beneficiaries 85 years and older ($20,384 to $22,138).

Analyses Stratified by Service Use Characteristics

Some types of inpatient episodes had larger declines in the number of episodes, including episodes with planned admissions for the index hospital stay (28.8% decline from 68 to 48 episodes per 1000 beneficiaries) and episodes without post–acute care use (23.9% decline from 169 to 129 episodes per 1000 beneficiaries) (Appendix Table). In contrast, declines in the number of episodes were similar for index hospital admissions that did or did not involve ICU use (17.8% and 18.3% reduction in mean number of episodes per 1000 beneficiaries, respectively) or that included a surgical procedure or not (17.1% versus 18.6%, respectively). Several types of inpatient episodes had larger increases in spending per episode, such as a 15.1% increase for planned admissions and a 13.2% increase for hospitalizations without ICU use.

According to diagnosis information for an episode’s index hospital stay, inpatient episodes related to conditions affecting the circulatory system had the largest decline in mean number of episodes, decreasing by 31.8% from 78 to 53 episodes per 1000 beneficiaries (Appendix Table). Episodes for other diseases had much smaller declines in volume. Admissions for diagnoses of pneumonia or sepsis had notable increases in the volume of episodes, increasing by 20.7% from 25 to 30 admissions per 1000 beneficiaries.

Medicare spending per beneficiary on inpatient episodes, including services provided pre- and post hospitalization, declined by 8.9% from 2009 to 2017 after adjusting for payment rate changes. This decline reflects two components. First, the number of episodes per 1000 beneficiaries declined by 18.2%. Although the extent of this decrease varied across populations, every group examined had declines in inpatient use. In particular, hospitalizations for conditions affecting the circulatory system, such as heart attacks and cardiac procedures, decreased. Second, as inpatient volume declined, spending per episode increased by 11.4% to an average of $23,273 in 2017. This increase in spending per episode offset how much overall Medicare spending on episode-related care declined.

Medicare is increasingly challenging hospitals to demonstrate the value of inpatient services and associated treatment, which requires hospital leaders to recognize how their facilities’ spending trends relate to these national patterns. Understanding how much national episode-related spending has decreased over time with declining inpatient volume can help an accountable care organization evaluate whether it is feasible to achieve significant savings by reducing hospitalizations. Bundled payment providers focused on managing spending per episode can benefit from identifying which types of hospitalizations have increased spending per episode, especially for certain diagnoses.

These results also highlight the continued importance of a perennial factor in Medicare spending: payment rates. If Medicare payment rates had not increased over our study period, Medicare spending per inpatient episode would have increased by only 11%. Actual Medicare spending per episode increased by 25%, demonstrating that over half of the relative increase in spending per episode reflected increases in Medicare’s payment rates.

Increased spending per episode, even after adjustment for payment rate changes, suggests that services provided during an episode have increased in intensity or shifted toward higher-cost treatments. In order to understand how Medicare episode-related spending changed without introducing assumptions about factors underlying that change, our analysis did not adjust for inpatient acuity. We observed a smaller decline in the volume of hospitalizations with post–acute care use but similar decreases in the volume of hospitalizations with and without ICU use. This finding is consistent with previous evidence suggesting inpatient acuity has increased, with some caveats. The case-mix index for inpatients increased in Medicare claims,¹¹ but some of this increase may reflect expanded opportunities for hospitals to document comorbidities.¹² Geographic areas with larger decreases in inpatient admissions between 2010 and 2013 experienced greater risk-adjusted mortality among inpatients, consistent with a higher level of acuity among inpatients.¹³ The volume of ICU admissions declined, but ICU patients were more likely to have organ failure and to use complex services, such as mechanical ventilation, than patients admitted in earlier years.¹⁴

When interpreting these trends, several points are notable. The underlying health of the Medicare population may contribute to declining inpatient use but is difficult to quantify. The observed decline in cardiac-related hospitalizations is consistent with evidence that the impact of ischemic heart disease, the leading source of disease or injury in the US population, has dramatically declined over recent decades¹⁵ and that the Medicare program has experienced large declines in overall spending and use related to cardiac conditions.^16-18

Other potential factors include a shift toward hospitals treating Medicare beneficiaries as outpatients during an observation stay instead of admitting them as inpatients. Observation stays have increased as traditional Medicare implemented measures to penalize readmissions and limit payments for short inpatient stays.^19-21 Even so, the increase in observation stays would have to be at least three times as large as described in other work to fully substitute for the decrease in inpatient stays: between the years 2007 and 2018, the number of observation stays per 1000 beneficiaries increased by only 26 stays, whereas the number of hospitalizations per 1000 beneficiaries decreased by 83 hospitalizations.²⁰

Outpatient services may also broaden treatment availability in alternative settings or enable beneficiaries to avoid inpatient treatment with appropriate preventative care.^22-27 These considerations are even more relevant as the COVID-19 pandemic spurred reduced admissions and shifted acute services outside of hospitals.^28,29 Some services, such as elective surgeries, have probably shifted from an inpatient to an outpatient setting, which would be consistent with our finding that there are larger relative declines in planned hospitalizations. Although this analysis does not capture spending for outpatient services that are not linked to an inpatient admission, prior work demonstrates that annual growth in total Medicare spending per beneficiary (episode related or not) has recently declined for the inpatient sector but increased for outpatient and physician sectors.³⁰ By offering other outpatient services, hospitals may be able to recoup some declining inpatient revenues. However, outpatient services are reimbursed at a lower rate than inpatient services, suggesting these trends may create financial pressure for hospitals.

There are several limitations to our analysis. First, our analysis is not designed to uncover the reason for the shift away from inpatient services nor to analyze how it has affected beneficiaries’ overall quality of care. Second, in accounting for payment rate changes, we do not consider that facilities may have changed their behavior in response to payment rate changes. If the profitability of treating Medicare patients declined, then facilities may no longer have as much financial incentive to offer services that attract Medicare beneficiaries as inpatients. Third, our analysis excludes the Medicare Advantage population, which more than doubled over this time period and experienced smaller declines in inpatient use over our study years.^31,32 Fourth, our analysis does not include spending on services provided outside of inpatient episodes, so we do not estimate how much declines in episode-related spending contributed to overall Medicare spending. Finally, as with the trends noted for sepsis and pneumonia,⁹ some of the changes in diagnosis categories might reflect changes in coding practices to ensure that conditions with higher DRG payment rates are listed as the primary diagnosis, even if the actual services rendered or conditions treated did not change.

Over an 8-year period, Medicare spending per beneficiary on inpatient episodes, including all services immediately preceding and following hospitalizations, declined by 8.9% after taking into account payment rate increases. This broad shift away from inpatient services among all Medicare beneficiaries suggests policymakers should aim for payment policies that balance financial sustainability for hospitals and associated facilities with more efficient use of inpatient and related services.

The authors thank Sunita Thapa, Lucas Stewart, Christine Lai, and Liliana Podczerwinski for contributions in data analysis and manuscript preparation.

The rate of inpatient admissions among adults aged 65 years and older has decreased by approximately 25% since 2000.^1,2 This long-term trend raises important questions about inpatient-related spending in the traditional Medicare program for hospitals and providers who treat beneficiaries after a hospitalization. As traditional Medicare’s most expensive sector (accounting for 21% of all Medicare spending³), reducing hospitalizations is often championed as an opportunity to moderate Medicare spending growth.

Medicare’s ability to achieve significant savings from declining inpatient use may be tempered by a shift toward more expensive hospitalizations. If marginal hospitalizations among healthier beneficiaries are avoided, then the remaining inpatient users may be sicker and have greater spending per hospitalization and greater need for follow-up services. This study examines trends in Medicare spending related to episodes initiated by an inpatient stay because of its importance to overall Medicare spending and the implications for several Medicare value-based payment initiatives. In care models seeking to contain spending at a population level, such as accountable care organizations and managed care plans, reducing inpatient use and associated services may have the largest impact in curbing overall spending growth per beneficiary. Other models focused on spending at an episode level, including bundled payment initiatives, may face challenges if inpatient episodes become more expensive over time.

As Medicare shifts toward value-based payments, hospitalists and other hospital leaders are often involved in redesigning care delivery models for the hospital or accountable care organization (eg, through readmission reduction initiatives, post–acute care coordination, and bundled-care delivery programs). Not all savings strategies rely on providers to change how services are delivered; Medicare can modify payment rates, such as Affordable Care Act provisions that slowed how quickly Medicare payment rates increased.⁴ For clinicians to navigate the shift toward new payment models, it is important to recognize how each of these elements—declining hospital admissions, spending per inpatient episode, and payment rates—affect spending trends for inpatient services and associated care. Previous articles on overall Medicare inpatient spending have examined inpatient stays alone⁵ or focused mainly on spending per episode^6,7 without quantifying how these elements contributed to overall episode-related Medicare spending per beneficiary. This article addresses this gap by demonstrating how inpatient-related spending trends reflect each component.

This study examined trends in Medicare’s spending on inpatient episodes during the years 2009 to 2017. We described changes in the volume and spending on inpatient-initiated episodes across several dimensions, including beneficiary-level and hospitalization-level factors. We examined whether declines in spending associated with fewer inpatient-initiated episodes have been offset by increased spending per episode and how spending would have differed without changes in Medicare payment rates.

Episode Definition

We constructed an episode measure that captured traditional Medicare spending for 30 days prior to hospital admission, hospitalization duration, and 90 days following hospital discharge (additional details in the Appendix). As with bundled payments that include pre- and postdischarge services, this window allowed us to observe any services that preceded or followed a hospitalization. Using Medicare Part A, B, and D claims data for the years 2008 to 2018, we captured spending across all sectors for episodes with an index hospital admission in the years 2009 to 2017. If a beneficiary died during an episode, we measured healthcare spending until death. We excluded episodes where beneficiaries did not have traditional Medicare Part A and B for the entire episode or lived outside the 50 states or Washington, DC.

Any acute hospitalization triggered a new episode, with one exception: if a beneficiary was discharged and readmitted within 90 days for the same diagnosis-related group (DRG), then the readmission did not trigger a new episode. The spending for that readmission was attributed to the prior hospital stay. In effect, the annual number of episodes is equivalent to the annual number of hospital admissions minus subsequent rehospitalizations for the same DRG. Neither observation stays nor hospitalizations in inpatient rehabilitation, psychiatric, or long-term facilities were considered acute hospital admissions.

We assigned claims from noninpatient sectors to an episode based on whether the claim start date fell within the episode window. All traditional Medicare sectors were measured, including outpatient services, physician claims, post–acute care services, and Medicare Part D prescription drug events.

Our analysis aimed to measure all spending related to inpatient episodes without double-counting spending for overlapping episodes. If episodes overlapped, then spending for overlapping days was weighted to be evenly divided across episodes.

Outcome Measures

The study’s main outcomes summarized episode trends across the entire traditional Medicare population, including beneficiaries without an episode, in annual mean number of episodes per beneficiary and annual mean episode-related spending per beneficiary. The denominator of these measures is person-years, or total number of beneficiary months with Medicare Part A and B coverage divided by 12. The annual mean number of episodes per beneficiary is the total number of episodes initiated in a calendar year divided by person-years. The annual mean episode-related spending per beneficiary is the total amount of spending attributed to episodes divided by person-years. We also measured annual mean spending per episode, or total amount of spending attributed to episodes divided by the total number of episodes.

Medicare annually updates each sector’s payment rates for several factors, including inflation. We constructed an index for each sector to adjust for these annual payment rate changes. We also accounted for sequestration measures in effect since April 2013 that reduced Medicare payments to all sectors by 2%. We report our spending measures twice, with and without adjusting for changes in payment rates. Adjusted numbers reflect payment rates in effect in 2015.

Analysis Approach

We present annual trends on changes in the number of inpatient episodes per beneficiary, mean episode-related spending per beneficiary, and mean spending per episode. To quantify how changes in episode-related spending per beneficiary reflect changes in the number of episodes per beneficiary vs changes in spending per episode, we modified an approach implemented by Rosen and colleagues.⁸We calculated how much episode-related spending per beneficiary would have changed between 2009 and 2017 if spending per episode remained at 2009 levels but the number of episodes per beneficiary declined, as observed between 2009 and 2017 (see example calculation in Appendix). Conversely, we estimated how much episode-related spending per beneficiary would have changed if the number of episodes per beneficiary remained at 2009 levels but spending per episode increased, as observed between 2009 and 2017. In reality, the number of episodes per beneficiary and spending per episode concurrently changed, so the decomposition also includes an interaction term that quantifies how much of the change in spending reflects changes in both factors. We present these estimates for all sectors and separately for inpatient and all other sectors.

To better understand which beneficiaries have declining inpatient use, we performed stratified analyses describing changes in the number of episodes per beneficiary between 2009 and 2017, spending per episode, and total episode-related spending per beneficiary. We report these measures for several subpopulations defined by age, sex, race, dual-eligible status, and whether the beneficiary used long-term nursing home services during the episode’s calendar year. Descriptive statistics also detail how these measures changed between 2009 and 2017 for episodes stratified by characteristics of the index hospital stay: planned vs unplanned, medical vs surgical, and any use of intensive care unit (ICU) or coronary care unit services. We also stratify study measures by whether an episode included any use of post–acute care services (skilled nursing facility, home health, or inpatient rehabilitation facility use). Finally, we aggregate the episodes into major diagnostic categories (MDCs) based on the index hospital stay’s DRG to report study outcomes by condition. Because of a shift in coding hospitalizations for pneumonia as sepsis,^9,10 we exclude these two diseases from their respective MDCs and analyze them jointly as a unique category.

Changes in Number of Inpatient Episodes and Related Spending

From 2009 to 2017, the number of inpatient episodes per 1000 traditional Medicare beneficiaries declined from 326 to 267 (Table 1), or a relative decline of 18.2% (Figure 1). The total volume of inpatient episodes declined by only 13.4%, from 10.2 million to 8.8 million, reflecting that the size of the traditional Medicare population grew during these years. Over the same years, mean payment-rate–adjusted spending per episode increased 11.4% from $20,891 to $23,273.

When considering overall episode-related spending, the large decline in the volume of episodes outweighed increased spending per episode: the mean amount of episode-related Medicare spending per beneficiary decreased 8.9% from $6810 to $6206 (Table 1), or a net change of $604 (Figure 2). This net change reflects decreased spending due to fewer episodes per beneficiary ($1239 reduction in episode-related spending) offset by increased spending per episode (translating to a $776 increase in episode-related spending per beneficiary). These two factors, plus their interaction reflecting the combined influence of these factors ($141), comprise the overall change in episode-related spending per beneficiary over this period.

When these estimates are calculated separately for the inpatient sector and all other sectors, the inpatient sector experienced small increases in spending associated with greater spending per episode ($304) compared with noninpatient sectors ($472). Accordingly, the inpatient sector had a larger net decline in episode-related spending per beneficiary ($420) than noninpatient sectors ($184) after taking into account declining episode volume.

As expected, episode-related spending increased more when measures were not adjusted for annual payment rate increases. Without such adjustment, mean spending per episode increased 25.5%, and episode-related spending per beneficiary was nearly flat (2.6% between 2009 and 2017 [Figure 1]). The decline in unadjusted spending associated with fewer episodes ($1138) was offset by the spending increase associated with higher spending per episode ($1592) (Figure 2).

Analyses Stratified by Beneficiary Characteristics

Every population examined had declines in the number of inpatient episodes, even beneficiaries with more frequent inpatient use (Table 2). Among Medicare beneficiaries aged 85 years and older, the mean number of episodes per 1000 beneficiaries declined by 12.7%, from 524 to 457. Populations with less frequent inpatient use often experienced larger relative declines in number of episodes than populations with more frequent inpatient use. For example, the mean number of episodes per 1000 beneficiaries decreased by 17.7% for beneficiaries without nursing home use (306 to 252), as compared with an 8.1% decline for beneficiaries with nursing home use (from 888 to 816). In contrast, populations with less frequent inpatient use had larger relative increases in spending per episode with adjustment for payment rate changes. For example, spending per episode increased by 13.1% for beneficiaries aged 65 to 74 years ($20,904 to $23,644), but only by 8.6% for beneficiaries 85 years and older ($20,384 to $22,138).

Analyses Stratified by Service Use Characteristics

Some types of inpatient episodes had larger declines in the number of episodes, including episodes with planned admissions for the index hospital stay (28.8% decline from 68 to 48 episodes per 1000 beneficiaries) and episodes without post–acute care use (23.9% decline from 169 to 129 episodes per 1000 beneficiaries) (Appendix Table). In contrast, declines in the number of episodes were similar for index hospital admissions that did or did not involve ICU use (17.8% and 18.3% reduction in mean number of episodes per 1000 beneficiaries, respectively) or that included a surgical procedure or not (17.1% versus 18.6%, respectively). Several types of inpatient episodes had larger increases in spending per episode, such as a 15.1% increase for planned admissions and a 13.2% increase for hospitalizations without ICU use.

According to diagnosis information for an episode’s index hospital stay, inpatient episodes related to conditions affecting the circulatory system had the largest decline in mean number of episodes, decreasing by 31.8% from 78 to 53 episodes per 1000 beneficiaries (Appendix Table). Episodes for other diseases had much smaller declines in volume. Admissions for diagnoses of pneumonia or sepsis had notable increases in the volume of episodes, increasing by 20.7% from 25 to 30 admissions per 1000 beneficiaries.

Medicare spending per beneficiary on inpatient episodes, including services provided pre- and post hospitalization, declined by 8.9% from 2009 to 2017 after adjusting for payment rate changes. This decline reflects two components. First, the number of episodes per 1000 beneficiaries declined by 18.2%. Although the extent of this decrease varied across populations, every group examined had declines in inpatient use. In particular, hospitalizations for conditions affecting the circulatory system, such as heart attacks and cardiac procedures, decreased. Second, as inpatient volume declined, spending per episode increased by 11.4% to an average of $23,273 in 2017. This increase in spending per episode offset how much overall Medicare spending on episode-related care declined.

Medicare is increasingly challenging hospitals to demonstrate the value of inpatient services and associated treatment, which requires hospital leaders to recognize how their facilities’ spending trends relate to these national patterns. Understanding how much national episode-related spending has decreased over time with declining inpatient volume can help an accountable care organization evaluate whether it is feasible to achieve significant savings by reducing hospitalizations. Bundled payment providers focused on managing spending per episode can benefit from identifying which types of hospitalizations have increased spending per episode, especially for certain diagnoses.

These results also highlight the continued importance of a perennial factor in Medicare spending: payment rates. If Medicare payment rates had not increased over our study period, Medicare spending per inpatient episode would have increased by only 11%. Actual Medicare spending per episode increased by 25%, demonstrating that over half of the relative increase in spending per episode reflected increases in Medicare’s payment rates.

Increased spending per episode, even after adjustment for payment rate changes, suggests that services provided during an episode have increased in intensity or shifted toward higher-cost treatments. In order to understand how Medicare episode-related spending changed without introducing assumptions about factors underlying that change, our analysis did not adjust for inpatient acuity. We observed a smaller decline in the volume of hospitalizations with post–acute care use but similar decreases in the volume of hospitalizations with and without ICU use. This finding is consistent with previous evidence suggesting inpatient acuity has increased, with some caveats. The case-mix index for inpatients increased in Medicare claims,¹¹ but some of this increase may reflect expanded opportunities for hospitals to document comorbidities.¹² Geographic areas with larger decreases in inpatient admissions between 2010 and 2013 experienced greater risk-adjusted mortality among inpatients, consistent with a higher level of acuity among inpatients.¹³ The volume of ICU admissions declined, but ICU patients were more likely to have organ failure and to use complex services, such as mechanical ventilation, than patients admitted in earlier years.¹⁴

When interpreting these trends, several points are notable. The underlying health of the Medicare population may contribute to declining inpatient use but is difficult to quantify. The observed decline in cardiac-related hospitalizations is consistent with evidence that the impact of ischemic heart disease, the leading source of disease or injury in the US population, has dramatically declined over recent decades¹⁵ and that the Medicare program has experienced large declines in overall spending and use related to cardiac conditions.^16-18

Other potential factors include a shift toward hospitals treating Medicare beneficiaries as outpatients during an observation stay instead of admitting them as inpatients. Observation stays have increased as traditional Medicare implemented measures to penalize readmissions and limit payments for short inpatient stays.^19-21 Even so, the increase in observation stays would have to be at least three times as large as described in other work to fully substitute for the decrease in inpatient stays: between the years 2007 and 2018, the number of observation stays per 1000 beneficiaries increased by only 26 stays, whereas the number of hospitalizations per 1000 beneficiaries decreased by 83 hospitalizations.²⁰

Outpatient services may also broaden treatment availability in alternative settings or enable beneficiaries to avoid inpatient treatment with appropriate preventative care.^22-27 These considerations are even more relevant as the COVID-19 pandemic spurred reduced admissions and shifted acute services outside of hospitals.^28,29 Some services, such as elective surgeries, have probably shifted from an inpatient to an outpatient setting, which would be consistent with our finding that there are larger relative declines in planned hospitalizations. Although this analysis does not capture spending for outpatient services that are not linked to an inpatient admission, prior work demonstrates that annual growth in total Medicare spending per beneficiary (episode related or not) has recently declined for the inpatient sector but increased for outpatient and physician sectors.³⁰ By offering other outpatient services, hospitals may be able to recoup some declining inpatient revenues. However, outpatient services are reimbursed at a lower rate than inpatient services, suggesting these trends may create financial pressure for hospitals.

There are several limitations to our analysis. First, our analysis is not designed to uncover the reason for the shift away from inpatient services nor to analyze how it has affected beneficiaries’ overall quality of care. Second, in accounting for payment rate changes, we do not consider that facilities may have changed their behavior in response to payment rate changes. If the profitability of treating Medicare patients declined, then facilities may no longer have as much financial incentive to offer services that attract Medicare beneficiaries as inpatients. Third, our analysis excludes the Medicare Advantage population, which more than doubled over this time period and experienced smaller declines in inpatient use over our study years.^31,32 Fourth, our analysis does not include spending on services provided outside of inpatient episodes, so we do not estimate how much declines in episode-related spending contributed to overall Medicare spending. Finally, as with the trends noted for sepsis and pneumonia,⁹ some of the changes in diagnosis categories might reflect changes in coding practices to ensure that conditions with higher DRG payment rates are listed as the primary diagnosis, even if the actual services rendered or conditions treated did not change.

Over an 8-year period, Medicare spending per beneficiary on inpatient episodes, including all services immediately preceding and following hospitalizations, declined by 8.9% after taking into account payment rate increases. This broad shift away from inpatient services among all Medicare beneficiaries suggests policymakers should aim for payment policies that balance financial sustainability for hospitals and associated facilities with more efficient use of inpatient and related services.

The authors thank Sunita Thapa, Lucas Stewart, Christine Lai, and Liliana Podczerwinski for contributions in data analysis and manuscript preparation.