Hospital Medicine

Siloed, adj.:

Kept in isolation in a way that hinders communication and cooperation . . .

—Merriam-Webster’s Dictionary

Humans naturally separate into groups, and the medical field is no exception. Being a member of a likeminded group, such as one’s specialty, can improve self-esteem and provide social organization: it feels good to identify with people we admire. Through culture, these specialty-based groups implicitly and explicitly guide and encourage positive attributes or behaviors like a hospitalist’s thoroughness or an emergency medicine physician’s steady management of unstable patients. Our specialties also provide support and understanding in challenging times. 

Despite these positive aspects, such divisions can negatively affect interprofessional relationships when our specialties become siloed. A potential side-effect of building up ourselves and our own groups is that we can implicitly put others down. For example, a hospitalist who spends extra time on the phone regularly updating each patient’s family will appropriately take pride in their practice, but over time this can also lead to an unreasonable assumption that physicians in other departments with different routines are not as committed to outstanding communication.

These rigid separations facilitate the fundamental attribution error, the tendency to ascribe a problem or disagreement to a colleague’s substandard character or ability. Imagine that the aforementioned hospitalist’s phone call delays a response to an admission page from the emergency room. The emergency medicine physician, who is waiting to sign out the admission while simultaneously managing many sick and complex patients, could assume the hospitalist is being disrespectful, rather than also working hard to provide the best care. Our siloed specialty identities can lead us to imagine the worst in each other and exacerbate intergroup conflict.¹

Silos in medicine also adversely affect patients. Poor communication and lack of information-sharing across disciplines can lead to medical error² and stifle dissemination of safer practices.³ Further, the unintentional disparaging of other medical specialties undermines the confidence our patients have in all of us; a patient within earshot of the hospitalist expressing annoyance at the “impatient” emergency medicine physician who “won’t stop paging,” or the emergency medicine physician complaining about the hospitalist who “refuses to call back,” will lose trust in each of their providers. 

We suggest three steps to reduce the negative impact of specialty silos in medicine: 

1. Get to know each other personally. Friendly conversation during work hours and social interaction outside the hospital can inoculate against interspecialty conflict by putting a human face on our colleagues. The resultant relationships make it easier to work together and see things from another’s perspective. 
2. Emphasize our shared affiliations.⁴ The greater the salience of a mutual identity as “healthcare providers,” the more likely we are to recognize each other’s unique contributions and question the stereotypes we imagine about one another. 
3. Consider projects across specialties. Interdepartmental data-sharing and joint meetings, including educational conferences, can facilitate situational awareness, synergy, and efficient problem-solving. 

Our medical specialties will continue to group together. While these groups can be a source of strength and meaning, silos can interfere with professional alliances and effective patient care. Mitigating the harmful effects of silos can benefit all of us and our patients.

Authors’ note: This article was previously published using the term “tribalism,” which we have since learned is derogatory to Indigenous Americans and others. We apologize for any harm. We have retracted and republished the article without this language. We appreciate readers teaching us how to choose better words so all people feel respected and valued.

Siloed, adj.:

Kept in isolation in a way that hinders communication and cooperation . . .

—Merriam-Webster’s Dictionary

Humans naturally separate into groups, and the medical field is no exception. Being a member of a likeminded group, such as one’s specialty, can improve self-esteem and provide social organization: it feels good to identify with people we admire. Through culture, these specialty-based groups implicitly and explicitly guide and encourage positive attributes or behaviors like a hospitalist’s thoroughness or an emergency medicine physician’s steady management of unstable patients. Our specialties also provide support and understanding in challenging times. 

Despite these positive aspects, such divisions can negatively affect interprofessional relationships when our specialties become siloed. A potential side-effect of building up ourselves and our own groups is that we can implicitly put others down. For example, a hospitalist who spends extra time on the phone regularly updating each patient’s family will appropriately take pride in their practice, but over time this can also lead to an unreasonable assumption that physicians in other departments with different routines are not as committed to outstanding communication.

These rigid separations facilitate the fundamental attribution error, the tendency to ascribe a problem or disagreement to a colleague’s substandard character or ability. Imagine that the aforementioned hospitalist’s phone call delays a response to an admission page from the emergency room. The emergency medicine physician, who is waiting to sign out the admission while simultaneously managing many sick and complex patients, could assume the hospitalist is being disrespectful, rather than also working hard to provide the best care. Our siloed specialty identities can lead us to imagine the worst in each other and exacerbate intergroup conflict.¹

Silos in medicine also adversely affect patients. Poor communication and lack of information-sharing across disciplines can lead to medical error² and stifle dissemination of safer practices.³ Further, the unintentional disparaging of other medical specialties undermines the confidence our patients have in all of us; a patient within earshot of the hospitalist expressing annoyance at the “impatient” emergency medicine physician who “won’t stop paging,” or the emergency medicine physician complaining about the hospitalist who “refuses to call back,” will lose trust in each of their providers. 

We suggest three steps to reduce the negative impact of specialty silos in medicine: 

1. Get to know each other personally. Friendly conversation during work hours and social interaction outside the hospital can inoculate against interspecialty conflict by putting a human face on our colleagues. The resultant relationships make it easier to work together and see things from another’s perspective. 
2. Emphasize our shared affiliations.⁴ The greater the salience of a mutual identity as “healthcare providers,” the more likely we are to recognize each other’s unique contributions and question the stereotypes we imagine about one another. 
3. Consider projects across specialties. Interdepartmental data-sharing and joint meetings, including educational conferences, can facilitate situational awareness, synergy, and efficient problem-solving. 

Our medical specialties will continue to group together. While these groups can be a source of strength and meaning, silos can interfere with professional alliances and effective patient care. Mitigating the harmful effects of silos can benefit all of us and our patients.

Authors’ note: This article was previously published using the term “tribalism,” which we have since learned is derogatory to Indigenous Americans and others. We apologize for any harm. We have retracted and republished the article without this language. We appreciate readers teaching us how to choose better words so all people feel respected and valued.

Siloed, adj.:

Kept in isolation in a way that hinders communication and cooperation . . .

—Merriam-Webster’s Dictionary

Humans naturally separate into groups, and the medical field is no exception. Being a member of a likeminded group, such as one’s specialty, can improve self-esteem and provide social organization: it feels good to identify with people we admire. Through culture, these specialty-based groups implicitly and explicitly guide and encourage positive attributes or behaviors like a hospitalist’s thoroughness or an emergency medicine physician’s steady management of unstable patients. Our specialties also provide support and understanding in challenging times. 

Despite these positive aspects, such divisions can negatively affect interprofessional relationships when our specialties become siloed. A potential side-effect of building up ourselves and our own groups is that we can implicitly put others down. For example, a hospitalist who spends extra time on the phone regularly updating each patient’s family will appropriately take pride in their practice, but over time this can also lead to an unreasonable assumption that physicians in other departments with different routines are not as committed to outstanding communication.

These rigid separations facilitate the fundamental attribution error, the tendency to ascribe a problem or disagreement to a colleague’s substandard character or ability. Imagine that the aforementioned hospitalist’s phone call delays a response to an admission page from the emergency room. The emergency medicine physician, who is waiting to sign out the admission while simultaneously managing many sick and complex patients, could assume the hospitalist is being disrespectful, rather than also working hard to provide the best care. Our siloed specialty identities can lead us to imagine the worst in each other and exacerbate intergroup conflict.¹

Silos in medicine also adversely affect patients. Poor communication and lack of information-sharing across disciplines can lead to medical error² and stifle dissemination of safer practices.³ Further, the unintentional disparaging of other medical specialties undermines the confidence our patients have in all of us; a patient within earshot of the hospitalist expressing annoyance at the “impatient” emergency medicine physician who “won’t stop paging,” or the emergency medicine physician complaining about the hospitalist who “refuses to call back,” will lose trust in each of their providers. 

We suggest three steps to reduce the negative impact of specialty silos in medicine: 

1. Get to know each other personally. Friendly conversation during work hours and social interaction outside the hospital can inoculate against interspecialty conflict by putting a human face on our colleagues. The resultant relationships make it easier to work together and see things from another’s perspective. 
2. Emphasize our shared affiliations.⁴ The greater the salience of a mutual identity as “healthcare providers,” the more likely we are to recognize each other’s unique contributions and question the stereotypes we imagine about one another. 
3. Consider projects across specialties. Interdepartmental data-sharing and joint meetings, including educational conferences, can facilitate situational awareness, synergy, and efficient problem-solving. 

Our medical specialties will continue to group together. While these groups can be a source of strength and meaning, silos can interfere with professional alliances and effective patient care. Mitigating the harmful effects of silos can benefit all of us and our patients.

Authors’ note: This article was previously published using the term “tribalism,” which we have since learned is derogatory to Indigenous Americans and others. We apologize for any harm. We have retracted and republished the article without this language. We appreciate readers teaching us how to choose better words so all people feel respected and valued.

There is substantial interest in improving patient flow and reducing hospital length of stay (LOS).^1-4 Impaired hospital flow may negatively impact both patient satisfaction and safety through, for example, emergency department (ED) overcrowding.^5,6 Impaired hospital flow is associated with downstream effects on patient care, hospital costs, and availability of beds.^7-9

A number of quality-improvement interventions aim to improve patient flow, including efforts to increase the number of discharges that occur before noon.^10,11 Morning discharges have been hypothesized to free hospital beds earlier, thus reducing ED wait times for incoming patients and increasing beds for elective surgeries.¹¹ Morning discharges may also be more predictable for staff and patients. However, it is unclear whether efforts to increase the number of morning discharges have a negative impact on inpatient LOS by incentivizing physicians to keep patients in the hospital for an extra night to facilitate discharge in the early morning rather than the late afternoon. Morning discharges have been associated with both increased¹² and decreased LOS.^10,11,13-15

The purpose of this study was to examine the associations between morning discharges and ED LOS and hospital LOS in general internal medicine (GIM) at seven hospitals. GIM patients represent nearly 40% of ED admissions to a hospital,¹⁶ and thus are an important determinant of patient flow through the ED and hospital. We hypothesized that patients who were admitted to GIM on days with more morning discharges would have shorter ED LOS and hospital LOS.

Design, Setting, and Participants

This was a retrospective cohort study conducted using the General Medicine Inpatient Initiative (GEMINI) clinical dataset.¹⁶ The dataset includes all GIM admissions at seven large hospital sites in Toronto and Mississauga, Ontario, Canada. These include five academic hospitals and two community-based teaching hospitals. Each hospital is publicly funded and provides tertiary and/or quaternary care to diverse multiethnic populations. Research ethics board approval was obtained from all participating sites.

GIM care is delivered by several interdisciplinary clinical teams functioning in parallel. Attending physicians are predominantly internists who practice as hospitalists in discrete service blocks, typically lasting 2 weeks at a time. Although GIM patients are preferentially admitted to GIM wards, participating hospitals did not have strict policies regarding cohorting GIM patients to specific wards (ie, holding patients in ED until a specific bed becomes available) that would confound the association between morning discharge and ED wait times. Approximately 75% of GIM patients are cared for on dedicated GIM wards at participating hospitals, with the remainder cared for on other medical or surgical wards.

We included all hospitalized patients who were admitted to hospital and discharged from GIM between April 1, 2010, and October 31, 2017, from the seven GEMINI hospitals. We included only patients admitted through the ED. As such, we did not include elective admissions or interfacility transfers who would not experience ED wait times. We excluded patients who were discharged without a provincial health insurance number (N = 2,169; 1.1% of total sample) because they could not be linked across visits to measure readmissions.

Data Source

The GEMINI dataset has been rigorously validated and previously described in detail.¹⁶ GEMINI collects both administrative health data reported to the Canadian Institute for Health Information (including data about patient demographics, comorbidities, and discharge destination) as well as electronic clinical data extracted from hospital computer systems (including attending physicians, in-hospital patient room transfers, and laboratory test results). Data are collected for each individual hospital encounter, and the provincial health insurance number is used to link patients across encounters.

Exposures and Outcomes

The primary exposure was the number of GIM patients discharged alive between 8:00 am and 12:00 pm (ie, morning GIM discharges) on the day of admission for each hospital encounter. This time window to define morning discharges was selected based on previous literature.^10-13 To report admission characteristics and unadjusted outcomes, hospital days were categorized into quartiles within each hospital based on the number of morning GIM discharges. Hospital days with the lowest number of morning discharges were classified into Q1 and the highest number of morning discharges into Q4, and quartiles were pooled across hospitals.

The two primary outcomes were ED LOS and hospital LOS. ED LOS was calculated as the difference between the time from triage by nursing staff to a patient’s exit from the ED, measured in hours. We also examined 30-day readmission to GIM at any participating hospital as a balancing measure against premature discharges and inpatient mortality because it could modify hospital LOS.

Patient Characteristics

Baseline patient characteristics were measured, including age, sex, Charlson Comorbidity Index score,¹⁷ day of admission (categorized as weekend/holiday or weekday), time of admission to hospital (categorized as daytime, 8:00 am to 4:59 pm, or nighttime, 5:00 pm to 07:59 am), study month, hospital site, and whether patients had been admitted to GIM in the prior 30 days. We used laboratory data to calculate the baseline Laboratory-based Acute Physiology Score (LAPS), which is a validated predictor of inpatient mortality when combined with age and Charlson Comorbidity Index score.^18,19 GIM census on day of admission was calculated in order to include overall patient volumes as an important adjustment variable that could confound the association between morning discharges and patient flow.

Statistical Analysis

The study population and physician characteristics were summarized with descriptive statistics. The balance of baseline patient characteristics across morning discharge quartiles was assessed using standardized differences. A standardized difference of less than 0.1 reflects good balance.²⁰

Unadjusted estimates of patient outcomes were reported across morning discharge quartiles. To model the overall association between morning discharge and outcomes, the number of morning GIM discharges on the day of admission was subtracted from the mean number of morning discharges at each hospital and considered as a continuous exposure. We used generalized linear mixed models to estimate the effect of morning discharges on patient outcomes. We fit negative binomial regression models with log link to examine the association between the number of morning discharges (centered by subtracting the hospital mean) and the two main outcomes, ED LOS and hospital LOS. Given the overdispersion of the study population due to the unequal mean and variance, a negative binomial model was preferred over a Poisson regression, as the mean and variance were not equal.²¹ For our secondary outcomes of binary measures (30-day readmission and morality), we fit logistic regression models. Adjustment for multiple comparisons was not performed.

Multivariable analysis was conducted to adjust for the baseline characteristics described above as well as the total number of GIM discharges on the day of admission and GIM census on the day of admission. Hospital and study month (to account for secular time trends) were included as fixed effects, and patients and admitting physicians were included as crossed random effects to account for the nested structure of admissions within patients and admissions within physicians within hospitals.

A sensitivity analysis was performed to assess for nonlinear associations between morning discharges and the four outcomes (hospital LOS, ED LOS, in-hospital mortality, and readmission) by inputting the term as a restricted cubic spline, with up to five knots, in multivariable regression models. We compared the Akaike information criteria (AIC), computed using the log-likelihood, to determine the goodness-of-fit from the negative binomial models.²² We replicated models for each individual hospital to examine whether any hospital-specific associations existed. Two additional sensitivity analyses were performed to examine heterogeneity in hospital-specific effects. Regression models were fit, including interaction terms between morning discharge and hospital, as well as interaction terms between hospital and the total number of GIM discharges and GIM census. The findings of these models (data not presented) were qualitatively similar to the overall results.

Study Population and Patient Characteristics

The study population consisted of 189,781 hospitalizations involving 115,630 unique patients. The median patient age was 73 years (interquartile range [IQR], 57-84), 50.3% were female, 43.8% had a high Charlson Comorbidity Index score, and 11.1% were admitted to GIM in the prior 30 days (Table 1). The median ED LOS was 14.5 hours (IQR, 10.0-23.1), and the mean was 18.1 hours (SD, 12.2). The median hospital LOS was 4.6 days (IQR, 2.4-9.0), and the mean was 8.6 days (SD, 18.7).

In total, 36,043 (19.0%) discharges occurred between 8:00 am and 12:00 pm. The average number of total daily discharges per hospital was 8.4 (SD, 4.6), and the average number of morning discharges was 1.7 (SD, 1.4). Morning discharges varied across hospitals, ranging from 0.9 per day (SD, 1.1) to 2.1 (SD, 1.6) (Appendix Table 1). The average number of morning discharges in the lowest quartile (Q1) was 0.3 (SD, 0.5) and was 3.3 (SD, 1.2) in the highest quartile (Q4). Baseline patient characteristics were well balanced across the quartiles, with standardized differences less than 0.1 (Table 1), except day of admission and GIM census. Days with a greater number of morning discharges were more likely to be weekdays rather than weekends/holidays (89.2% weekday admissions in Q4 compared with 53.9% in Q1). The median GIM census was 93 patients (IQR, 79-109).

Outcomes

Unadjusted clinical outcomes by number of morning discharges are presented in Table 2. The median unadjusted ED LOS was 14.4 (SD, 14.1), 14.3 (SD, 13.2), 14.5 (SD, 13.0), and 14.8 (SD, 13.0) hours for the first to fourth quartiles (fewest to largest number of morning discharges), respectively. The median unadjusted hospital LOS was 4.6 (SD, 6.5), 4.6 (SD, 6.9), 4.7 (SD, 6.4), and 4.6 (SD, 6.4) days for the first to fourth quartiles, respectively.

Unadjusted inpatient mortality was 6.1%, 5.5%, 5.5%, and 5.2% across the first to fourth quartiles, respectively. Unadjusted 30-day readmission to GIM was 12.2%, 12.6%, 12.6%, and 12.5% across the first to fourth quartiles, respectively.

After multivariable adjustment, there was no significant association between morning discharge and hospital LOS (aRR, 1.000; 95% CI, 0.996-1.000; P = .997), ED LOS (aRR, 0.999; 95% CI, 0.997-1.000; P = .307), in-hospital mortality (aRR, 0.967; 95% CI, 0.920-1.020; P =.183), or 30-day readmission (aRR, 1.010; 95% CI, 0.991-1.020; P = .471) (Table 3, Appendix Table 2, Appendix Table 3, Appendix Table 4, Appendix Table 5). When examining each hospital separately, we found that morning discharge was significantly associated with hospital LOS at only one hospital (Hospital D; aRR, 0.981; 95% CI, 0.966-0.996; P = .013). Morning discharge was statistically significantly associated with ED LOS at three hospitals (A, B, and C), but the aRR was at least 0.99 in all three cases (Table 4).

In sensitivity analyses, we found no improvements in model fit when adding spline terms to the model, suggesting no significant nonlinear associations between morning discharges and the outcomes of interest.

DISCUSSION

This large multicenter cohort study found no significant overall association between the number of morning discharges and ED or hospital LOS in GIM. At one hospital, there was a 1.9% reduction in adjusted ED LOS for every additional morning discharge, but no difference in hospital LOS. We also did not observe differences in readmission or inpatient mortality associated with the number of morning discharges. Our observational findings suggest that there is unlikely to be a strong association between morning discharge and patient throughput in GIM. Given that there may be other downstream benefits of morning discharge, such as freeing beds for daytime surgeries,²³ further research is needed to determine the effectiveness of specific interventions.

Several studies have posited morning discharge as a method of improving both patient care and hospital flow metrics.^{10,11,13-15,23} Quality improvement initiatives targeting morning discharges have included stakeholder meetings, incentives programs, discharge-centered breakfast programs, and creating deadlines for discharge orders.^24-29 Although these initiatives have gained support, critics have suggested that their supporting evidence is not robust. Werthemier et al¹⁰ found a 9.0% reduction of observed to expected LOS associated with increasing the number of early discharges. However, a response article suggested that their findings were confounded by other hospital initiatives, such as allocation of medical and social services to weekends.³⁰ Other observational studies have concluded that hospital LOS is not affected by the number of morning discharges, but this research has been limited by single-center analysis and relatively smaller sample sizes.¹² Our study further calls into question the association between morning discharge and patient throughput.

An additional reason for the controversy is that physicians may actively work to discharge patients late in the day to avoid an additional night in hospital. A qualitative study by Minichiello et al³¹ evaluated staff perceptions regarding afternoon discharges. Physicians and medical students believed that afternoon discharges were a result of waiting for test results and procedures, with staff aiming to discharge patients immediately after obtaining results or finishing necessary procedures. As such, there are concerns that incentivizing morning discharge may lead physicians in the opposite direction, to consciously or unconsciously keep patients overnight in order to facilitate an early morning discharge.³⁰

Our study’s greatest strength was the large sample size over 7 years at seven hospitals in two cities, including both academic and community hospitals with different models of care. To our knowledge, this is the first cohort study that has analyzed the association between early discharge and LOS using multiple centers. To avoid the confounding and reverse causality that may exist when examining the relationship between LOS and morning discharge at the patient level (eg, patients who stay in hospital longer may have more “planned” discharges and leave in the morning), we examined the association based on variation across different days within the GIM service of each hospital. Further, we included robust risk adjustment using clinical and laboratory data. Finally, since our study included a diverse patient population served by participating centers in a system with universal insurance for hospital care, our findings are likely generalizable to other urban and suburban hospitals.

There are several important limitations of our analysis. First, we could only include GIM patients, who represent nearly 40% of ED admissions to hospital at participating centers. A more holistic analysis across all hospital services could be justified; however, given that many quality improvement initiatives occur at the level of a single hospital service, we felt our approach would be informative for future research and improvement efforts. Approximately 75% of GIM patients at participating hospitals were cared for on a GIM ward, with 25% cared for on off-service units. We were unable to include the total hospital census in our models, and this could affect LOS and waiting times for GIM patients, particularly those admitted to off-service units. GIM census is likely highly correlated with hospital census, and we were able to adjust for this. Nevertheless, this remains an important potential source of unmeasured confounding. Second, we did not model the effects of morning discharges from GIM on patient-flow measures for non-GIM patients. Given the lack of effects for GIM patients, who would be more likely to be directly affected, it is unlikely that large effects would be seen for other hospital patients, but we did not measure effects on surgical delays or cancellations, for example.²³ Third, we report 30-day readmission to GIM at participating hospitals only, rather than all readmissions. However, prior research in our region demonstrated that 82% of hospital readmissions occur to the same site.³² Thus, our measure, which includes admission to any participating hospital, likely captures more than 80% of all readmissions, and this was a secondary outcome in our analysis. Finally, qualitative metrics, such as patient or provider satisfaction, were not measured in our study. Earlier discharge may impact patient care in other ways by being more predictable for staff, improving bed allocation for daytime procedures, making medication pick-ups easier to arrange, or making consultations with allied health services more convenient.^11,28,33 Conversely, if pressured to discharge before noon, providers may feel rushed to complete tasks and may face disruptions to typical workflow.²⁴ As such, future research is needed to provide a more complete understanding of the impact of early-morning discharge beyond hospital flow. Given the lack of a strong association observed between morning discharge and patient throughput in our study, further research should also consider the opportunity costs of interventions designed to improve morning discharge.

CONCLUSION

The number of morning discharges was not significantly associated with shorter ED LOS or hospital LOS for GIM patients. Our observational findings suggest that increasing morning discharges alone may not substantially improve patient flow in GIM. Further research is needed to evaluate specific morning discharge interventions and assess hospital-wide effects.

There is substantial interest in improving patient flow and reducing hospital length of stay (LOS).^1-4 Impaired hospital flow may negatively impact both patient satisfaction and safety through, for example, emergency department (ED) overcrowding.^5,6 Impaired hospital flow is associated with downstream effects on patient care, hospital costs, and availability of beds.^7-9

A number of quality-improvement interventions aim to improve patient flow, including efforts to increase the number of discharges that occur before noon.^10,11 Morning discharges have been hypothesized to free hospital beds earlier, thus reducing ED wait times for incoming patients and increasing beds for elective surgeries.¹¹ Morning discharges may also be more predictable for staff and patients. However, it is unclear whether efforts to increase the number of morning discharges have a negative impact on inpatient LOS by incentivizing physicians to keep patients in the hospital for an extra night to facilitate discharge in the early morning rather than the late afternoon. Morning discharges have been associated with both increased¹² and decreased LOS.^10,11,13-15

The purpose of this study was to examine the associations between morning discharges and ED LOS and hospital LOS in general internal medicine (GIM) at seven hospitals. GIM patients represent nearly 40% of ED admissions to a hospital,¹⁶ and thus are an important determinant of patient flow through the ED and hospital. We hypothesized that patients who were admitted to GIM on days with more morning discharges would have shorter ED LOS and hospital LOS.

Design, Setting, and Participants

This was a retrospective cohort study conducted using the General Medicine Inpatient Initiative (GEMINI) clinical dataset.¹⁶ The dataset includes all GIM admissions at seven large hospital sites in Toronto and Mississauga, Ontario, Canada. These include five academic hospitals and two community-based teaching hospitals. Each hospital is publicly funded and provides tertiary and/or quaternary care to diverse multiethnic populations. Research ethics board approval was obtained from all participating sites.

GIM care is delivered by several interdisciplinary clinical teams functioning in parallel. Attending physicians are predominantly internists who practice as hospitalists in discrete service blocks, typically lasting 2 weeks at a time. Although GIM patients are preferentially admitted to GIM wards, participating hospitals did not have strict policies regarding cohorting GIM patients to specific wards (ie, holding patients in ED until a specific bed becomes available) that would confound the association between morning discharge and ED wait times. Approximately 75% of GIM patients are cared for on dedicated GIM wards at participating hospitals, with the remainder cared for on other medical or surgical wards.

We included all hospitalized patients who were admitted to hospital and discharged from GIM between April 1, 2010, and October 31, 2017, from the seven GEMINI hospitals. We included only patients admitted through the ED. As such, we did not include elective admissions or interfacility transfers who would not experience ED wait times. We excluded patients who were discharged without a provincial health insurance number (N = 2,169; 1.1% of total sample) because they could not be linked across visits to measure readmissions.

Data Source

The GEMINI dataset has been rigorously validated and previously described in detail.¹⁶ GEMINI collects both administrative health data reported to the Canadian Institute for Health Information (including data about patient demographics, comorbidities, and discharge destination) as well as electronic clinical data extracted from hospital computer systems (including attending physicians, in-hospital patient room transfers, and laboratory test results). Data are collected for each individual hospital encounter, and the provincial health insurance number is used to link patients across encounters.

Exposures and Outcomes

The primary exposure was the number of GIM patients discharged alive between 8:00 am and 12:00 pm (ie, morning GIM discharges) on the day of admission for each hospital encounter. This time window to define morning discharges was selected based on previous literature.^10-13 To report admission characteristics and unadjusted outcomes, hospital days were categorized into quartiles within each hospital based on the number of morning GIM discharges. Hospital days with the lowest number of morning discharges were classified into Q1 and the highest number of morning discharges into Q4, and quartiles were pooled across hospitals.

The two primary outcomes were ED LOS and hospital LOS. ED LOS was calculated as the difference between the time from triage by nursing staff to a patient’s exit from the ED, measured in hours. We also examined 30-day readmission to GIM at any participating hospital as a balancing measure against premature discharges and inpatient mortality because it could modify hospital LOS.

Patient Characteristics

Baseline patient characteristics were measured, including age, sex, Charlson Comorbidity Index score,¹⁷ day of admission (categorized as weekend/holiday or weekday), time of admission to hospital (categorized as daytime, 8:00 am to 4:59 pm, or nighttime, 5:00 pm to 07:59 am), study month, hospital site, and whether patients had been admitted to GIM in the prior 30 days. We used laboratory data to calculate the baseline Laboratory-based Acute Physiology Score (LAPS), which is a validated predictor of inpatient mortality when combined with age and Charlson Comorbidity Index score.^18,19 GIM census on day of admission was calculated in order to include overall patient volumes as an important adjustment variable that could confound the association between morning discharges and patient flow.

Statistical Analysis

The study population and physician characteristics were summarized with descriptive statistics. The balance of baseline patient characteristics across morning discharge quartiles was assessed using standardized differences. A standardized difference of less than 0.1 reflects good balance.²⁰

Unadjusted estimates of patient outcomes were reported across morning discharge quartiles. To model the overall association between morning discharge and outcomes, the number of morning GIM discharges on the day of admission was subtracted from the mean number of morning discharges at each hospital and considered as a continuous exposure. We used generalized linear mixed models to estimate the effect of morning discharges on patient outcomes. We fit negative binomial regression models with log link to examine the association between the number of morning discharges (centered by subtracting the hospital mean) and the two main outcomes, ED LOS and hospital LOS. Given the overdispersion of the study population due to the unequal mean and variance, a negative binomial model was preferred over a Poisson regression, as the mean and variance were not equal.²¹ For our secondary outcomes of binary measures (30-day readmission and morality), we fit logistic regression models. Adjustment for multiple comparisons was not performed.

Multivariable analysis was conducted to adjust for the baseline characteristics described above as well as the total number of GIM discharges on the day of admission and GIM census on the day of admission. Hospital and study month (to account for secular time trends) were included as fixed effects, and patients and admitting physicians were included as crossed random effects to account for the nested structure of admissions within patients and admissions within physicians within hospitals.

A sensitivity analysis was performed to assess for nonlinear associations between morning discharges and the four outcomes (hospital LOS, ED LOS, in-hospital mortality, and readmission) by inputting the term as a restricted cubic spline, with up to five knots, in multivariable regression models. We compared the Akaike information criteria (AIC), computed using the log-likelihood, to determine the goodness-of-fit from the negative binomial models.²² We replicated models for each individual hospital to examine whether any hospital-specific associations existed. Two additional sensitivity analyses were performed to examine heterogeneity in hospital-specific effects. Regression models were fit, including interaction terms between morning discharge and hospital, as well as interaction terms between hospital and the total number of GIM discharges and GIM census. The findings of these models (data not presented) were qualitatively similar to the overall results.

Study Population and Patient Characteristics

The study population consisted of 189,781 hospitalizations involving 115,630 unique patients. The median patient age was 73 years (interquartile range [IQR], 57-84), 50.3% were female, 43.8% had a high Charlson Comorbidity Index score, and 11.1% were admitted to GIM in the prior 30 days (Table 1). The median ED LOS was 14.5 hours (IQR, 10.0-23.1), and the mean was 18.1 hours (SD, 12.2). The median hospital LOS was 4.6 days (IQR, 2.4-9.0), and the mean was 8.6 days (SD, 18.7).

In total, 36,043 (19.0%) discharges occurred between 8:00 am and 12:00 pm. The average number of total daily discharges per hospital was 8.4 (SD, 4.6), and the average number of morning discharges was 1.7 (SD, 1.4). Morning discharges varied across hospitals, ranging from 0.9 per day (SD, 1.1) to 2.1 (SD, 1.6) (Appendix Table 1). The average number of morning discharges in the lowest quartile (Q1) was 0.3 (SD, 0.5) and was 3.3 (SD, 1.2) in the highest quartile (Q4). Baseline patient characteristics were well balanced across the quartiles, with standardized differences less than 0.1 (Table 1), except day of admission and GIM census. Days with a greater number of morning discharges were more likely to be weekdays rather than weekends/holidays (89.2% weekday admissions in Q4 compared with 53.9% in Q1). The median GIM census was 93 patients (IQR, 79-109).

Outcomes

Unadjusted clinical outcomes by number of morning discharges are presented in Table 2. The median unadjusted ED LOS was 14.4 (SD, 14.1), 14.3 (SD, 13.2), 14.5 (SD, 13.0), and 14.8 (SD, 13.0) hours for the first to fourth quartiles (fewest to largest number of morning discharges), respectively. The median unadjusted hospital LOS was 4.6 (SD, 6.5), 4.6 (SD, 6.9), 4.7 (SD, 6.4), and 4.6 (SD, 6.4) days for the first to fourth quartiles, respectively.

Unadjusted inpatient mortality was 6.1%, 5.5%, 5.5%, and 5.2% across the first to fourth quartiles, respectively. Unadjusted 30-day readmission to GIM was 12.2%, 12.6%, 12.6%, and 12.5% across the first to fourth quartiles, respectively.

After multivariable adjustment, there was no significant association between morning discharge and hospital LOS (aRR, 1.000; 95% CI, 0.996-1.000; P = .997), ED LOS (aRR, 0.999; 95% CI, 0.997-1.000; P = .307), in-hospital mortality (aRR, 0.967; 95% CI, 0.920-1.020; P =.183), or 30-day readmission (aRR, 1.010; 95% CI, 0.991-1.020; P = .471) (Table 3, Appendix Table 2, Appendix Table 3, Appendix Table 4, Appendix Table 5). When examining each hospital separately, we found that morning discharge was significantly associated with hospital LOS at only one hospital (Hospital D; aRR, 0.981; 95% CI, 0.966-0.996; P = .013). Morning discharge was statistically significantly associated with ED LOS at three hospitals (A, B, and C), but the aRR was at least 0.99 in all three cases (Table 4).

In sensitivity analyses, we found no improvements in model fit when adding spline terms to the model, suggesting no significant nonlinear associations between morning discharges and the outcomes of interest.

DISCUSSION

This large multicenter cohort study found no significant overall association between the number of morning discharges and ED or hospital LOS in GIM. At one hospital, there was a 1.9% reduction in adjusted ED LOS for every additional morning discharge, but no difference in hospital LOS. We also did not observe differences in readmission or inpatient mortality associated with the number of morning discharges. Our observational findings suggest that there is unlikely to be a strong association between morning discharge and patient throughput in GIM. Given that there may be other downstream benefits of morning discharge, such as freeing beds for daytime surgeries,²³ further research is needed to determine the effectiveness of specific interventions.

Several studies have posited morning discharge as a method of improving both patient care and hospital flow metrics.^{10,11,13-15,23} Quality improvement initiatives targeting morning discharges have included stakeholder meetings, incentives programs, discharge-centered breakfast programs, and creating deadlines for discharge orders.^24-29 Although these initiatives have gained support, critics have suggested that their supporting evidence is not robust. Werthemier et al¹⁰ found a 9.0% reduction of observed to expected LOS associated with increasing the number of early discharges. However, a response article suggested that their findings were confounded by other hospital initiatives, such as allocation of medical and social services to weekends.³⁰ Other observational studies have concluded that hospital LOS is not affected by the number of morning discharges, but this research has been limited by single-center analysis and relatively smaller sample sizes.¹² Our study further calls into question the association between morning discharge and patient throughput.

An additional reason for the controversy is that physicians may actively work to discharge patients late in the day to avoid an additional night in hospital. A qualitative study by Minichiello et al³¹ evaluated staff perceptions regarding afternoon discharges. Physicians and medical students believed that afternoon discharges were a result of waiting for test results and procedures, with staff aiming to discharge patients immediately after obtaining results or finishing necessary procedures. As such, there are concerns that incentivizing morning discharge may lead physicians in the opposite direction, to consciously or unconsciously keep patients overnight in order to facilitate an early morning discharge.³⁰

Our study’s greatest strength was the large sample size over 7 years at seven hospitals in two cities, including both academic and community hospitals with different models of care. To our knowledge, this is the first cohort study that has analyzed the association between early discharge and LOS using multiple centers. To avoid the confounding and reverse causality that may exist when examining the relationship between LOS and morning discharge at the patient level (eg, patients who stay in hospital longer may have more “planned” discharges and leave in the morning), we examined the association based on variation across different days within the GIM service of each hospital. Further, we included robust risk adjustment using clinical and laboratory data. Finally, since our study included a diverse patient population served by participating centers in a system with universal insurance for hospital care, our findings are likely generalizable to other urban and suburban hospitals.

There are several important limitations of our analysis. First, we could only include GIM patients, who represent nearly 40% of ED admissions to hospital at participating centers. A more holistic analysis across all hospital services could be justified; however, given that many quality improvement initiatives occur at the level of a single hospital service, we felt our approach would be informative for future research and improvement efforts. Approximately 75% of GIM patients at participating hospitals were cared for on a GIM ward, with 25% cared for on off-service units. We were unable to include the total hospital census in our models, and this could affect LOS and waiting times for GIM patients, particularly those admitted to off-service units. GIM census is likely highly correlated with hospital census, and we were able to adjust for this. Nevertheless, this remains an important potential source of unmeasured confounding. Second, we did not model the effects of morning discharges from GIM on patient-flow measures for non-GIM patients. Given the lack of effects for GIM patients, who would be more likely to be directly affected, it is unlikely that large effects would be seen for other hospital patients, but we did not measure effects on surgical delays or cancellations, for example.²³ Third, we report 30-day readmission to GIM at participating hospitals only, rather than all readmissions. However, prior research in our region demonstrated that 82% of hospital readmissions occur to the same site.³² Thus, our measure, which includes admission to any participating hospital, likely captures more than 80% of all readmissions, and this was a secondary outcome in our analysis. Finally, qualitative metrics, such as patient or provider satisfaction, were not measured in our study. Earlier discharge may impact patient care in other ways by being more predictable for staff, improving bed allocation for daytime procedures, making medication pick-ups easier to arrange, or making consultations with allied health services more convenient.^11,28,33 Conversely, if pressured to discharge before noon, providers may feel rushed to complete tasks and may face disruptions to typical workflow.²⁴ As such, future research is needed to provide a more complete understanding of the impact of early-morning discharge beyond hospital flow. Given the lack of a strong association observed between morning discharge and patient throughput in our study, further research should also consider the opportunity costs of interventions designed to improve morning discharge.

CONCLUSION

The number of morning discharges was not significantly associated with shorter ED LOS or hospital LOS for GIM patients. Our observational findings suggest that increasing morning discharges alone may not substantially improve patient flow in GIM. Further research is needed to evaluate specific morning discharge interventions and assess hospital-wide effects.

There is substantial interest in improving patient flow and reducing hospital length of stay (LOS).^1-4 Impaired hospital flow may negatively impact both patient satisfaction and safety through, for example, emergency department (ED) overcrowding.^5,6 Impaired hospital flow is associated with downstream effects on patient care, hospital costs, and availability of beds.^7-9

A number of quality-improvement interventions aim to improve patient flow, including efforts to increase the number of discharges that occur before noon.^10,11 Morning discharges have been hypothesized to free hospital beds earlier, thus reducing ED wait times for incoming patients and increasing beds for elective surgeries.¹¹ Morning discharges may also be more predictable for staff and patients. However, it is unclear whether efforts to increase the number of morning discharges have a negative impact on inpatient LOS by incentivizing physicians to keep patients in the hospital for an extra night to facilitate discharge in the early morning rather than the late afternoon. Morning discharges have been associated with both increased¹² and decreased LOS.^10,11,13-15

The purpose of this study was to examine the associations between morning discharges and ED LOS and hospital LOS in general internal medicine (GIM) at seven hospitals. GIM patients represent nearly 40% of ED admissions to a hospital,¹⁶ and thus are an important determinant of patient flow through the ED and hospital. We hypothesized that patients who were admitted to GIM on days with more morning discharges would have shorter ED LOS and hospital LOS.

Design, Setting, and Participants

This was a retrospective cohort study conducted using the General Medicine Inpatient Initiative (GEMINI) clinical dataset.¹⁶ The dataset includes all GIM admissions at seven large hospital sites in Toronto and Mississauga, Ontario, Canada. These include five academic hospitals and two community-based teaching hospitals. Each hospital is publicly funded and provides tertiary and/or quaternary care to diverse multiethnic populations. Research ethics board approval was obtained from all participating sites.

GIM care is delivered by several interdisciplinary clinical teams functioning in parallel. Attending physicians are predominantly internists who practice as hospitalists in discrete service blocks, typically lasting 2 weeks at a time. Although GIM patients are preferentially admitted to GIM wards, participating hospitals did not have strict policies regarding cohorting GIM patients to specific wards (ie, holding patients in ED until a specific bed becomes available) that would confound the association between morning discharge and ED wait times. Approximately 75% of GIM patients are cared for on dedicated GIM wards at participating hospitals, with the remainder cared for on other medical or surgical wards.

We included all hospitalized patients who were admitted to hospital and discharged from GIM between April 1, 2010, and October 31, 2017, from the seven GEMINI hospitals. We included only patients admitted through the ED. As such, we did not include elective admissions or interfacility transfers who would not experience ED wait times. We excluded patients who were discharged without a provincial health insurance number (N = 2,169; 1.1% of total sample) because they could not be linked across visits to measure readmissions.

Data Source

The GEMINI dataset has been rigorously validated and previously described in detail.¹⁶ GEMINI collects both administrative health data reported to the Canadian Institute for Health Information (including data about patient demographics, comorbidities, and discharge destination) as well as electronic clinical data extracted from hospital computer systems (including attending physicians, in-hospital patient room transfers, and laboratory test results). Data are collected for each individual hospital encounter, and the provincial health insurance number is used to link patients across encounters.

Exposures and Outcomes

The primary exposure was the number of GIM patients discharged alive between 8:00 am and 12:00 pm (ie, morning GIM discharges) on the day of admission for each hospital encounter. This time window to define morning discharges was selected based on previous literature.^10-13 To report admission characteristics and unadjusted outcomes, hospital days were categorized into quartiles within each hospital based on the number of morning GIM discharges. Hospital days with the lowest number of morning discharges were classified into Q1 and the highest number of morning discharges into Q4, and quartiles were pooled across hospitals.

The two primary outcomes were ED LOS and hospital LOS. ED LOS was calculated as the difference between the time from triage by nursing staff to a patient’s exit from the ED, measured in hours. We also examined 30-day readmission to GIM at any participating hospital as a balancing measure against premature discharges and inpatient mortality because it could modify hospital LOS.

Patient Characteristics

Baseline patient characteristics were measured, including age, sex, Charlson Comorbidity Index score,¹⁷ day of admission (categorized as weekend/holiday or weekday), time of admission to hospital (categorized as daytime, 8:00 am to 4:59 pm, or nighttime, 5:00 pm to 07:59 am), study month, hospital site, and whether patients had been admitted to GIM in the prior 30 days. We used laboratory data to calculate the baseline Laboratory-based Acute Physiology Score (LAPS), which is a validated predictor of inpatient mortality when combined with age and Charlson Comorbidity Index score.^18,19 GIM census on day of admission was calculated in order to include overall patient volumes as an important adjustment variable that could confound the association between morning discharges and patient flow.

Statistical Analysis

The study population and physician characteristics were summarized with descriptive statistics. The balance of baseline patient characteristics across morning discharge quartiles was assessed using standardized differences. A standardized difference of less than 0.1 reflects good balance.²⁰

Unadjusted estimates of patient outcomes were reported across morning discharge quartiles. To model the overall association between morning discharge and outcomes, the number of morning GIM discharges on the day of admission was subtracted from the mean number of morning discharges at each hospital and considered as a continuous exposure. We used generalized linear mixed models to estimate the effect of morning discharges on patient outcomes. We fit negative binomial regression models with log link to examine the association between the number of morning discharges (centered by subtracting the hospital mean) and the two main outcomes, ED LOS and hospital LOS. Given the overdispersion of the study population due to the unequal mean and variance, a negative binomial model was preferred over a Poisson regression, as the mean and variance were not equal.²¹ For our secondary outcomes of binary measures (30-day readmission and morality), we fit logistic regression models. Adjustment for multiple comparisons was not performed.

Multivariable analysis was conducted to adjust for the baseline characteristics described above as well as the total number of GIM discharges on the day of admission and GIM census on the day of admission. Hospital and study month (to account for secular time trends) were included as fixed effects, and patients and admitting physicians were included as crossed random effects to account for the nested structure of admissions within patients and admissions within physicians within hospitals.

A sensitivity analysis was performed to assess for nonlinear associations between morning discharges and the four outcomes (hospital LOS, ED LOS, in-hospital mortality, and readmission) by inputting the term as a restricted cubic spline, with up to five knots, in multivariable regression models. We compared the Akaike information criteria (AIC), computed using the log-likelihood, to determine the goodness-of-fit from the negative binomial models.²² We replicated models for each individual hospital to examine whether any hospital-specific associations existed. Two additional sensitivity analyses were performed to examine heterogeneity in hospital-specific effects. Regression models were fit, including interaction terms between morning discharge and hospital, as well as interaction terms between hospital and the total number of GIM discharges and GIM census. The findings of these models (data not presented) were qualitatively similar to the overall results.

Study Population and Patient Characteristics

The study population consisted of 189,781 hospitalizations involving 115,630 unique patients. The median patient age was 73 years (interquartile range [IQR], 57-84), 50.3% were female, 43.8% had a high Charlson Comorbidity Index score, and 11.1% were admitted to GIM in the prior 30 days (Table 1). The median ED LOS was 14.5 hours (IQR, 10.0-23.1), and the mean was 18.1 hours (SD, 12.2). The median hospital LOS was 4.6 days (IQR, 2.4-9.0), and the mean was 8.6 days (SD, 18.7).

In total, 36,043 (19.0%) discharges occurred between 8:00 am and 12:00 pm. The average number of total daily discharges per hospital was 8.4 (SD, 4.6), and the average number of morning discharges was 1.7 (SD, 1.4). Morning discharges varied across hospitals, ranging from 0.9 per day (SD, 1.1) to 2.1 (SD, 1.6) (Appendix Table 1). The average number of morning discharges in the lowest quartile (Q1) was 0.3 (SD, 0.5) and was 3.3 (SD, 1.2) in the highest quartile (Q4). Baseline patient characteristics were well balanced across the quartiles, with standardized differences less than 0.1 (Table 1), except day of admission and GIM census. Days with a greater number of morning discharges were more likely to be weekdays rather than weekends/holidays (89.2% weekday admissions in Q4 compared with 53.9% in Q1). The median GIM census was 93 patients (IQR, 79-109).

Outcomes

Unadjusted clinical outcomes by number of morning discharges are presented in Table 2. The median unadjusted ED LOS was 14.4 (SD, 14.1), 14.3 (SD, 13.2), 14.5 (SD, 13.0), and 14.8 (SD, 13.0) hours for the first to fourth quartiles (fewest to largest number of morning discharges), respectively. The median unadjusted hospital LOS was 4.6 (SD, 6.5), 4.6 (SD, 6.9), 4.7 (SD, 6.4), and 4.6 (SD, 6.4) days for the first to fourth quartiles, respectively.

Unadjusted inpatient mortality was 6.1%, 5.5%, 5.5%, and 5.2% across the first to fourth quartiles, respectively. Unadjusted 30-day readmission to GIM was 12.2%, 12.6%, 12.6%, and 12.5% across the first to fourth quartiles, respectively.

After multivariable adjustment, there was no significant association between morning discharge and hospital LOS (aRR, 1.000; 95% CI, 0.996-1.000; P = .997), ED LOS (aRR, 0.999; 95% CI, 0.997-1.000; P = .307), in-hospital mortality (aRR, 0.967; 95% CI, 0.920-1.020; P =.183), or 30-day readmission (aRR, 1.010; 95% CI, 0.991-1.020; P = .471) (Table 3, Appendix Table 2, Appendix Table 3, Appendix Table 4, Appendix Table 5). When examining each hospital separately, we found that morning discharge was significantly associated with hospital LOS at only one hospital (Hospital D; aRR, 0.981; 95% CI, 0.966-0.996; P = .013). Morning discharge was statistically significantly associated with ED LOS at three hospitals (A, B, and C), but the aRR was at least 0.99 in all three cases (Table 4).

In sensitivity analyses, we found no improvements in model fit when adding spline terms to the model, suggesting no significant nonlinear associations between morning discharges and the outcomes of interest.

DISCUSSION

This large multicenter cohort study found no significant overall association between the number of morning discharges and ED or hospital LOS in GIM. At one hospital, there was a 1.9% reduction in adjusted ED LOS for every additional morning discharge, but no difference in hospital LOS. We also did not observe differences in readmission or inpatient mortality associated with the number of morning discharges. Our observational findings suggest that there is unlikely to be a strong association between morning discharge and patient throughput in GIM. Given that there may be other downstream benefits of morning discharge, such as freeing beds for daytime surgeries,²³ further research is needed to determine the effectiveness of specific interventions.

Several studies have posited morning discharge as a method of improving both patient care and hospital flow metrics.^{10,11,13-15,23} Quality improvement initiatives targeting morning discharges have included stakeholder meetings, incentives programs, discharge-centered breakfast programs, and creating deadlines for discharge orders.^24-29 Although these initiatives have gained support, critics have suggested that their supporting evidence is not robust. Werthemier et al¹⁰ found a 9.0% reduction of observed to expected LOS associated with increasing the number of early discharges. However, a response article suggested that their findings were confounded by other hospital initiatives, such as allocation of medical and social services to weekends.³⁰ Other observational studies have concluded that hospital LOS is not affected by the number of morning discharges, but this research has been limited by single-center analysis and relatively smaller sample sizes.¹² Our study further calls into question the association between morning discharge and patient throughput.

An additional reason for the controversy is that physicians may actively work to discharge patients late in the day to avoid an additional night in hospital. A qualitative study by Minichiello et al³¹ evaluated staff perceptions regarding afternoon discharges. Physicians and medical students believed that afternoon discharges were a result of waiting for test results and procedures, with staff aiming to discharge patients immediately after obtaining results or finishing necessary procedures. As such, there are concerns that incentivizing morning discharge may lead physicians in the opposite direction, to consciously or unconsciously keep patients overnight in order to facilitate an early morning discharge.³⁰

Our study’s greatest strength was the large sample size over 7 years at seven hospitals in two cities, including both academic and community hospitals with different models of care. To our knowledge, this is the first cohort study that has analyzed the association between early discharge and LOS using multiple centers. To avoid the confounding and reverse causality that may exist when examining the relationship between LOS and morning discharge at the patient level (eg, patients who stay in hospital longer may have more “planned” discharges and leave in the morning), we examined the association based on variation across different days within the GIM service of each hospital. Further, we included robust risk adjustment using clinical and laboratory data. Finally, since our study included a diverse patient population served by participating centers in a system with universal insurance for hospital care, our findings are likely generalizable to other urban and suburban hospitals.

There are several important limitations of our analysis. First, we could only include GIM patients, who represent nearly 40% of ED admissions to hospital at participating centers. A more holistic analysis across all hospital services could be justified; however, given that many quality improvement initiatives occur at the level of a single hospital service, we felt our approach would be informative for future research and improvement efforts. Approximately 75% of GIM patients at participating hospitals were cared for on a GIM ward, with 25% cared for on off-service units. We were unable to include the total hospital census in our models, and this could affect LOS and waiting times for GIM patients, particularly those admitted to off-service units. GIM census is likely highly correlated with hospital census, and we were able to adjust for this. Nevertheless, this remains an important potential source of unmeasured confounding. Second, we did not model the effects of morning discharges from GIM on patient-flow measures for non-GIM patients. Given the lack of effects for GIM patients, who would be more likely to be directly affected, it is unlikely that large effects would be seen for other hospital patients, but we did not measure effects on surgical delays or cancellations, for example.²³ Third, we report 30-day readmission to GIM at participating hospitals only, rather than all readmissions. However, prior research in our region demonstrated that 82% of hospital readmissions occur to the same site.³² Thus, our measure, which includes admission to any participating hospital, likely captures more than 80% of all readmissions, and this was a secondary outcome in our analysis. Finally, qualitative metrics, such as patient or provider satisfaction, were not measured in our study. Earlier discharge may impact patient care in other ways by being more predictable for staff, improving bed allocation for daytime procedures, making medication pick-ups easier to arrange, or making consultations with allied health services more convenient.^11,28,33 Conversely, if pressured to discharge before noon, providers may feel rushed to complete tasks and may face disruptions to typical workflow.²⁴ As such, future research is needed to provide a more complete understanding of the impact of early-morning discharge beyond hospital flow. Given the lack of a strong association observed between morning discharge and patient throughput in our study, further research should also consider the opportunity costs of interventions designed to improve morning discharge.

CONCLUSION

The number of morning discharges was not significantly associated with shorter ED LOS or hospital LOS for GIM patients. Our observational findings suggest that increasing morning discharges alone may not substantially improve patient flow in GIM. Further research is needed to evaluate specific morning discharge interventions and assess hospital-wide effects.

Nearly 48,000 Americans died from overdoses involving opioids in 2018, continuing a national crisis that has led to 446,000 deaths since 1999.¹ Annually, opioids are responsible for more than 500,000 admissions, approximately 1% of all hospitalizations, costing the United States nearly $15 billion.^2,3 Among hospitalized patients, chronic opioid use is associated with increased mortality, severe infectious complications, and higher rates of readmission.⁴ Opioid use disorder (OUD) is a chronic, relapsing medical condition with biopsychosocial origins and significant morbidity and mortality.⁵ Opioid agonist therapy (OAT) with buprenorphine or methadone maintenance, the evidence-based standard of treatment, reduces the mortality rate by half, decreases overdoses and hospital readmissions, and improves retention in care.^6-10

OAT maintenance refers to using buprenorphine or methadone for long-term treatment of OUD rather than for acute treatment of opioid withdrawal. Despite evidence supporting OAT maintenance, clinicians start medications for only 11% to 15% of hospitalized patients with OUD, depending on practice contexts.^11,12 Three significant barriers—stigma, insufficient clinician education, and restrictive regulations—prevent clinicians from starting OAT.¹³ Clinicians who do not have the Drug Enforcement Administration (DEA)–issued DATA-2000 waiver (X-waiver) for outpatient prescribing can order buprenorphine for admitted patients but cannot prescribe it at discharge.¹⁴ In hospitals where they exist, addiction medicine consult services offer primary teams guidance on pharmacotherapy, leading to reduced hospital readmissions and increased engagement in outpatient addiction treatment.^15-17 However, in most hospitals around the country, such specialty services do not exist.¹⁸ In some hospitals without addiction medicine consult services, hospitalists with expertise in OUD have started assisting primary teams in starting OAT, but to our knowledge, no prior studies have described the impact of these interventions on patients or clinician experience with OAT.¹⁹

This quality improvement project aimed to increase the rate at which internal medicine resident teams at Johns Hopkins Hospital (JHH) in Baltimore, Maryland, started hospitalized patients with OUD on buprenorphine maintenance. We hypothesized that resident education and measures to increase the availability of X-waivered physicians would increase the rate of initiating buprenorphine maintenance. We additionally hypothesized that these interventions would increase knowledge about and comfort with buprenorphine across the residency. This represents the first study to examine the effects of clinician education and a team of X-waivered residents and hospitalists who assist in starting buprenorphine maintenance in a hospital without an addiction medicine consult service.

Setting

This study took place from July 2018 to June 2019 at JHH, a large, academic, urban hospital in Baltimore. Prior to the intervention, internal medicine residents at JHH commonly used short courses of buprenorphine to treat withdrawal, but they did not have access to hospital-specific resources to assist with starting maintenance OAT. During the study period, JHH had a Substance Use Disorders team staffed by peer recovery specialists that could be consulted by hospitalists and residents to provide psychosocial support and link admitted patients to treatment after discharge. There were no providers on the team to guide pharmacotherapy or to write discharge buprenorphine prescriptions. The Osler Medical Residency Training Program at JHH has 140 internal medicine residents and 16 combined medicine-pediatrics residents. All residents receive 1 hour of formal education about opioid use disorder annually. In addition, 28 of those 156 residents, those in the Urban Health Primary Care track, spend 1 month on an Addiction Medicine rotation in which they complete the 8-hour training required to receive the X-waiver. Those residents are encouraged to apply for the X-waiver once they obtain a medical license subsidized by a Health Resources & Services Administration (HRSA) grant. Four internal medicine attending physicians on teaching services and one resident had X-waivers prior to the intervention.

Intervention

In November 2018, we administered a survey to residents to identify barriers to starting buprenorphine maintenance and to measure knowledge and confidence with using buprenorphine for OUD (Appendix Figure 1 and Figure 2). We focused on buprenorphine because providers at JHH were familiar with this medication and because Baltimore has widespread access to buprenorphine, with more than 490 local buprenorphine providers.²⁰ Five residents piloted the survey and provided feedback. We then administered the survey to all internal medicine and medicine-pediatrics residents. Based on the results, we developed a targeted educational conference and also created the Buprenorphine Bridge Team (BBT).

In January 2019, we presented the educational conference for residents devoted to the use of buprenorphine for OUD and introduced the BBT. The conference started with a patient testimonial and included peer recovery specialists, pharmacists, nurses, and social workers. We summarized the evidence for buprenorphine and offered a practical guide to start treatment in a one-page protocol. This protocol included guidance on selecting patients, shared decision-making around OUD treatment, avoiding precipitated withdrawal, dosing buprenorphine, and establishing follow-up (Appendix Figure 3). We asked for input on this protocol from nursing leadership, social work teams, and peer recovery specialists. Dosing was adapted from the Guidelines from the American Society of Addiction Medicine, with expert input from physicians from the Addiction Medicine Consult service at Johns Hopkins Bayview Medical Center, also in Baltimore.⁵ We instructed residents to obtain discharge buprenorphine prescriptions from an X-waivered physician on their team or from the newly established BBT. We asked resident teams to set up a postdischarge appointment for patients with an X-waivered provider, either in a community practice or at the JHH After Care Clinic, a transitional care clinic for discharged patients.²¹

The BBT is a resident-led group of X-waivered JHH residents and hospitalists who volunteer to write discharge buprenorphine prescriptions for patients. The BBT serves to ensure primary teams have access to an X-waivered prescriber. It is not a consult service. We asked primary teams to contact the BBT after initiating buprenorphine and after securing a follow-up appointment. In response to each request, a member of the BBT reviews the patient chart, confirms the follow-up plan, writes a prescription for buprenorphine along with intranasal naloxone, and leaves a brief note. During the 6-month postintervention period, the team consisted of three residents and three hospitalist attendings. Each week, two members (residents or attendings) staffed the team Monday to Friday, 8 am to 5 pm. Most weeks were staffed by two residents. One resident provided services after hours and during weekends. Resident team members ensured that the buprenorphine plan was discussed with the primary team’s attending. For dosing questions beyond the BBT’s scope of knowledge, a member of the BBT relayed questions to physicians from the Addiction Medicine Consult team at Johns Hopkins Bayview Medical Center.

In May 2019, 5 months after the education session and implementation of the BBT, we administered a follow-up survey.

Outcomes

The primary outcome was the percent of inpatients eligible to start OAT who were discharged on buprenorphine maintenance. We obtained data from the electronic medical record. The denominator consisted of patients with OUD not on buprenorphine or methadone maintenance on admission. We identified patients with OUD by an opioid-related International Classification of Diseases, Tenth Revision (ICD-10) diagnosis code or by a standing or as-needed order for buprenorphine or methadone during hospitalization.²² We reviewed admission and discharge documentation to identify patients with OUD who were not in active treatment with buprenorphine or methadone maintenance.

As a secondary outcome, we measured engagement in OUD treatment after discharge by calculating the proportion of patients started on buprenorphine who filled a buprenorphine prescription within 30 days after discharge. We chose 30 days based on the National Committee for Quality Assurance’s Healthcare Effectiveness Data and Information Set (HEDIS) measure for engagement of treatment for alcohol and other drugs.²³ We obtained the data from the Chesapeake Regional Information System for our Patients (CRISP) Prescription Drug Monitoring Program, which monitors all prescriptions for controlled substances dispensed in Maryland and five neighboring states. As a balancing measure, we counted patients newly started on methadone maintenance for OUD before and after the intervention. Additional secondary process outcomes included frequency of BBT requests, the volume of buprenorphine prescriptions written by the team, and time required to complete a BBT request.

Clinician-level outcomes, measured with electronically administered pre- and postintervention surveys to residents, included knowledge about and comfort with buprenorphine. Of the 16 questions in the pre- and postimplementation surveys, we analyzed the 6 questions concerning knowledge and comfort that remained identical in the pre- and postintervention surveys and used 5-point Likert scale responses. As an incentive, we randomly distributed three $50 gift cards to survey completers.

Analysis

We used an interrupted time series analysis to evaluate the association between the intervention bundle and a change in the rate that medical teams started patients with OUD on buprenorphine maintenance. This approach allowed us to control for preintervention trends. To evaluate the impact of our interventions, our pre- and postintervention periods include the same residents during the 2018-2019 academic year. Both periods consisted of twelve 2-week intervals (preintervention: July 26, 2019, to January 9, 2019; postintervention: January 10, 2019, to June 26, 2019).

To evaluate for changes in engagement in OUD treatment after discharge, we used two-sample t tests. To evaluate for changes in resident-reported comfort and knowledge with initiating buprenorphine maintenance, we used Wilcoxon rank sum tests for survey data and Wilcoxon signed rank tests for paired data. All analyses employed two-sided P values with statistical significance evaluated at the .05 alpha level. We analyzed data using R version 3.6.3 (Foundation for Statistical Computing). The Institutional Review Board at JHH reviewed and approved the study protocol as a quality improvement project (IRB00193365).

During the 24-week preintervention period, internal medicine resident teams started 30 out of 305 eligible patients (10%) on buprenorphine maintenance vs 64 out of 270 eligible patients (24%) during the 24-week postintervention period. Our interrupted time series analysis showed a significant increase in the percent of eligible patients started on buprenorphine maintenance (expected number of patients started postintervention, 27; actual, 64; absolute increase in percent, 14.4%; 95% CI, 3.6%-25.3%; P = .017) (Figure). There was no significant trend during the preintervention period and no significant trend during the postintervention period.

Before the intervention, 13 of the 30 patients (40%) newly started on buprenorphine maintenance during their admission filled a follow-up buprenorphine prescription within 30 days of discharge. After the intervention, 31 of 64 patients (46%) filled a buprenorphine prescription within 30 days (P = .612). Two patients were started on methadone maintenance, one prior to and one after the intervention.

During the 6-month postintervention period, the BBT received 75 requests and wrote 70 prescriptions for buprenorphine. The median time required to complete a BBT request was 15 minutes (minimum, 5 minutes; maximum, 60 minutes).

Of 156 internal medicine and medicine-pediatrics residents, 89 residents (57%) completed the baseline survey and 66 residents (42%) completed the follow-up survey. Forty residents completed both surveys. After the intervention, residents were significantly more likely to feel comfortable dosing buprenorphine (P < .0001) and counseling patients about its use (P = .0237) and were more likely to report ease of establishing follow-up (P < .0001). Self-reported knowledge about preventing precipitated withdrawal increased significantly (P = .0191), as did knowledge about the effectiveness of buprenorphine (P = .0003) independent of formal drug counseling (P = .0066) (Table). Paired survey data also found statistically significant results for all questions except those about preventing precipitated withdrawal and efficacy. For the latter, respondents who completed both surveys were more knowledgeable before the intervention than the overall group that completed the baseline survey (Appendix Table).

This study shows how a resident-led quality improvement project comprising clinician education and implementation of a novel BBT was associated with an increased rate of starting buprenorphine maintenance in hospitalized patients with OUD and improved resident knowledge about and comfort with buprenorphine. To our knowledge, this is the first study demonstrating how education and a team of X-waivered generalists can help primary teams initiate and discharge patients on buprenorphine maintenance in a hospital without an addiction medicine consult service.

Prior to the intervention, resident internal medicine teams at JHH started 10% of hospitalized patients with OUD on buprenorphine maintenance, consistent with prior studies showing rates of 11% to 15% for initiating OAT for hospitalized patients.^11,12 After the intervention, the rate of initiating buprenorphine maintenance more than doubled, rising to 24% of eligible patients. Resident internal medicine teams at JHH started buprenorphine maintenance for 37 more patients over the 24-week postintervention period than would have been predicted prior to the intervention, or an additional three patients every 2 weeks.

Between 40% and 46% of hospitalized patients newly started on buprenorphine maintenance filled an outpatient buprenorphine prescription within 30 days of discharge. We are not aware of comparative data for 30-day follow-up for hospitalized patients newly started on buprenorphine maintenance. Data from other contexts show 5% to 10% of veterans were engaged in addiction treatment 30 days after initiation from inpatient or outpatient encounters. An analysis of an academic medical center in Oregon found engagement with an addiction medicine consult service increased after hospital engagement for patients with any substance use disorder from 23% to 39% using the 34-day HEDIS measure for engagement.^17,24,25

The BBT required approximately 15 minutes per request and wrote an average of three prescriptions per week, demonstrating the feasibility of this approach and the high demand for this service. One strength of our approach is that residents gained experience starting buprenorphine independently using the aforementioned protocol instead of deferring to a full consult service. It is likely that this resident engagement in initiating longitudinal OUD care contributed to the success of this initiative, as did existing resident familiarity with using buprenorphine for opioid withdrawal.

This approach to resident education—promoting direct, first-person experience with medications in a clinical context—aligns with recommendations from a recent review about substance use disorder education for health professionals.²⁶ Our interventions increased resident knowledge and comfort with buprenorphine, consistent with prior studies showing increased resident confidence in management of substance use disorders after curricular innovations.^24,25

A few contextual features were essential for this project’s viability. Maryland allows American medical graduates to obtain a medical license after 1 year of postgraduate training. This allowed three residents to obtain X-waivers. These residents had access to HRSA funding to subsidize the expenses of applying for state licensure and DEA registration. BBT members volunteered their time while working on other services. Last, we were able to take advantage of buprenorphine-providing clinics in Baltimore, including the JHH After Care Clinic, to accept patients for follow-up appointments after discharge.

Limitations

The BBT required motivated clinicians willing to volunteer for additional clinical responsibilities during inpatient rotations and supportive faculty and residency leadership. Attending physicians, nurse practitioners, or physician assistants could staff a similar BBT in hospitals without residents or in hospitals where residents cannot obtain DEA registration. Crucially, other hospitals may not have access to practices with X-waivered physicians for outpatient follow-up. A recent study found X-waivered primary care physicians were less likely to be affiliated with hospital health systems. Other studies have shown limitations in access to buprenorphine at the county level based on geography and racial/ethnic segregation.^27-29

Most patients hospitalized with OUD did not have ICD-10 codes associated with OUD. We addressed this by assuming patients had OUD if buprenorphine or methadone was ordered during their hospitalization, even if the medication was never administered. This may have overcounted patients prescribed these medications for indications other than OUD, and it may have undercounted patients with OUD for whom buprenorphine or methadone were never considered. The opioid withdrawal order set at JHH automatically offers an option to use buprenorphine to treat withdrawal. Patients with OUD for whom buprenorphine or methadone were never ordered likely did not experience withdrawal or were in withdrawal so mild that it escaped the attention of the team, which limits the generalizability of our intervention.

We identified several limitations to the internal validity of our study. First, we used a before-and-after study design without a control group. We could not ethically withhold access to evidence-based, mortality-reducing medications from patients. Without a control group, we cannot rule out the possibility that underlying temporal trends made residents more likely to start buprenorphine maintenance independent of our intervention. We attempted to control for unmeasured confounders by using an interrupted time series analysis to control for preintervention trends, comparing the same group of residents before and after our interventions, and selecting an intervention period during which residents were given only educational sessions and materials provided by our team. Our results may be biased by clustered data because certain residents may have been more likely to initiate buprenorphine, but these effects are likely marginal because resident schedules are balanced between outpatient and inpatient rotations during each 6-month period.

Finally, this project focused on buprenorphine, not on other medications for OUD, including methadone or naltrexone, or nonpharmacologic treatments for OUD.

Since the start of the BBT in January 2019, five additional PGY-2 residents obtained their medical licenses and X-waivers. These residents, with the support of two attending hospitalists, led the BBT and coordinated education sessions that were incorporated into the curriculum during the 2019-2020 academic year. These educational sessions will continue indefinitely. In 2020, JHH started an Addiction Medicine Consult Service staffed by physicians, NPs, and a pharmacist. The BBT continues to operate in conjunction with this service.

We found substantial variability in the rate of buprenorphine maintenance initiation despite our interventions. This is an area for future improvement. In a free-response prompt in our follow-up survey, residents requested additional education sessions and an order set to assist with initiation of buprenorphine. To address these gaps, three educational sessions were added, one of which included education on starting methadone maintenance therapy. We also added a new order set for starting buprenorphine maintenance. We hypothesize that these interventions will improve consistency.

In order for a similar program to be disseminated to other institutions, educational initiatives and a team of dedicated X-waivered prescribers are key. Materials to assist with this process are available in the Appendix.

This study shows how a resident-led intervention comprising clinician education and a team of X-waivered generalists was associated with improved treatment of OUD for hospitalized patients. We encourage residents and all clinicians at other hospitals without addiction medicine consult services to design, implement, and study similar interventions that directly increase the use of buprenorphine or methadone maintenance to treat OUD.

Preliminary results from this project were presented at the AMERSA National Conference on November 7, 2019.

Nearly 48,000 Americans died from overdoses involving opioids in 2018, continuing a national crisis that has led to 446,000 deaths since 1999.¹ Annually, opioids are responsible for more than 500,000 admissions, approximately 1% of all hospitalizations, costing the United States nearly $15 billion.^2,3 Among hospitalized patients, chronic opioid use is associated with increased mortality, severe infectious complications, and higher rates of readmission.⁴ Opioid use disorder (OUD) is a chronic, relapsing medical condition with biopsychosocial origins and significant morbidity and mortality.⁵ Opioid agonist therapy (OAT) with buprenorphine or methadone maintenance, the evidence-based standard of treatment, reduces the mortality rate by half, decreases overdoses and hospital readmissions, and improves retention in care.^6-10

OAT maintenance refers to using buprenorphine or methadone for long-term treatment of OUD rather than for acute treatment of opioid withdrawal. Despite evidence supporting OAT maintenance, clinicians start medications for only 11% to 15% of hospitalized patients with OUD, depending on practice contexts.^11,12 Three significant barriers—stigma, insufficient clinician education, and restrictive regulations—prevent clinicians from starting OAT.¹³ Clinicians who do not have the Drug Enforcement Administration (DEA)–issued DATA-2000 waiver (X-waiver) for outpatient prescribing can order buprenorphine for admitted patients but cannot prescribe it at discharge.¹⁴ In hospitals where they exist, addiction medicine consult services offer primary teams guidance on pharmacotherapy, leading to reduced hospital readmissions and increased engagement in outpatient addiction treatment.^15-17 However, in most hospitals around the country, such specialty services do not exist.¹⁸ In some hospitals without addiction medicine consult services, hospitalists with expertise in OUD have started assisting primary teams in starting OAT, but to our knowledge, no prior studies have described the impact of these interventions on patients or clinician experience with OAT.¹⁹

This quality improvement project aimed to increase the rate at which internal medicine resident teams at Johns Hopkins Hospital (JHH) in Baltimore, Maryland, started hospitalized patients with OUD on buprenorphine maintenance. We hypothesized that resident education and measures to increase the availability of X-waivered physicians would increase the rate of initiating buprenorphine maintenance. We additionally hypothesized that these interventions would increase knowledge about and comfort with buprenorphine across the residency. This represents the first study to examine the effects of clinician education and a team of X-waivered residents and hospitalists who assist in starting buprenorphine maintenance in a hospital without an addiction medicine consult service.

Setting

This study took place from July 2018 to June 2019 at JHH, a large, academic, urban hospital in Baltimore. Prior to the intervention, internal medicine residents at JHH commonly used short courses of buprenorphine to treat withdrawal, but they did not have access to hospital-specific resources to assist with starting maintenance OAT. During the study period, JHH had a Substance Use Disorders team staffed by peer recovery specialists that could be consulted by hospitalists and residents to provide psychosocial support and link admitted patients to treatment after discharge. There were no providers on the team to guide pharmacotherapy or to write discharge buprenorphine prescriptions. The Osler Medical Residency Training Program at JHH has 140 internal medicine residents and 16 combined medicine-pediatrics residents. All residents receive 1 hour of formal education about opioid use disorder annually. In addition, 28 of those 156 residents, those in the Urban Health Primary Care track, spend 1 month on an Addiction Medicine rotation in which they complete the 8-hour training required to receive the X-waiver. Those residents are encouraged to apply for the X-waiver once they obtain a medical license subsidized by a Health Resources & Services Administration (HRSA) grant. Four internal medicine attending physicians on teaching services and one resident had X-waivers prior to the intervention.

Intervention

In November 2018, we administered a survey to residents to identify barriers to starting buprenorphine maintenance and to measure knowledge and confidence with using buprenorphine for OUD (Appendix Figure 1 and Figure 2). We focused on buprenorphine because providers at JHH were familiar with this medication and because Baltimore has widespread access to buprenorphine, with more than 490 local buprenorphine providers.²⁰ Five residents piloted the survey and provided feedback. We then administered the survey to all internal medicine and medicine-pediatrics residents. Based on the results, we developed a targeted educational conference and also created the Buprenorphine Bridge Team (BBT).

In January 2019, we presented the educational conference for residents devoted to the use of buprenorphine for OUD and introduced the BBT. The conference started with a patient testimonial and included peer recovery specialists, pharmacists, nurses, and social workers. We summarized the evidence for buprenorphine and offered a practical guide to start treatment in a one-page protocol. This protocol included guidance on selecting patients, shared decision-making around OUD treatment, avoiding precipitated withdrawal, dosing buprenorphine, and establishing follow-up (Appendix Figure 3). We asked for input on this protocol from nursing leadership, social work teams, and peer recovery specialists. Dosing was adapted from the Guidelines from the American Society of Addiction Medicine, with expert input from physicians from the Addiction Medicine Consult service at Johns Hopkins Bayview Medical Center, also in Baltimore.⁵ We instructed residents to obtain discharge buprenorphine prescriptions from an X-waivered physician on their team or from the newly established BBT. We asked resident teams to set up a postdischarge appointment for patients with an X-waivered provider, either in a community practice or at the JHH After Care Clinic, a transitional care clinic for discharged patients.²¹

The BBT is a resident-led group of X-waivered JHH residents and hospitalists who volunteer to write discharge buprenorphine prescriptions for patients. The BBT serves to ensure primary teams have access to an X-waivered prescriber. It is not a consult service. We asked primary teams to contact the BBT after initiating buprenorphine and after securing a follow-up appointment. In response to each request, a member of the BBT reviews the patient chart, confirms the follow-up plan, writes a prescription for buprenorphine along with intranasal naloxone, and leaves a brief note. During the 6-month postintervention period, the team consisted of three residents and three hospitalist attendings. Each week, two members (residents or attendings) staffed the team Monday to Friday, 8 am to 5 pm. Most weeks were staffed by two residents. One resident provided services after hours and during weekends. Resident team members ensured that the buprenorphine plan was discussed with the primary team’s attending. For dosing questions beyond the BBT’s scope of knowledge, a member of the BBT relayed questions to physicians from the Addiction Medicine Consult team at Johns Hopkins Bayview Medical Center.

In May 2019, 5 months after the education session and implementation of the BBT, we administered a follow-up survey.

Outcomes

The primary outcome was the percent of inpatients eligible to start OAT who were discharged on buprenorphine maintenance. We obtained data from the electronic medical record. The denominator consisted of patients with OUD not on buprenorphine or methadone maintenance on admission. We identified patients with OUD by an opioid-related International Classification of Diseases, Tenth Revision (ICD-10) diagnosis code or by a standing or as-needed order for buprenorphine or methadone during hospitalization.²² We reviewed admission and discharge documentation to identify patients with OUD who were not in active treatment with buprenorphine or methadone maintenance.

As a secondary outcome, we measured engagement in OUD treatment after discharge by calculating the proportion of patients started on buprenorphine who filled a buprenorphine prescription within 30 days after discharge. We chose 30 days based on the National Committee for Quality Assurance’s Healthcare Effectiveness Data and Information Set (HEDIS) measure for engagement of treatment for alcohol and other drugs.²³ We obtained the data from the Chesapeake Regional Information System for our Patients (CRISP) Prescription Drug Monitoring Program, which monitors all prescriptions for controlled substances dispensed in Maryland and five neighboring states. As a balancing measure, we counted patients newly started on methadone maintenance for OUD before and after the intervention. Additional secondary process outcomes included frequency of BBT requests, the volume of buprenorphine prescriptions written by the team, and time required to complete a BBT request.

Clinician-level outcomes, measured with electronically administered pre- and postintervention surveys to residents, included knowledge about and comfort with buprenorphine. Of the 16 questions in the pre- and postimplementation surveys, we analyzed the 6 questions concerning knowledge and comfort that remained identical in the pre- and postintervention surveys and used 5-point Likert scale responses. As an incentive, we randomly distributed three $50 gift cards to survey completers.

Analysis

We used an interrupted time series analysis to evaluate the association between the intervention bundle and a change in the rate that medical teams started patients with OUD on buprenorphine maintenance. This approach allowed us to control for preintervention trends. To evaluate the impact of our interventions, our pre- and postintervention periods include the same residents during the 2018-2019 academic year. Both periods consisted of twelve 2-week intervals (preintervention: July 26, 2019, to January 9, 2019; postintervention: January 10, 2019, to June 26, 2019).

To evaluate for changes in engagement in OUD treatment after discharge, we used two-sample t tests. To evaluate for changes in resident-reported comfort and knowledge with initiating buprenorphine maintenance, we used Wilcoxon rank sum tests for survey data and Wilcoxon signed rank tests for paired data. All analyses employed two-sided P values with statistical significance evaluated at the .05 alpha level. We analyzed data using R version 3.6.3 (Foundation for Statistical Computing). The Institutional Review Board at JHH reviewed and approved the study protocol as a quality improvement project (IRB00193365).

During the 24-week preintervention period, internal medicine resident teams started 30 out of 305 eligible patients (10%) on buprenorphine maintenance vs 64 out of 270 eligible patients (24%) during the 24-week postintervention period. Our interrupted time series analysis showed a significant increase in the percent of eligible patients started on buprenorphine maintenance (expected number of patients started postintervention, 27; actual, 64; absolute increase in percent, 14.4%; 95% CI, 3.6%-25.3%; P = .017) (Figure). There was no significant trend during the preintervention period and no significant trend during the postintervention period.

Before the intervention, 13 of the 30 patients (40%) newly started on buprenorphine maintenance during their admission filled a follow-up buprenorphine prescription within 30 days of discharge. After the intervention, 31 of 64 patients (46%) filled a buprenorphine prescription within 30 days (P = .612). Two patients were started on methadone maintenance, one prior to and one after the intervention.

During the 6-month postintervention period, the BBT received 75 requests and wrote 70 prescriptions for buprenorphine. The median time required to complete a BBT request was 15 minutes (minimum, 5 minutes; maximum, 60 minutes).

Of 156 internal medicine and medicine-pediatrics residents, 89 residents (57%) completed the baseline survey and 66 residents (42%) completed the follow-up survey. Forty residents completed both surveys. After the intervention, residents were significantly more likely to feel comfortable dosing buprenorphine (P < .0001) and counseling patients about its use (P = .0237) and were more likely to report ease of establishing follow-up (P < .0001). Self-reported knowledge about preventing precipitated withdrawal increased significantly (P = .0191), as did knowledge about the effectiveness of buprenorphine (P = .0003) independent of formal drug counseling (P = .0066) (Table). Paired survey data also found statistically significant results for all questions except those about preventing precipitated withdrawal and efficacy. For the latter, respondents who completed both surveys were more knowledgeable before the intervention than the overall group that completed the baseline survey (Appendix Table).

This study shows how a resident-led quality improvement project comprising clinician education and implementation of a novel BBT was associated with an increased rate of starting buprenorphine maintenance in hospitalized patients with OUD and improved resident knowledge about and comfort with buprenorphine. To our knowledge, this is the first study demonstrating how education and a team of X-waivered generalists can help primary teams initiate and discharge patients on buprenorphine maintenance in a hospital without an addiction medicine consult service.

Prior to the intervention, resident internal medicine teams at JHH started 10% of hospitalized patients with OUD on buprenorphine maintenance, consistent with prior studies showing rates of 11% to 15% for initiating OAT for hospitalized patients.^11,12 After the intervention, the rate of initiating buprenorphine maintenance more than doubled, rising to 24% of eligible patients. Resident internal medicine teams at JHH started buprenorphine maintenance for 37 more patients over the 24-week postintervention period than would have been predicted prior to the intervention, or an additional three patients every 2 weeks.

Between 40% and 46% of hospitalized patients newly started on buprenorphine maintenance filled an outpatient buprenorphine prescription within 30 days of discharge. We are not aware of comparative data for 30-day follow-up for hospitalized patients newly started on buprenorphine maintenance. Data from other contexts show 5% to 10% of veterans were engaged in addiction treatment 30 days after initiation from inpatient or outpatient encounters. An analysis of an academic medical center in Oregon found engagement with an addiction medicine consult service increased after hospital engagement for patients with any substance use disorder from 23% to 39% using the 34-day HEDIS measure for engagement.^17,24,25

The BBT required approximately 15 minutes per request and wrote an average of three prescriptions per week, demonstrating the feasibility of this approach and the high demand for this service. One strength of our approach is that residents gained experience starting buprenorphine independently using the aforementioned protocol instead of deferring to a full consult service. It is likely that this resident engagement in initiating longitudinal OUD care contributed to the success of this initiative, as did existing resident familiarity with using buprenorphine for opioid withdrawal.

This approach to resident education—promoting direct, first-person experience with medications in a clinical context—aligns with recommendations from a recent review about substance use disorder education for health professionals.²⁶ Our interventions increased resident knowledge and comfort with buprenorphine, consistent with prior studies showing increased resident confidence in management of substance use disorders after curricular innovations.^24,25

A few contextual features were essential for this project’s viability. Maryland allows American medical graduates to obtain a medical license after 1 year of postgraduate training. This allowed three residents to obtain X-waivers. These residents had access to HRSA funding to subsidize the expenses of applying for state licensure and DEA registration. BBT members volunteered their time while working on other services. Last, we were able to take advantage of buprenorphine-providing clinics in Baltimore, including the JHH After Care Clinic, to accept patients for follow-up appointments after discharge.

Limitations

The BBT required motivated clinicians willing to volunteer for additional clinical responsibilities during inpatient rotations and supportive faculty and residency leadership. Attending physicians, nurse practitioners, or physician assistants could staff a similar BBT in hospitals without residents or in hospitals where residents cannot obtain DEA registration. Crucially, other hospitals may not have access to practices with X-waivered physicians for outpatient follow-up. A recent study found X-waivered primary care physicians were less likely to be affiliated with hospital health systems. Other studies have shown limitations in access to buprenorphine at the county level based on geography and racial/ethnic segregation.^27-29

Most patients hospitalized with OUD did not have ICD-10 codes associated with OUD. We addressed this by assuming patients had OUD if buprenorphine or methadone was ordered during their hospitalization, even if the medication was never administered. This may have overcounted patients prescribed these medications for indications other than OUD, and it may have undercounted patients with OUD for whom buprenorphine or methadone were never considered. The opioid withdrawal order set at JHH automatically offers an option to use buprenorphine to treat withdrawal. Patients with OUD for whom buprenorphine or methadone were never ordered likely did not experience withdrawal or were in withdrawal so mild that it escaped the attention of the team, which limits the generalizability of our intervention.

We identified several limitations to the internal validity of our study. First, we used a before-and-after study design without a control group. We could not ethically withhold access to evidence-based, mortality-reducing medications from patients. Without a control group, we cannot rule out the possibility that underlying temporal trends made residents more likely to start buprenorphine maintenance independent of our intervention. We attempted to control for unmeasured confounders by using an interrupted time series analysis to control for preintervention trends, comparing the same group of residents before and after our interventions, and selecting an intervention period during which residents were given only educational sessions and materials provided by our team. Our results may be biased by clustered data because certain residents may have been more likely to initiate buprenorphine, but these effects are likely marginal because resident schedules are balanced between outpatient and inpatient rotations during each 6-month period.

Finally, this project focused on buprenorphine, not on other medications for OUD, including methadone or naltrexone, or nonpharmacologic treatments for OUD.

Since the start of the BBT in January 2019, five additional PGY-2 residents obtained their medical licenses and X-waivers. These residents, with the support of two attending hospitalists, led the BBT and coordinated education sessions that were incorporated into the curriculum during the 2019-2020 academic year. These educational sessions will continue indefinitely. In 2020, JHH started an Addiction Medicine Consult Service staffed by physicians, NPs, and a pharmacist. The BBT continues to operate in conjunction with this service.

We found substantial variability in the rate of buprenorphine maintenance initiation despite our interventions. This is an area for future improvement. In a free-response prompt in our follow-up survey, residents requested additional education sessions and an order set to assist with initiation of buprenorphine. To address these gaps, three educational sessions were added, one of which included education on starting methadone maintenance therapy. We also added a new order set for starting buprenorphine maintenance. We hypothesize that these interventions will improve consistency.

In order for a similar program to be disseminated to other institutions, educational initiatives and a team of dedicated X-waivered prescribers are key. Materials to assist with this process are available in the Appendix.

This study shows how a resident-led intervention comprising clinician education and a team of X-waivered generalists was associated with improved treatment of OUD for hospitalized patients. We encourage residents and all clinicians at other hospitals without addiction medicine consult services to design, implement, and study similar interventions that directly increase the use of buprenorphine or methadone maintenance to treat OUD.

Preliminary results from this project were presented at the AMERSA National Conference on November 7, 2019.

Nearly 48,000 Americans died from overdoses involving opioids in 2018, continuing a national crisis that has led to 446,000 deaths since 1999.¹ Annually, opioids are responsible for more than 500,000 admissions, approximately 1% of all hospitalizations, costing the United States nearly $15 billion.^2,3 Among hospitalized patients, chronic opioid use is associated with increased mortality, severe infectious complications, and higher rates of readmission.⁴ Opioid use disorder (OUD) is a chronic, relapsing medical condition with biopsychosocial origins and significant morbidity and mortality.⁵ Opioid agonist therapy (OAT) with buprenorphine or methadone maintenance, the evidence-based standard of treatment, reduces the mortality rate by half, decreases overdoses and hospital readmissions, and improves retention in care.^6-10

OAT maintenance refers to using buprenorphine or methadone for long-term treatment of OUD rather than for acute treatment of opioid withdrawal. Despite evidence supporting OAT maintenance, clinicians start medications for only 11% to 15% of hospitalized patients with OUD, depending on practice contexts.^11,12 Three significant barriers—stigma, insufficient clinician education, and restrictive regulations—prevent clinicians from starting OAT.¹³ Clinicians who do not have the Drug Enforcement Administration (DEA)–issued DATA-2000 waiver (X-waiver) for outpatient prescribing can order buprenorphine for admitted patients but cannot prescribe it at discharge.¹⁴ In hospitals where they exist, addiction medicine consult services offer primary teams guidance on pharmacotherapy, leading to reduced hospital readmissions and increased engagement in outpatient addiction treatment.^15-17 However, in most hospitals around the country, such specialty services do not exist.¹⁸ In some hospitals without addiction medicine consult services, hospitalists with expertise in OUD have started assisting primary teams in starting OAT, but to our knowledge, no prior studies have described the impact of these interventions on patients or clinician experience with OAT.¹⁹

This quality improvement project aimed to increase the rate at which internal medicine resident teams at Johns Hopkins Hospital (JHH) in Baltimore, Maryland, started hospitalized patients with OUD on buprenorphine maintenance. We hypothesized that resident education and measures to increase the availability of X-waivered physicians would increase the rate of initiating buprenorphine maintenance. We additionally hypothesized that these interventions would increase knowledge about and comfort with buprenorphine across the residency. This represents the first study to examine the effects of clinician education and a team of X-waivered residents and hospitalists who assist in starting buprenorphine maintenance in a hospital without an addiction medicine consult service.

Setting

This study took place from July 2018 to June 2019 at JHH, a large, academic, urban hospital in Baltimore. Prior to the intervention, internal medicine residents at JHH commonly used short courses of buprenorphine to treat withdrawal, but they did not have access to hospital-specific resources to assist with starting maintenance OAT. During the study period, JHH had a Substance Use Disorders team staffed by peer recovery specialists that could be consulted by hospitalists and residents to provide psychosocial support and link admitted patients to treatment after discharge. There were no providers on the team to guide pharmacotherapy or to write discharge buprenorphine prescriptions. The Osler Medical Residency Training Program at JHH has 140 internal medicine residents and 16 combined medicine-pediatrics residents. All residents receive 1 hour of formal education about opioid use disorder annually. In addition, 28 of those 156 residents, those in the Urban Health Primary Care track, spend 1 month on an Addiction Medicine rotation in which they complete the 8-hour training required to receive the X-waiver. Those residents are encouraged to apply for the X-waiver once they obtain a medical license subsidized by a Health Resources & Services Administration (HRSA) grant. Four internal medicine attending physicians on teaching services and one resident had X-waivers prior to the intervention.

Intervention

In November 2018, we administered a survey to residents to identify barriers to starting buprenorphine maintenance and to measure knowledge and confidence with using buprenorphine for OUD (Appendix Figure 1 and Figure 2). We focused on buprenorphine because providers at JHH were familiar with this medication and because Baltimore has widespread access to buprenorphine, with more than 490 local buprenorphine providers.²⁰ Five residents piloted the survey and provided feedback. We then administered the survey to all internal medicine and medicine-pediatrics residents. Based on the results, we developed a targeted educational conference and also created the Buprenorphine Bridge Team (BBT).

In January 2019, we presented the educational conference for residents devoted to the use of buprenorphine for OUD and introduced the BBT. The conference started with a patient testimonial and included peer recovery specialists, pharmacists, nurses, and social workers. We summarized the evidence for buprenorphine and offered a practical guide to start treatment in a one-page protocol. This protocol included guidance on selecting patients, shared decision-making around OUD treatment, avoiding precipitated withdrawal, dosing buprenorphine, and establishing follow-up (Appendix Figure 3). We asked for input on this protocol from nursing leadership, social work teams, and peer recovery specialists. Dosing was adapted from the Guidelines from the American Society of Addiction Medicine, with expert input from physicians from the Addiction Medicine Consult service at Johns Hopkins Bayview Medical Center, also in Baltimore.⁵ We instructed residents to obtain discharge buprenorphine prescriptions from an X-waivered physician on their team or from the newly established BBT. We asked resident teams to set up a postdischarge appointment for patients with an X-waivered provider, either in a community practice or at the JHH After Care Clinic, a transitional care clinic for discharged patients.²¹

The BBT is a resident-led group of X-waivered JHH residents and hospitalists who volunteer to write discharge buprenorphine prescriptions for patients. The BBT serves to ensure primary teams have access to an X-waivered prescriber. It is not a consult service. We asked primary teams to contact the BBT after initiating buprenorphine and after securing a follow-up appointment. In response to each request, a member of the BBT reviews the patient chart, confirms the follow-up plan, writes a prescription for buprenorphine along with intranasal naloxone, and leaves a brief note. During the 6-month postintervention period, the team consisted of three residents and three hospitalist attendings. Each week, two members (residents or attendings) staffed the team Monday to Friday, 8 am to 5 pm. Most weeks were staffed by two residents. One resident provided services after hours and during weekends. Resident team members ensured that the buprenorphine plan was discussed with the primary team’s attending. For dosing questions beyond the BBT’s scope of knowledge, a member of the BBT relayed questions to physicians from the Addiction Medicine Consult team at Johns Hopkins Bayview Medical Center.

In May 2019, 5 months after the education session and implementation of the BBT, we administered a follow-up survey.

Outcomes

The primary outcome was the percent of inpatients eligible to start OAT who were discharged on buprenorphine maintenance. We obtained data from the electronic medical record. The denominator consisted of patients with OUD not on buprenorphine or methadone maintenance on admission. We identified patients with OUD by an opioid-related International Classification of Diseases, Tenth Revision (ICD-10) diagnosis code or by a standing or as-needed order for buprenorphine or methadone during hospitalization.²² We reviewed admission and discharge documentation to identify patients with OUD who were not in active treatment with buprenorphine or methadone maintenance.

As a secondary outcome, we measured engagement in OUD treatment after discharge by calculating the proportion of patients started on buprenorphine who filled a buprenorphine prescription within 30 days after discharge. We chose 30 days based on the National Committee for Quality Assurance’s Healthcare Effectiveness Data and Information Set (HEDIS) measure for engagement of treatment for alcohol and other drugs.²³ We obtained the data from the Chesapeake Regional Information System for our Patients (CRISP) Prescription Drug Monitoring Program, which monitors all prescriptions for controlled substances dispensed in Maryland and five neighboring states. As a balancing measure, we counted patients newly started on methadone maintenance for OUD before and after the intervention. Additional secondary process outcomes included frequency of BBT requests, the volume of buprenorphine prescriptions written by the team, and time required to complete a BBT request.

Clinician-level outcomes, measured with electronically administered pre- and postintervention surveys to residents, included knowledge about and comfort with buprenorphine. Of the 16 questions in the pre- and postimplementation surveys, we analyzed the 6 questions concerning knowledge and comfort that remained identical in the pre- and postintervention surveys and used 5-point Likert scale responses. As an incentive, we randomly distributed three $50 gift cards to survey completers.

Analysis

We used an interrupted time series analysis to evaluate the association between the intervention bundle and a change in the rate that medical teams started patients with OUD on buprenorphine maintenance. This approach allowed us to control for preintervention trends. To evaluate the impact of our interventions, our pre- and postintervention periods include the same residents during the 2018-2019 academic year. Both periods consisted of twelve 2-week intervals (preintervention: July 26, 2019, to January 9, 2019; postintervention: January 10, 2019, to June 26, 2019).

To evaluate for changes in engagement in OUD treatment after discharge, we used two-sample t tests. To evaluate for changes in resident-reported comfort and knowledge with initiating buprenorphine maintenance, we used Wilcoxon rank sum tests for survey data and Wilcoxon signed rank tests for paired data. All analyses employed two-sided P values with statistical significance evaluated at the .05 alpha level. We analyzed data using R version 3.6.3 (Foundation for Statistical Computing). The Institutional Review Board at JHH reviewed and approved the study protocol as a quality improvement project (IRB00193365).

During the 24-week preintervention period, internal medicine resident teams started 30 out of 305 eligible patients (10%) on buprenorphine maintenance vs 64 out of 270 eligible patients (24%) during the 24-week postintervention period. Our interrupted time series analysis showed a significant increase in the percent of eligible patients started on buprenorphine maintenance (expected number of patients started postintervention, 27; actual, 64; absolute increase in percent, 14.4%; 95% CI, 3.6%-25.3%; P = .017) (Figure). There was no significant trend during the preintervention period and no significant trend during the postintervention period.

Before the intervention, 13 of the 30 patients (40%) newly started on buprenorphine maintenance during their admission filled a follow-up buprenorphine prescription within 30 days of discharge. After the intervention, 31 of 64 patients (46%) filled a buprenorphine prescription within 30 days (P = .612). Two patients were started on methadone maintenance, one prior to and one after the intervention.

During the 6-month postintervention period, the BBT received 75 requests and wrote 70 prescriptions for buprenorphine. The median time required to complete a BBT request was 15 minutes (minimum, 5 minutes; maximum, 60 minutes).

Of 156 internal medicine and medicine-pediatrics residents, 89 residents (57%) completed the baseline survey and 66 residents (42%) completed the follow-up survey. Forty residents completed both surveys. After the intervention, residents were significantly more likely to feel comfortable dosing buprenorphine (P < .0001) and counseling patients about its use (P = .0237) and were more likely to report ease of establishing follow-up (P < .0001). Self-reported knowledge about preventing precipitated withdrawal increased significantly (P = .0191), as did knowledge about the effectiveness of buprenorphine (P = .0003) independent of formal drug counseling (P = .0066) (Table). Paired survey data also found statistically significant results for all questions except those about preventing precipitated withdrawal and efficacy. For the latter, respondents who completed both surveys were more knowledgeable before the intervention than the overall group that completed the baseline survey (Appendix Table).

This study shows how a resident-led quality improvement project comprising clinician education and implementation of a novel BBT was associated with an increased rate of starting buprenorphine maintenance in hospitalized patients with OUD and improved resident knowledge about and comfort with buprenorphine. To our knowledge, this is the first study demonstrating how education and a team of X-waivered generalists can help primary teams initiate and discharge patients on buprenorphine maintenance in a hospital without an addiction medicine consult service.

Prior to the intervention, resident internal medicine teams at JHH started 10% of hospitalized patients with OUD on buprenorphine maintenance, consistent with prior studies showing rates of 11% to 15% for initiating OAT for hospitalized patients.^11,12 After the intervention, the rate of initiating buprenorphine maintenance more than doubled, rising to 24% of eligible patients. Resident internal medicine teams at JHH started buprenorphine maintenance for 37 more patients over the 24-week postintervention period than would have been predicted prior to the intervention, or an additional three patients every 2 weeks.

Between 40% and 46% of hospitalized patients newly started on buprenorphine maintenance filled an outpatient buprenorphine prescription within 30 days of discharge. We are not aware of comparative data for 30-day follow-up for hospitalized patients newly started on buprenorphine maintenance. Data from other contexts show 5% to 10% of veterans were engaged in addiction treatment 30 days after initiation from inpatient or outpatient encounters. An analysis of an academic medical center in Oregon found engagement with an addiction medicine consult service increased after hospital engagement for patients with any substance use disorder from 23% to 39% using the 34-day HEDIS measure for engagement.^17,24,25

The BBT required approximately 15 minutes per request and wrote an average of three prescriptions per week, demonstrating the feasibility of this approach and the high demand for this service. One strength of our approach is that residents gained experience starting buprenorphine independently using the aforementioned protocol instead of deferring to a full consult service. It is likely that this resident engagement in initiating longitudinal OUD care contributed to the success of this initiative, as did existing resident familiarity with using buprenorphine for opioid withdrawal.

This approach to resident education—promoting direct, first-person experience with medications in a clinical context—aligns with recommendations from a recent review about substance use disorder education for health professionals.²⁶ Our interventions increased resident knowledge and comfort with buprenorphine, consistent with prior studies showing increased resident confidence in management of substance use disorders after curricular innovations.^24,25

A few contextual features were essential for this project’s viability. Maryland allows American medical graduates to obtain a medical license after 1 year of postgraduate training. This allowed three residents to obtain X-waivers. These residents had access to HRSA funding to subsidize the expenses of applying for state licensure and DEA registration. BBT members volunteered their time while working on other services. Last, we were able to take advantage of buprenorphine-providing clinics in Baltimore, including the JHH After Care Clinic, to accept patients for follow-up appointments after discharge.

Limitations

The BBT required motivated clinicians willing to volunteer for additional clinical responsibilities during inpatient rotations and supportive faculty and residency leadership. Attending physicians, nurse practitioners, or physician assistants could staff a similar BBT in hospitals without residents or in hospitals where residents cannot obtain DEA registration. Crucially, other hospitals may not have access to practices with X-waivered physicians for outpatient follow-up. A recent study found X-waivered primary care physicians were less likely to be affiliated with hospital health systems. Other studies have shown limitations in access to buprenorphine at the county level based on geography and racial/ethnic segregation.^27-29

Most patients hospitalized with OUD did not have ICD-10 codes associated with OUD. We addressed this by assuming patients had OUD if buprenorphine or methadone was ordered during their hospitalization, even if the medication was never administered. This may have overcounted patients prescribed these medications for indications other than OUD, and it may have undercounted patients with OUD for whom buprenorphine or methadone were never considered. The opioid withdrawal order set at JHH automatically offers an option to use buprenorphine to treat withdrawal. Patients with OUD for whom buprenorphine or methadone were never ordered likely did not experience withdrawal or were in withdrawal so mild that it escaped the attention of the team, which limits the generalizability of our intervention.

We identified several limitations to the internal validity of our study. First, we used a before-and-after study design without a control group. We could not ethically withhold access to evidence-based, mortality-reducing medications from patients. Without a control group, we cannot rule out the possibility that underlying temporal trends made residents more likely to start buprenorphine maintenance independent of our intervention. We attempted to control for unmeasured confounders by using an interrupted time series analysis to control for preintervention trends, comparing the same group of residents before and after our interventions, and selecting an intervention period during which residents were given only educational sessions and materials provided by our team. Our results may be biased by clustered data because certain residents may have been more likely to initiate buprenorphine, but these effects are likely marginal because resident schedules are balanced between outpatient and inpatient rotations during each 6-month period.

Finally, this project focused on buprenorphine, not on other medications for OUD, including methadone or naltrexone, or nonpharmacologic treatments for OUD.

Since the start of the BBT in January 2019, five additional PGY-2 residents obtained their medical licenses and X-waivers. These residents, with the support of two attending hospitalists, led the BBT and coordinated education sessions that were incorporated into the curriculum during the 2019-2020 academic year. These educational sessions will continue indefinitely. In 2020, JHH started an Addiction Medicine Consult Service staffed by physicians, NPs, and a pharmacist. The BBT continues to operate in conjunction with this service.

We found substantial variability in the rate of buprenorphine maintenance initiation despite our interventions. This is an area for future improvement. In a free-response prompt in our follow-up survey, residents requested additional education sessions and an order set to assist with initiation of buprenorphine. To address these gaps, three educational sessions were added, one of which included education on starting methadone maintenance therapy. We also added a new order set for starting buprenorphine maintenance. We hypothesize that these interventions will improve consistency.

In order for a similar program to be disseminated to other institutions, educational initiatives and a team of dedicated X-waivered prescribers are key. Materials to assist with this process are available in the Appendix.

This study shows how a resident-led intervention comprising clinician education and a team of X-waivered generalists was associated with improved treatment of OUD for hospitalized patients. We encourage residents and all clinicians at other hospitals without addiction medicine consult services to design, implement, and study similar interventions that directly increase the use of buprenorphine or methadone maintenance to treat OUD.

Preliminary results from this project were presented at the AMERSA National Conference on November 7, 2019.

Hospitalizations related to opioid use disorder (OUD) have increased and now account for up to 6% of hospital admissions in certain areas of the United States.¹ Patients with OUD who are started on buprenorphine during hospitalization are more likely to enter outpatient treatment, stay in treatment longer, and have more drug-free days compared with patients who only receive a referral for outpatient treatment.^2,3 Therefore, a crucial comprehensive strategy for OUD care should include hospital-based programs that support initiation of treatment in the inpatient setting and strong bridges to outpatient care. One of the common barriers to initiating treatment in the inpatient setting, however, is a lack of access to addiction medicine specialists.^4-6

In 2017, we created a hospitalist-led interprofessional team called the B-Team (Buprenorphine Team) to help primary care teams identify patients with OUD, initiate and maintain buprenorphine therapy during hospitalization, provide warm handoffs to outpatient treatment programs, and reduce institutional stigma related to people with substance use disorders.

Program Description

The B-Team is led by a hospital medicine physician assistant and includes physicians from internal medicine, consult-liaison psychiatry, and palliative care; advanced practice and bedside nurses; a social worker; a pharmacist; a chaplain; a peer-recovery specialist; and medical trainees. The B-Team is notified of potential candidates for buprenorphine through a secure texting platform, one that is accessible to any healthcare provider at the hospital. Patients who are referred to the B-Team either self-identify or are identified by their primary team as having an underlying OUD. One of the B-Team providers assesses the patient to determine if they are eligible to receive inpatient therapy. Patients are considered eligible for the program if they meet Diagnostic and Statistical Manual of Mental Disorders (5th edition) criteria for OUD, have a desire to cease opioid use, and receive medical clearance to take buprenorphine.

For eligible patients, the B-Team provider orders a nurse-driven protocol to initiate buprenorphine for OUD. The chaplain offers psychospiritual counseling, and the social worker provides counseling and coordination of care. The B-Team partners with a nonhospital-affiliated, publicly-funded, office-based opioid treatment (OBOT) program that combines primary care with behavioral health programming. A follow-up outpatient appointment is secured prior to hospital discharge, and a member of the B-Team who has Drug Addiction Treatment Act of 2000 (DATA 2000) X-waiver certification prescribes buprenorphine as a bridge until the follow-up appointment. The medication is dispensed from the hospital’s retail pharmacy, and the patient leaves the hospital with the medication in-hand.

Patients who are not eligible for buprenorphine therapy are offered a harm-reduction intervention or referral to the psychiatry consult liaison service to assess for alternative diagnoses or treatment. These patients are also offered psychospiritual counseling and a prescription for naloxone.

Prior to the creation of the B-Team at our hospital, there was no structure in place to facilitate initiation of buprenorphine therapy during hospitalization and no linkage to outpatient treatment after discharge; furthermore, none of the hospitalists or other providers (including consulting psychiatrists) had an X-waiver to prescribe buprenorphine for OUD.

Program Evaluation

Study data were collected using Research Electronic Data Capture software. Inpatient and outpatient data were entered by a B-Team provider or a researcher via chart review. Patients were considered to be engaged in care if they attended at least one outpatient appointment for buprenorphine therapy during each of the following time periods: (1) 0 to 27 days (initial follow-up), 28-89 days (1- to 3-month follow-up), 90-179 days (3- to 6-month follow-up), and 180 days or more (>6-month follow-up). Only visits specifically for buprenorphine maintenance therapy were counted. If multiple encounters occurred within one time frame, the encounter closest to 0, 30, 90, or 180 days from discharge was used. If a patient did not attend any encounters during a specified time frame, they were considered to no longer be engaged in care and were no longer tracked for purposes of the evaluation. Data for the percentage of patients engaged in outpatient care are presented as the number of patients who attended at least one appointment during each of the follow-up periods (1 to 3 months, 3 to 6 months, or after 6 months, as noted above) divided by the number of patients who had been discharged with coordinated follow-up.

The number of patients admitted per month for whom there was an order to initiate inpatient buprenorphine therapy was analyzed using a statistical process control chart, in addition to the number of OUD admissions based on the inclusion of the International Classification of Disease, Tenth Revision (ICD-10) F11 code (opioid-related disorders) in any position in the discharge diagnoses.

This program and study were considered quality improvement by The University of Texas Institutional Review Board and did not meet criteria for human subjects research.

RESULTS

During the first 2 years of the program (September 2018-September 2020), the B-Team received 260 patient referrals. Most of the patients were White (72%), male (62%), and between ages 25 and 44 years (53%) (Appendix Table). The team initiated buprenorphine therapy in 132 hospitalized patients. In the year prior to the creation of the B-Team program, the average number of hospitalized patients receiving buprenorphine for OUD per month was three; after the launch of the B-Team program, this number increased to 12 encounters per month (Figure 1A). The sudden decrease observed in August 2020 is likely related to a surge in COVID-19 admissions. The number of monthly admissions for OUD is also shown (Figure 1B).

The B-Team saw a total of 132 eligible patients; members of the team provided counseling, support, and resources regarding buprenorphine therapy. In addition, the B-Team’s chaplain provided emotional support and spiritual connection (if desired) to 40 of these patients (30%). In the study, no cases of precipitated withdrawal were identified. Of the 132 patients seen, 110 (83%) were accepted to an outpatient OUD program upon discharge from the hospital; 98 (89%) of these patients were accepted at our partner OBOT clinic. The remaining patients were not interested in continuing OUD treatment (13%) or were denied acceptance to an outpatient program based on administrative and/or financial eligibility guidelines (4%). Patients who would not be attending an outpatient program were discontinued on buprenorphine therapy prior to discharge, counseled about naloxone, and provided printed resources.

Outpatient appointment attendance was used to measure ongoing treatment engagement of the 110 patients who were discharged with coordinated follow-up care. A total of 65 patients (59%) attended their first outpatient appointment; the average time between discharge and the first outpatient appointment was 5.9 days. Forty-two patients (38%) attended at least one appointment between 1 and 3 months; 29 (26%) between 3 and 6 months; and 24 (22%) after 6 months (Figure 2).

Of the 128 patients who were not administered buprenorphine therapy, 64 (50%) were not interested in starting treatment and/or were not ready to engage in treatment; 36 (28%) did not meet criteria for OUD treatment; 28 (22%) were already receiving treatment or preferred another type of OUD treatment; and 13 (10%) had severe comorbid addiction and/or illness requiring treatment that contraindicates the use of buprenorphine.

A volunteer hospitalist-led interprofessional team providing evidence-based care for hospitalized patients with OUD was associated with a substantial increase in patients receiving buprenorphine therapy—both during hospitalization and after discharge. In the program, 59% of patients attended initial follow-up appointments, and 22% of patients were still engaged at 6 months. These outpatient follow-up rates appear to be similar to, or higher than, other programs described in the literature. For example, a buprenorphine OUD-treatment initiative led by the psychiatry consult service at a Boston academic medical center resulted in less than half of patients receiving buprenorphine treatment within 2 months of discharge.⁷ In another study wherein an addiction medicine consult service administered buprenorphine to patients with OUD during hospitalization, 39%, 27%, and 18% of patients were retained in outpatient treatment at 30, 90, and 180 days, respectively.⁸

The B-Team model is likely generalizable to other hospital medicine groups that may not otherwise have access to inpatient care for substance use disorder. The B-Team is not an addiction medicine consultation service; rather, it is a hospitalist-led quality improvement initiative seeking to improve the standard of care for hospitalized patients with OUD.

A significant barrier is ensuring ongoing support for patients with OUD after discharge. In the B-Team program, a parallel OBOT program was created by a local nonaffiliated federally qualified health center. Although 89% of patients received treatment at this OBOT clinic, the inpatient team also has relationships with other local treatment centers, including programs that provide methadone. Another important barrier to high-quality outpatient care for OUD is the requirement of an X-waiver. To help overcome this barrier, our inpatient program partnered with a regional medical society to offer periodic X-waiver training to outpatient providers. In less than a year, more than 100 regional prescribers participated in this program.

Our study has several limitations. There was likely some degree of selection bias among the hospitalized patients who received initial buprenorphine treatment. To our knowledge, there is no specific validated screening tool for OUD in the inpatient acute care setting; moreover, we have been unable to implement standardized screening for OUD into the electronic health record. As such, we rely on the totality of the clinical circumstances approach to identify patients with OUD.

Furthermore, we had neither a comparison group nor a prospective plan to follow patients who did not remain engaged in care after discharge. In addition, our analysis of OUD admissions included F11 ICD-10 codes, which are limited by clinical documentation.^9,10 Our program focuses exclusively on buprenorphine initiation due to insufficient immediate outpatient capacity for methadone initiated during hospitalization and lack of coverage for extended-release naltrexone. Limitations to outpatient data-sharing prevented the reporting of outpatient appointments external to the identified partner program; since these appointments were included in the analysis as “lost to follow-up,” actual engagement rates may be higher than those reported.

Moving forward, the B-Team is continuing to serve as a role model for appropriate, patient-centered, evidence-based care for hospitalized patients with OUD. Attending physicians and residents with an X-waiver are now encouraged to initiate buprenorphine treatment on their own. In June 2020, we added peer-recovery support services to the program, which has improved care for patients and increased adoption of hospital-initiated substance use disorder interventions.¹¹ Lessons learned from inpatient implementation are being applied to our hospital’s emergency department and to an inpatient obstetrics unit at a partner hospital; they are also being employed to further empower hospitalists to diagnose and treat other substance use disorders, such as alcohol use disorder.

Hospitalizations related to opioid use disorder (OUD) have increased and now account for up to 6% of hospital admissions in certain areas of the United States.¹ Patients with OUD who are started on buprenorphine during hospitalization are more likely to enter outpatient treatment, stay in treatment longer, and have more drug-free days compared with patients who only receive a referral for outpatient treatment.^2,3 Therefore, a crucial comprehensive strategy for OUD care should include hospital-based programs that support initiation of treatment in the inpatient setting and strong bridges to outpatient care. One of the common barriers to initiating treatment in the inpatient setting, however, is a lack of access to addiction medicine specialists.^4-6

In 2017, we created a hospitalist-led interprofessional team called the B-Team (Buprenorphine Team) to help primary care teams identify patients with OUD, initiate and maintain buprenorphine therapy during hospitalization, provide warm handoffs to outpatient treatment programs, and reduce institutional stigma related to people with substance use disorders.

Program Description

The B-Team is led by a hospital medicine physician assistant and includes physicians from internal medicine, consult-liaison psychiatry, and palliative care; advanced practice and bedside nurses; a social worker; a pharmacist; a chaplain; a peer-recovery specialist; and medical trainees. The B-Team is notified of potential candidates for buprenorphine through a secure texting platform, one that is accessible to any healthcare provider at the hospital. Patients who are referred to the B-Team either self-identify or are identified by their primary team as having an underlying OUD. One of the B-Team providers assesses the patient to determine if they are eligible to receive inpatient therapy. Patients are considered eligible for the program if they meet Diagnostic and Statistical Manual of Mental Disorders (5th edition) criteria for OUD, have a desire to cease opioid use, and receive medical clearance to take buprenorphine.

For eligible patients, the B-Team provider orders a nurse-driven protocol to initiate buprenorphine for OUD. The chaplain offers psychospiritual counseling, and the social worker provides counseling and coordination of care. The B-Team partners with a nonhospital-affiliated, publicly-funded, office-based opioid treatment (OBOT) program that combines primary care with behavioral health programming. A follow-up outpatient appointment is secured prior to hospital discharge, and a member of the B-Team who has Drug Addiction Treatment Act of 2000 (DATA 2000) X-waiver certification prescribes buprenorphine as a bridge until the follow-up appointment. The medication is dispensed from the hospital’s retail pharmacy, and the patient leaves the hospital with the medication in-hand.

Patients who are not eligible for buprenorphine therapy are offered a harm-reduction intervention or referral to the psychiatry consult liaison service to assess for alternative diagnoses or treatment. These patients are also offered psychospiritual counseling and a prescription for naloxone.

Prior to the creation of the B-Team at our hospital, there was no structure in place to facilitate initiation of buprenorphine therapy during hospitalization and no linkage to outpatient treatment after discharge; furthermore, none of the hospitalists or other providers (including consulting psychiatrists) had an X-waiver to prescribe buprenorphine for OUD.

Program Evaluation

Study data were collected using Research Electronic Data Capture software. Inpatient and outpatient data were entered by a B-Team provider or a researcher via chart review. Patients were considered to be engaged in care if they attended at least one outpatient appointment for buprenorphine therapy during each of the following time periods: (1) 0 to 27 days (initial follow-up), 28-89 days (1- to 3-month follow-up), 90-179 days (3- to 6-month follow-up), and 180 days or more (>6-month follow-up). Only visits specifically for buprenorphine maintenance therapy were counted. If multiple encounters occurred within one time frame, the encounter closest to 0, 30, 90, or 180 days from discharge was used. If a patient did not attend any encounters during a specified time frame, they were considered to no longer be engaged in care and were no longer tracked for purposes of the evaluation. Data for the percentage of patients engaged in outpatient care are presented as the number of patients who attended at least one appointment during each of the follow-up periods (1 to 3 months, 3 to 6 months, or after 6 months, as noted above) divided by the number of patients who had been discharged with coordinated follow-up.

The number of patients admitted per month for whom there was an order to initiate inpatient buprenorphine therapy was analyzed using a statistical process control chart, in addition to the number of OUD admissions based on the inclusion of the International Classification of Disease, Tenth Revision (ICD-10) F11 code (opioid-related disorders) in any position in the discharge diagnoses.

This program and study were considered quality improvement by The University of Texas Institutional Review Board and did not meet criteria for human subjects research.

RESULTS

During the first 2 years of the program (September 2018-September 2020), the B-Team received 260 patient referrals. Most of the patients were White (72%), male (62%), and between ages 25 and 44 years (53%) (Appendix Table). The team initiated buprenorphine therapy in 132 hospitalized patients. In the year prior to the creation of the B-Team program, the average number of hospitalized patients receiving buprenorphine for OUD per month was three; after the launch of the B-Team program, this number increased to 12 encounters per month (Figure 1A). The sudden decrease observed in August 2020 is likely related to a surge in COVID-19 admissions. The number of monthly admissions for OUD is also shown (Figure 1B).

The B-Team saw a total of 132 eligible patients; members of the team provided counseling, support, and resources regarding buprenorphine therapy. In addition, the B-Team’s chaplain provided emotional support and spiritual connection (if desired) to 40 of these patients (30%). In the study, no cases of precipitated withdrawal were identified. Of the 132 patients seen, 110 (83%) were accepted to an outpatient OUD program upon discharge from the hospital; 98 (89%) of these patients were accepted at our partner OBOT clinic. The remaining patients were not interested in continuing OUD treatment (13%) or were denied acceptance to an outpatient program based on administrative and/or financial eligibility guidelines (4%). Patients who would not be attending an outpatient program were discontinued on buprenorphine therapy prior to discharge, counseled about naloxone, and provided printed resources.

Outpatient appointment attendance was used to measure ongoing treatment engagement of the 110 patients who were discharged with coordinated follow-up care. A total of 65 patients (59%) attended their first outpatient appointment; the average time between discharge and the first outpatient appointment was 5.9 days. Forty-two patients (38%) attended at least one appointment between 1 and 3 months; 29 (26%) between 3 and 6 months; and 24 (22%) after 6 months (Figure 2).

Of the 128 patients who were not administered buprenorphine therapy, 64 (50%) were not interested in starting treatment and/or were not ready to engage in treatment; 36 (28%) did not meet criteria for OUD treatment; 28 (22%) were already receiving treatment or preferred another type of OUD treatment; and 13 (10%) had severe comorbid addiction and/or illness requiring treatment that contraindicates the use of buprenorphine.

A volunteer hospitalist-led interprofessional team providing evidence-based care for hospitalized patients with OUD was associated with a substantial increase in patients receiving buprenorphine therapy—both during hospitalization and after discharge. In the program, 59% of patients attended initial follow-up appointments, and 22% of patients were still engaged at 6 months. These outpatient follow-up rates appear to be similar to, or higher than, other programs described in the literature. For example, a buprenorphine OUD-treatment initiative led by the psychiatry consult service at a Boston academic medical center resulted in less than half of patients receiving buprenorphine treatment within 2 months of discharge.⁷ In another study wherein an addiction medicine consult service administered buprenorphine to patients with OUD during hospitalization, 39%, 27%, and 18% of patients were retained in outpatient treatment at 30, 90, and 180 days, respectively.⁸

The B-Team model is likely generalizable to other hospital medicine groups that may not otherwise have access to inpatient care for substance use disorder. The B-Team is not an addiction medicine consultation service; rather, it is a hospitalist-led quality improvement initiative seeking to improve the standard of care for hospitalized patients with OUD.

A significant barrier is ensuring ongoing support for patients with OUD after discharge. In the B-Team program, a parallel OBOT program was created by a local nonaffiliated federally qualified health center. Although 89% of patients received treatment at this OBOT clinic, the inpatient team also has relationships with other local treatment centers, including programs that provide methadone. Another important barrier to high-quality outpatient care for OUD is the requirement of an X-waiver. To help overcome this barrier, our inpatient program partnered with a regional medical society to offer periodic X-waiver training to outpatient providers. In less than a year, more than 100 regional prescribers participated in this program.

Our study has several limitations. There was likely some degree of selection bias among the hospitalized patients who received initial buprenorphine treatment. To our knowledge, there is no specific validated screening tool for OUD in the inpatient acute care setting; moreover, we have been unable to implement standardized screening for OUD into the electronic health record. As such, we rely on the totality of the clinical circumstances approach to identify patients with OUD.

Furthermore, we had neither a comparison group nor a prospective plan to follow patients who did not remain engaged in care after discharge. In addition, our analysis of OUD admissions included F11 ICD-10 codes, which are limited by clinical documentation.^9,10 Our program focuses exclusively on buprenorphine initiation due to insufficient immediate outpatient capacity for methadone initiated during hospitalization and lack of coverage for extended-release naltrexone. Limitations to outpatient data-sharing prevented the reporting of outpatient appointments external to the identified partner program; since these appointments were included in the analysis as “lost to follow-up,” actual engagement rates may be higher than those reported.

Moving forward, the B-Team is continuing to serve as a role model for appropriate, patient-centered, evidence-based care for hospitalized patients with OUD. Attending physicians and residents with an X-waiver are now encouraged to initiate buprenorphine treatment on their own. In June 2020, we added peer-recovery support services to the program, which has improved care for patients and increased adoption of hospital-initiated substance use disorder interventions.¹¹ Lessons learned from inpatient implementation are being applied to our hospital’s emergency department and to an inpatient obstetrics unit at a partner hospital; they are also being employed to further empower hospitalists to diagnose and treat other substance use disorders, such as alcohol use disorder.

Hospitalizations related to opioid use disorder (OUD) have increased and now account for up to 6% of hospital admissions in certain areas of the United States.¹ Patients with OUD who are started on buprenorphine during hospitalization are more likely to enter outpatient treatment, stay in treatment longer, and have more drug-free days compared with patients who only receive a referral for outpatient treatment.^2,3 Therefore, a crucial comprehensive strategy for OUD care should include hospital-based programs that support initiation of treatment in the inpatient setting and strong bridges to outpatient care. One of the common barriers to initiating treatment in the inpatient setting, however, is a lack of access to addiction medicine specialists.^4-6

In 2017, we created a hospitalist-led interprofessional team called the B-Team (Buprenorphine Team) to help primary care teams identify patients with OUD, initiate and maintain buprenorphine therapy during hospitalization, provide warm handoffs to outpatient treatment programs, and reduce institutional stigma related to people with substance use disorders.

Program Description

The B-Team is led by a hospital medicine physician assistant and includes physicians from internal medicine, consult-liaison psychiatry, and palliative care; advanced practice and bedside nurses; a social worker; a pharmacist; a chaplain; a peer-recovery specialist; and medical trainees. The B-Team is notified of potential candidates for buprenorphine through a secure texting platform, one that is accessible to any healthcare provider at the hospital. Patients who are referred to the B-Team either self-identify or are identified by their primary team as having an underlying OUD. One of the B-Team providers assesses the patient to determine if they are eligible to receive inpatient therapy. Patients are considered eligible for the program if they meet Diagnostic and Statistical Manual of Mental Disorders (5th edition) criteria for OUD, have a desire to cease opioid use, and receive medical clearance to take buprenorphine.

For eligible patients, the B-Team provider orders a nurse-driven protocol to initiate buprenorphine for OUD. The chaplain offers psychospiritual counseling, and the social worker provides counseling and coordination of care. The B-Team partners with a nonhospital-affiliated, publicly-funded, office-based opioid treatment (OBOT) program that combines primary care with behavioral health programming. A follow-up outpatient appointment is secured prior to hospital discharge, and a member of the B-Team who has Drug Addiction Treatment Act of 2000 (DATA 2000) X-waiver certification prescribes buprenorphine as a bridge until the follow-up appointment. The medication is dispensed from the hospital’s retail pharmacy, and the patient leaves the hospital with the medication in-hand.

Patients who are not eligible for buprenorphine therapy are offered a harm-reduction intervention or referral to the psychiatry consult liaison service to assess for alternative diagnoses or treatment. These patients are also offered psychospiritual counseling and a prescription for naloxone.

Prior to the creation of the B-Team at our hospital, there was no structure in place to facilitate initiation of buprenorphine therapy during hospitalization and no linkage to outpatient treatment after discharge; furthermore, none of the hospitalists or other providers (including consulting psychiatrists) had an X-waiver to prescribe buprenorphine for OUD.

Program Evaluation

Study data were collected using Research Electronic Data Capture software. Inpatient and outpatient data were entered by a B-Team provider or a researcher via chart review. Patients were considered to be engaged in care if they attended at least one outpatient appointment for buprenorphine therapy during each of the following time periods: (1) 0 to 27 days (initial follow-up), 28-89 days (1- to 3-month follow-up), 90-179 days (3- to 6-month follow-up), and 180 days or more (>6-month follow-up). Only visits specifically for buprenorphine maintenance therapy were counted. If multiple encounters occurred within one time frame, the encounter closest to 0, 30, 90, or 180 days from discharge was used. If a patient did not attend any encounters during a specified time frame, they were considered to no longer be engaged in care and were no longer tracked for purposes of the evaluation. Data for the percentage of patients engaged in outpatient care are presented as the number of patients who attended at least one appointment during each of the follow-up periods (1 to 3 months, 3 to 6 months, or after 6 months, as noted above) divided by the number of patients who had been discharged with coordinated follow-up.

The number of patients admitted per month for whom there was an order to initiate inpatient buprenorphine therapy was analyzed using a statistical process control chart, in addition to the number of OUD admissions based on the inclusion of the International Classification of Disease, Tenth Revision (ICD-10) F11 code (opioid-related disorders) in any position in the discharge diagnoses.

This program and study were considered quality improvement by The University of Texas Institutional Review Board and did not meet criteria for human subjects research.

RESULTS

During the first 2 years of the program (September 2018-September 2020), the B-Team received 260 patient referrals. Most of the patients were White (72%), male (62%), and between ages 25 and 44 years (53%) (Appendix Table). The team initiated buprenorphine therapy in 132 hospitalized patients. In the year prior to the creation of the B-Team program, the average number of hospitalized patients receiving buprenorphine for OUD per month was three; after the launch of the B-Team program, this number increased to 12 encounters per month (Figure 1A). The sudden decrease observed in August 2020 is likely related to a surge in COVID-19 admissions. The number of monthly admissions for OUD is also shown (Figure 1B).

The B-Team saw a total of 132 eligible patients; members of the team provided counseling, support, and resources regarding buprenorphine therapy. In addition, the B-Team’s chaplain provided emotional support and spiritual connection (if desired) to 40 of these patients (30%). In the study, no cases of precipitated withdrawal were identified. Of the 132 patients seen, 110 (83%) were accepted to an outpatient OUD program upon discharge from the hospital; 98 (89%) of these patients were accepted at our partner OBOT clinic. The remaining patients were not interested in continuing OUD treatment (13%) or were denied acceptance to an outpatient program based on administrative and/or financial eligibility guidelines (4%). Patients who would not be attending an outpatient program were discontinued on buprenorphine therapy prior to discharge, counseled about naloxone, and provided printed resources.

Outpatient appointment attendance was used to measure ongoing treatment engagement of the 110 patients who were discharged with coordinated follow-up care. A total of 65 patients (59%) attended their first outpatient appointment; the average time between discharge and the first outpatient appointment was 5.9 days. Forty-two patients (38%) attended at least one appointment between 1 and 3 months; 29 (26%) between 3 and 6 months; and 24 (22%) after 6 months (Figure 2).

Of the 128 patients who were not administered buprenorphine therapy, 64 (50%) were not interested in starting treatment and/or were not ready to engage in treatment; 36 (28%) did not meet criteria for OUD treatment; 28 (22%) were already receiving treatment or preferred another type of OUD treatment; and 13 (10%) had severe comorbid addiction and/or illness requiring treatment that contraindicates the use of buprenorphine.

A volunteer hospitalist-led interprofessional team providing evidence-based care for hospitalized patients with OUD was associated with a substantial increase in patients receiving buprenorphine therapy—both during hospitalization and after discharge. In the program, 59% of patients attended initial follow-up appointments, and 22% of patients were still engaged at 6 months. These outpatient follow-up rates appear to be similar to, or higher than, other programs described in the literature. For example, a buprenorphine OUD-treatment initiative led by the psychiatry consult service at a Boston academic medical center resulted in less than half of patients receiving buprenorphine treatment within 2 months of discharge.⁷ In another study wherein an addiction medicine consult service administered buprenorphine to patients with OUD during hospitalization, 39%, 27%, and 18% of patients were retained in outpatient treatment at 30, 90, and 180 days, respectively.⁸

The B-Team model is likely generalizable to other hospital medicine groups that may not otherwise have access to inpatient care for substance use disorder. The B-Team is not an addiction medicine consultation service; rather, it is a hospitalist-led quality improvement initiative seeking to improve the standard of care for hospitalized patients with OUD.

A significant barrier is ensuring ongoing support for patients with OUD after discharge. In the B-Team program, a parallel OBOT program was created by a local nonaffiliated federally qualified health center. Although 89% of patients received treatment at this OBOT clinic, the inpatient team also has relationships with other local treatment centers, including programs that provide methadone. Another important barrier to high-quality outpatient care for OUD is the requirement of an X-waiver. To help overcome this barrier, our inpatient program partnered with a regional medical society to offer periodic X-waiver training to outpatient providers. In less than a year, more than 100 regional prescribers participated in this program.

Our study has several limitations. There was likely some degree of selection bias among the hospitalized patients who received initial buprenorphine treatment. To our knowledge, there is no specific validated screening tool for OUD in the inpatient acute care setting; moreover, we have been unable to implement standardized screening for OUD into the electronic health record. As such, we rely on the totality of the clinical circumstances approach to identify patients with OUD.

Furthermore, we had neither a comparison group nor a prospective plan to follow patients who did not remain engaged in care after discharge. In addition, our analysis of OUD admissions included F11 ICD-10 codes, which are limited by clinical documentation.^9,10 Our program focuses exclusively on buprenorphine initiation due to insufficient immediate outpatient capacity for methadone initiated during hospitalization and lack of coverage for extended-release naltrexone. Limitations to outpatient data-sharing prevented the reporting of outpatient appointments external to the identified partner program; since these appointments were included in the analysis as “lost to follow-up,” actual engagement rates may be higher than those reported.

Moving forward, the B-Team is continuing to serve as a role model for appropriate, patient-centered, evidence-based care for hospitalized patients with OUD. Attending physicians and residents with an X-waiver are now encouraged to initiate buprenorphine treatment on their own. In June 2020, we added peer-recovery support services to the program, which has improved care for patients and increased adoption of hospital-initiated substance use disorder interventions.¹¹ Lessons learned from inpatient implementation are being applied to our hospital’s emergency department and to an inpatient obstetrics unit at a partner hospital; they are also being employed to further empower hospitalists to diagnose and treat other substance use disorders, such as alcohol use disorder.

There is growing evidence that gender may be associated with COVID-19 infection, presentation, and prognosis.^1-4 Most published evidence, however, has focused on individual aspects, such as specific symptoms or prognoses. We sought to provide a comprehensive analysis of gender and COVID-19 infection from admission to 30 days after discharge in a large, multinational cohort.

The registry HOPE-COVID-19 (Health Outcome Predictive Evaluation for COVID-19, NCT04334291) is an international investigator-initiated study.⁵ The study was approved by the ethics committee of the promoting center and was appraised and accepted by the institutional review board or local committee of each participating hospital. It was designed as an ambispective cohort study. Patients are eligible for enrollment when discharged (whether dead or alive) after an in-hospital admission with a positive COVID-19 test or if their attending physician considered them highly likely to have presented with SARS-CoV-2 infection. All decisions and clinical procedures were performed by the attending physician team independently of this study, following the local regular practice and protocols. The information presented here corresponds to the HOPE-COVID-19 Registry, with a cutoff date of April 18, 2020.

Study methods and definitions are available in Appendix 1 and Appendix 2, respectively, and detailed in a previous paper⁵ and online on the web page of the study.⁶

Enrolled patients were divided into two groups according to their gender, then propensity score matching (PSM) analysis was performed (1:1 nearest neighbor matching, caliper = 0.01, without replacement and maximizing execution performance). Our primary end point was all-cause mortality at 30 days. Other clinically relevant events were recorded as secondary end points: invasive mechanical ventilation, noninvasive mechanical ventilation, pronation, respiratory insufficiency, heart failure, renal failure, upper respiratory tract involvement, pneumonia, sepsis, systemic inflammatory response syndrome, clinically relevant bleeding, hemoptysis, and embolic events. Events were allocated based on HOPE-COVID-19 registry definitions, following local researchers’ criteria. Abnormal blood test values were classified according to the reference values of local laboratories (Appendix 2).

Statistical analysis methods are outlined in Appendix 1.

Of the 2,798 patients consecutively enrolled in the HOPE registry, 1,111 were women (39.7%) and 1,687 were men (60.3%). Of the 2,375 (84.9%) patients who had a nasopharyngeal swab positive for COVID-19, 962 were women and 1,413 were men. Among the 2,798 patients initially included in the analysis, 876 gender-balanced pairs were selected after PSM.

Baseline Characteristics and Clinical Presentation

The baseline characteristics and clinical presentation of the overall population included in the study are summarized in Appendix Table 1. In the raw population, men had a significantly higher prevalence of conventional cardiovascular risk factors, such as diabetes, dyslipidemia, and smoking history, as well as a history of lung and cardiovascular diseases. On presentation, the most common symptoms for all patients were fever, cough, and dyspnea. Fever was more common in men, whereas vomiting, diarrhea, and upper airway symptoms (eg, sore throat, hyposmia/anosmia, dysgeusia) were more common in women.

Most patients had increased values of acute phase reactants. C-reactive protein (CRP) was elevated in 90.2% and D-dimer in 64.2% of patients, both significantly more often in men. Lymphocytopenia was present in 75.4% of patients, more commonly among men. Bilateral pneumonia occurred in 69.2% of the population, more frequently in men.

After PSM analysis (Appendix Table 2), a higher prevalence of hyposmia/anosmia and gastrointestinal symptoms in women was confirmed, as well as a higher prevalence of fever in men. Laboratory tests in men still presented alterations consistent with a more severe COVID-19 infection (significantly higher CRP, troponin, transaminases, lymphocytopenia, thrombocytopenia, and ferritin). There was no significant difference in the time between onset of symptoms and hospital admission by gender (6.2 ± 7.1 days in women vs 5.9 ± 7.6 days in men; P = .472).

The main findings after PSM analysis are summarized in Appendix Figure 1 and Appendix Figure 2.

In-Hospital Management and Outcomes

The supportive and pharmacologic treatments of study patients and their outcomes are summarized in Appendix Table 3. During the in-hospital stay, men required oxygen supplementation more frequently than women. Noninvasive mechanical ventilation, invasive mechanical ventilation, and pronation were more commonly used in men. Chloroquine/hydroxychloroquine, antivirals, and antibiotics were the medications most widely used in our population (84.5%, 65.8%, and 74.4% of patients, respectively), without significant differences between male and female patients, with the exception of antibiotics, which were used more often in men (76.6% vs 71.1%). Immunomodulators (corticosteroids, tocilizumab, and interferon) were used more often in male patients.

After PSM (Table), men more frequently received immunomodulators (corticosteroids and tocilizumab), antibiotics, and pronation. No differences in invasive and noninvasive mechanical ventilation were observed.

Thirty-day outcome data were available for all patients included in the analysis. During the in-hospital stay, 48% of patients developed respiratory insufficiency, 18.8% systemic inflammatory response syndrome (SIRS), and 13.2% overt sepsis. Respiratory insufficiency and SIRS were more common in male patients. Mortality at 30 days in the raw population was 21.4%, and men died more often than women (23.5% vs 18.2%; P = .001).

The PSM analysis continued to show a higher 30-day mortality rate among men (Figure), as well as greater need for oxygen, pronation, and use of immunomodulators and antibiotics (Table).

The results of our study confirm that among patients with COVID-19, men have a poorer prognosis than women. Because of the design of the study, it is not possible to determine if men are more prone to SARS-CoV-2 infection in our population; however, given the prevalence of men in our unselected, all-comers population, we can assume that men are either infected more often and/or more frequently symptomatic.

After PSM analysis, the 30-day all-cause mortality remained higher among men than women. The poorer prognosis of male patients is attributable not only to a higher burden of cardiovascular risk factors, but may also be related to unmodifiable biological factors, such as sex differences in angiotensin-converting enzyme 2 expression.^7,8 The worse prognosis observed in our study confirms the higher incidence of death in male patients that was observed in previous studies.⁹ Liu et al questioned the role of gender as an independent prognostic factor in COVID-19¹⁰; however, that study included fewer patients, who were also younger and had less severe disease.

The clinical presentation of COVID-19 also differed by gender in our study. Gastrointestinal symptoms and hyposmia/anosmia were more common in women, whereas fever was more common in men. The prevalence of olfactory and gustatory dysfunction in women has already been described,^11,12 and these symptoms have been linked with milder disease.¹³ It is possible that women presenting to the hospital had milder forms of COVID-19, or that there were systematic differences in how men and women sought medical care. The results of our study emphasize the need for a high level of suspicion for COVID-19 infection in women, even in the presence of mild mucosal or gastrointestinal symptoms and/or relatively minor laboratory abnormalities.

Laboratory values indicative of more severe COVID-19 infection in men could suggest a higher inflammatory response to the infection. Men also received more immunomodulators and antibiotics in this study. A recent paper from Scully et al¹⁴ pointed out the different immune response to viruses observed in men that could partially explain the higher level of inflammation markers and the more severe disease observed in men.

Our study has several limitations. As an observational study of hospitalized patients, it may represent patients with more severe COVID-19. Men and women may have sought hospital care differently. Diagnosis, testing, and treatment were not standardized and may have been influenced by patient gender. Although we attempted to match patients on baseline medical conditions, we may not have completely controlled for differences in preexisting health. Finally, gender data were collected as binary and so did not capture other gender categories.

In our multicenter cohort of hospitalized COVID-19 patients, men had a higher burden of risk factors; different clinical presentations, with more fever and less olfactory and gastrointestinal symptoms; and a significantly poorer prognosis than women did at 30 days.

The authors thank Cardiovascular Excellence SL for their essential support regarding the database and HOPE web page as well as all HOPE researchers. The authors also thank Michael Andrews for his valuable contribution to the English revision.

There is growing evidence that gender may be associated with COVID-19 infection, presentation, and prognosis.^1-4 Most published evidence, however, has focused on individual aspects, such as specific symptoms or prognoses. We sought to provide a comprehensive analysis of gender and COVID-19 infection from admission to 30 days after discharge in a large, multinational cohort.

The registry HOPE-COVID-19 (Health Outcome Predictive Evaluation for COVID-19, NCT04334291) is an international investigator-initiated study.⁵ The study was approved by the ethics committee of the promoting center and was appraised and accepted by the institutional review board or local committee of each participating hospital. It was designed as an ambispective cohort study. Patients are eligible for enrollment when discharged (whether dead or alive) after an in-hospital admission with a positive COVID-19 test or if their attending physician considered them highly likely to have presented with SARS-CoV-2 infection. All decisions and clinical procedures were performed by the attending physician team independently of this study, following the local regular practice and protocols. The information presented here corresponds to the HOPE-COVID-19 Registry, with a cutoff date of April 18, 2020.

Study methods and definitions are available in Appendix 1 and Appendix 2, respectively, and detailed in a previous paper⁵ and online on the web page of the study.⁶

Enrolled patients were divided into two groups according to their gender, then propensity score matching (PSM) analysis was performed (1:1 nearest neighbor matching, caliper = 0.01, without replacement and maximizing execution performance). Our primary end point was all-cause mortality at 30 days. Other clinically relevant events were recorded as secondary end points: invasive mechanical ventilation, noninvasive mechanical ventilation, pronation, respiratory insufficiency, heart failure, renal failure, upper respiratory tract involvement, pneumonia, sepsis, systemic inflammatory response syndrome, clinically relevant bleeding, hemoptysis, and embolic events. Events were allocated based on HOPE-COVID-19 registry definitions, following local researchers’ criteria. Abnormal blood test values were classified according to the reference values of local laboratories (Appendix 2).

Statistical analysis methods are outlined in Appendix 1.

Of the 2,798 patients consecutively enrolled in the HOPE registry, 1,111 were women (39.7%) and 1,687 were men (60.3%). Of the 2,375 (84.9%) patients who had a nasopharyngeal swab positive for COVID-19, 962 were women and 1,413 were men. Among the 2,798 patients initially included in the analysis, 876 gender-balanced pairs were selected after PSM.

Baseline Characteristics and Clinical Presentation

The baseline characteristics and clinical presentation of the overall population included in the study are summarized in Appendix Table 1. In the raw population, men had a significantly higher prevalence of conventional cardiovascular risk factors, such as diabetes, dyslipidemia, and smoking history, as well as a history of lung and cardiovascular diseases. On presentation, the most common symptoms for all patients were fever, cough, and dyspnea. Fever was more common in men, whereas vomiting, diarrhea, and upper airway symptoms (eg, sore throat, hyposmia/anosmia, dysgeusia) were more common in women.

Most patients had increased values of acute phase reactants. C-reactive protein (CRP) was elevated in 90.2% and D-dimer in 64.2% of patients, both significantly more often in men. Lymphocytopenia was present in 75.4% of patients, more commonly among men. Bilateral pneumonia occurred in 69.2% of the population, more frequently in men.

After PSM analysis (Appendix Table 2), a higher prevalence of hyposmia/anosmia and gastrointestinal symptoms in women was confirmed, as well as a higher prevalence of fever in men. Laboratory tests in men still presented alterations consistent with a more severe COVID-19 infection (significantly higher CRP, troponin, transaminases, lymphocytopenia, thrombocytopenia, and ferritin). There was no significant difference in the time between onset of symptoms and hospital admission by gender (6.2 ± 7.1 days in women vs 5.9 ± 7.6 days in men; P = .472).

The main findings after PSM analysis are summarized in Appendix Figure 1 and Appendix Figure 2.

In-Hospital Management and Outcomes

The supportive and pharmacologic treatments of study patients and their outcomes are summarized in Appendix Table 3. During the in-hospital stay, men required oxygen supplementation more frequently than women. Noninvasive mechanical ventilation, invasive mechanical ventilation, and pronation were more commonly used in men. Chloroquine/hydroxychloroquine, antivirals, and antibiotics were the medications most widely used in our population (84.5%, 65.8%, and 74.4% of patients, respectively), without significant differences between male and female patients, with the exception of antibiotics, which were used more often in men (76.6% vs 71.1%). Immunomodulators (corticosteroids, tocilizumab, and interferon) were used more often in male patients.

After PSM (Table), men more frequently received immunomodulators (corticosteroids and tocilizumab), antibiotics, and pronation. No differences in invasive and noninvasive mechanical ventilation were observed.

Thirty-day outcome data were available for all patients included in the analysis. During the in-hospital stay, 48% of patients developed respiratory insufficiency, 18.8% systemic inflammatory response syndrome (SIRS), and 13.2% overt sepsis. Respiratory insufficiency and SIRS were more common in male patients. Mortality at 30 days in the raw population was 21.4%, and men died more often than women (23.5% vs 18.2%; P = .001).

The PSM analysis continued to show a higher 30-day mortality rate among men (Figure), as well as greater need for oxygen, pronation, and use of immunomodulators and antibiotics (Table).

The results of our study confirm that among patients with COVID-19, men have a poorer prognosis than women. Because of the design of the study, it is not possible to determine if men are more prone to SARS-CoV-2 infection in our population; however, given the prevalence of men in our unselected, all-comers population, we can assume that men are either infected more often and/or more frequently symptomatic.

After PSM analysis, the 30-day all-cause mortality remained higher among men than women. The poorer prognosis of male patients is attributable not only to a higher burden of cardiovascular risk factors, but may also be related to unmodifiable biological factors, such as sex differences in angiotensin-converting enzyme 2 expression.^7,8 The worse prognosis observed in our study confirms the higher incidence of death in male patients that was observed in previous studies.⁹ Liu et al questioned the role of gender as an independent prognostic factor in COVID-19¹⁰; however, that study included fewer patients, who were also younger and had less severe disease.

The clinical presentation of COVID-19 also differed by gender in our study. Gastrointestinal symptoms and hyposmia/anosmia were more common in women, whereas fever was more common in men. The prevalence of olfactory and gustatory dysfunction in women has already been described,^11,12 and these symptoms have been linked with milder disease.¹³ It is possible that women presenting to the hospital had milder forms of COVID-19, or that there were systematic differences in how men and women sought medical care. The results of our study emphasize the need for a high level of suspicion for COVID-19 infection in women, even in the presence of mild mucosal or gastrointestinal symptoms and/or relatively minor laboratory abnormalities.

Laboratory values indicative of more severe COVID-19 infection in men could suggest a higher inflammatory response to the infection. Men also received more immunomodulators and antibiotics in this study. A recent paper from Scully et al¹⁴ pointed out the different immune response to viruses observed in men that could partially explain the higher level of inflammation markers and the more severe disease observed in men.

Our study has several limitations. As an observational study of hospitalized patients, it may represent patients with more severe COVID-19. Men and women may have sought hospital care differently. Diagnosis, testing, and treatment were not standardized and may have been influenced by patient gender. Although we attempted to match patients on baseline medical conditions, we may not have completely controlled for differences in preexisting health. Finally, gender data were collected as binary and so did not capture other gender categories.

In our multicenter cohort of hospitalized COVID-19 patients, men had a higher burden of risk factors; different clinical presentations, with more fever and less olfactory and gastrointestinal symptoms; and a significantly poorer prognosis than women did at 30 days.

The authors thank Cardiovascular Excellence SL for their essential support regarding the database and HOPE web page as well as all HOPE researchers. The authors also thank Michael Andrews for his valuable contribution to the English revision.

There is growing evidence that gender may be associated with COVID-19 infection, presentation, and prognosis.^1-4 Most published evidence, however, has focused on individual aspects, such as specific symptoms or prognoses. We sought to provide a comprehensive analysis of gender and COVID-19 infection from admission to 30 days after discharge in a large, multinational cohort.

The registry HOPE-COVID-19 (Health Outcome Predictive Evaluation for COVID-19, NCT04334291) is an international investigator-initiated study.⁵ The study was approved by the ethics committee of the promoting center and was appraised and accepted by the institutional review board or local committee of each participating hospital. It was designed as an ambispective cohort study. Patients are eligible for enrollment when discharged (whether dead or alive) after an in-hospital admission with a positive COVID-19 test or if their attending physician considered them highly likely to have presented with SARS-CoV-2 infection. All decisions and clinical procedures were performed by the attending physician team independently of this study, following the local regular practice and protocols. The information presented here corresponds to the HOPE-COVID-19 Registry, with a cutoff date of April 18, 2020.

Study methods and definitions are available in Appendix 1 and Appendix 2, respectively, and detailed in a previous paper⁵ and online on the web page of the study.⁶

Enrolled patients were divided into two groups according to their gender, then propensity score matching (PSM) analysis was performed (1:1 nearest neighbor matching, caliper = 0.01, without replacement and maximizing execution performance). Our primary end point was all-cause mortality at 30 days. Other clinically relevant events were recorded as secondary end points: invasive mechanical ventilation, noninvasive mechanical ventilation, pronation, respiratory insufficiency, heart failure, renal failure, upper respiratory tract involvement, pneumonia, sepsis, systemic inflammatory response syndrome, clinically relevant bleeding, hemoptysis, and embolic events. Events were allocated based on HOPE-COVID-19 registry definitions, following local researchers’ criteria. Abnormal blood test values were classified according to the reference values of local laboratories (Appendix 2).

Statistical analysis methods are outlined in Appendix 1.

Of the 2,798 patients consecutively enrolled in the HOPE registry, 1,111 were women (39.7%) and 1,687 were men (60.3%). Of the 2,375 (84.9%) patients who had a nasopharyngeal swab positive for COVID-19, 962 were women and 1,413 were men. Among the 2,798 patients initially included in the analysis, 876 gender-balanced pairs were selected after PSM.

Baseline Characteristics and Clinical Presentation

The baseline characteristics and clinical presentation of the overall population included in the study are summarized in Appendix Table 1. In the raw population, men had a significantly higher prevalence of conventional cardiovascular risk factors, such as diabetes, dyslipidemia, and smoking history, as well as a history of lung and cardiovascular diseases. On presentation, the most common symptoms for all patients were fever, cough, and dyspnea. Fever was more common in men, whereas vomiting, diarrhea, and upper airway symptoms (eg, sore throat, hyposmia/anosmia, dysgeusia) were more common in women.

Most patients had increased values of acute phase reactants. C-reactive protein (CRP) was elevated in 90.2% and D-dimer in 64.2% of patients, both significantly more often in men. Lymphocytopenia was present in 75.4% of patients, more commonly among men. Bilateral pneumonia occurred in 69.2% of the population, more frequently in men.

After PSM analysis (Appendix Table 2), a higher prevalence of hyposmia/anosmia and gastrointestinal symptoms in women was confirmed, as well as a higher prevalence of fever in men. Laboratory tests in men still presented alterations consistent with a more severe COVID-19 infection (significantly higher CRP, troponin, transaminases, lymphocytopenia, thrombocytopenia, and ferritin). There was no significant difference in the time between onset of symptoms and hospital admission by gender (6.2 ± 7.1 days in women vs 5.9 ± 7.6 days in men; P = .472).

The main findings after PSM analysis are summarized in Appendix Figure 1 and Appendix Figure 2.

In-Hospital Management and Outcomes

The supportive and pharmacologic treatments of study patients and their outcomes are summarized in Appendix Table 3. During the in-hospital stay, men required oxygen supplementation more frequently than women. Noninvasive mechanical ventilation, invasive mechanical ventilation, and pronation were more commonly used in men. Chloroquine/hydroxychloroquine, antivirals, and antibiotics were the medications most widely used in our population (84.5%, 65.8%, and 74.4% of patients, respectively), without significant differences between male and female patients, with the exception of antibiotics, which were used more often in men (76.6% vs 71.1%). Immunomodulators (corticosteroids, tocilizumab, and interferon) were used more often in male patients.

After PSM (Table), men more frequently received immunomodulators (corticosteroids and tocilizumab), antibiotics, and pronation. No differences in invasive and noninvasive mechanical ventilation were observed.

Thirty-day outcome data were available for all patients included in the analysis. During the in-hospital stay, 48% of patients developed respiratory insufficiency, 18.8% systemic inflammatory response syndrome (SIRS), and 13.2% overt sepsis. Respiratory insufficiency and SIRS were more common in male patients. Mortality at 30 days in the raw population was 21.4%, and men died more often than women (23.5% vs 18.2%; P = .001).

The PSM analysis continued to show a higher 30-day mortality rate among men (Figure), as well as greater need for oxygen, pronation, and use of immunomodulators and antibiotics (Table).

The results of our study confirm that among patients with COVID-19, men have a poorer prognosis than women. Because of the design of the study, it is not possible to determine if men are more prone to SARS-CoV-2 infection in our population; however, given the prevalence of men in our unselected, all-comers population, we can assume that men are either infected more often and/or more frequently symptomatic.

After PSM analysis, the 30-day all-cause mortality remained higher among men than women. The poorer prognosis of male patients is attributable not only to a higher burden of cardiovascular risk factors, but may also be related to unmodifiable biological factors, such as sex differences in angiotensin-converting enzyme 2 expression.^7,8 The worse prognosis observed in our study confirms the higher incidence of death in male patients that was observed in previous studies.⁹ Liu et al questioned the role of gender as an independent prognostic factor in COVID-19¹⁰; however, that study included fewer patients, who were also younger and had less severe disease.

The clinical presentation of COVID-19 also differed by gender in our study. Gastrointestinal symptoms and hyposmia/anosmia were more common in women, whereas fever was more common in men. The prevalence of olfactory and gustatory dysfunction in women has already been described,^11,12 and these symptoms have been linked with milder disease.¹³ It is possible that women presenting to the hospital had milder forms of COVID-19, or that there were systematic differences in how men and women sought medical care. The results of our study emphasize the need for a high level of suspicion for COVID-19 infection in women, even in the presence of mild mucosal or gastrointestinal symptoms and/or relatively minor laboratory abnormalities.

Laboratory values indicative of more severe COVID-19 infection in men could suggest a higher inflammatory response to the infection. Men also received more immunomodulators and antibiotics in this study. A recent paper from Scully et al¹⁴ pointed out the different immune response to viruses observed in men that could partially explain the higher level of inflammation markers and the more severe disease observed in men.

Our study has several limitations. As an observational study of hospitalized patients, it may represent patients with more severe COVID-19. Men and women may have sought hospital care differently. Diagnosis, testing, and treatment were not standardized and may have been influenced by patient gender. Although we attempted to match patients on baseline medical conditions, we may not have completely controlled for differences in preexisting health. Finally, gender data were collected as binary and so did not capture other gender categories.

In our multicenter cohort of hospitalized COVID-19 patients, men had a higher burden of risk factors; different clinical presentations, with more fever and less olfactory and gastrointestinal symptoms; and a significantly poorer prognosis than women did at 30 days.

The authors thank Cardiovascular Excellence SL for their essential support regarding the database and HOPE web page as well as all HOPE researchers. The authors also thank Michael Andrews for his valuable contribution to the English revision.

The COVID-19 pandemic has disrupted the educational experience of medical trainees around the world, and this has been especially true for those in New York City (NYC), the early epicenter of the global outbreak.¹ The pandemic’s surge required redeployment of trainees away from scheduled rotations, focused didactics around emerging COVID-19 data, and seemingly narrowed trainees’ clinical exposure to a single respiratory infection.

While there is a small body of literature describing the programmatic responses^2,3 and educational adaptations^4-7 that have come about as a result of the pandemic’s disruptive force, a characterization of exactly how trainees’ clinical experiences have been affected is lacking. A detailed understanding of how trainees’ inpatient care activities evolved during the pandemic could provide valuable practice habits feedback, allow for comparisons across training sites, focus content selection for didactic learning and self-study, and potentially help forecast similar clinical changes in the event of a subsequent wave. Perhaps most important, as internal medicine (IM) trainees require broad exposure to diverse clinical conditions to mature toward independent practice, a characterization of exactly how the pandemic has narrowed the diversity of clinical exposure could inform changes in how trainees attain clinical competence.

Profiling IM residents’ clinical experiences in a meaningful way is particularly challenging given the extraordinary breadth of the field. We recently developed a strategy by which resident-attributed International Classification of Diseases, Tenth Revision (ICD-10) principal diagnosis codes are mapped to an educational taxonomy of medical content categories, yielding clinical exposure profiles.⁸ Here, we apply this mapping strategy to all four training hospitals of a large NYC IM residency program to catalogue the evolution of clinical diversity experienced by residents during the COVID-19 pandemic.

Study Population

The NYU IM Residency Program comprises 225 resident physicians rotating at four inpatient training sites: NYU Langone Hospital–Brooklyn (NYU-BK), NYU Langone Hospitals–Manhattan (NYU-MN), Bellevue Hospital (BH), and VA–New York Harbor Healthcare (VA). The study period was defined as February 1, 2020, to May 31, 2020, to capture clinical exposure during baseline, surge, and immediate post-surge periods. The NYU IM residency program declared pandemic emergency status on March 23, 2020, after which all residents were assigned to inpatient acute care and intensive care rotations to augment the inpatient workforce.

Data Source

Clinical data at each training hospital are collected and stored, allowing for asynchronous querying. Given differences in data reporting, strategies for collecting principal ICD-10 codes of patients discharged during the study period differed slightly across sites. Principal ICD-10 codes from patients discharged from NYU-BK and NYU-MN were filtered by nursing unit, allowing selection for resident-staffed units. Principal ICD-10 codes from BH were curated by care team, allowing selection for resident-staffed teams. Principal ICD-10 codes from VA were filtered by both hospital unit and provider service to attribute to resident providers. Dates of each discharge were included, and mortalities were included as discharges. All methods yielded a dataset of principal ICD-10 discharge diagnosis codes attributed primarily to IM residents. Given the rapid changes in hospital staffing to care for increasing patient volumes, in rare circumstances residents and other providers (such as advanced practice providers) shared hospital units. While ICD-10 codes mined from each hospital are attributed primarily to residents, this attribution is not entirely exclusive. Data were analyzed both by training site and in aggregate across the four training sites. No individually identifiable data were analyzed, the primary goal of the project was to improve education, and the data were collected as part of a required aspect of training; as a result, this project met criteria for certification as a quality improvement, and not a human subject, research project.

The Crosswalk Tool

We previously developed a crosswalk tool containing 4,854 ICD-10 diagnoses uniquely mapped to 16 broad medical content areas as defined by the American Board of Internal Medicine (ABIM).⁸ Custom programs (MATLAB, MathWorks, Inc) captured and subsequently mapped resident-attributed ICD-10 discharge codes to content areas if the syntax of the ICD-10 code in question exactly matched or was nested within an ICD-10 code in the crosswalk. This tool allowed us to measure the daily discharge frequency of each content area across the sites.

Analysis

The sensitivity of the crosswalk tool was calculated as the number of ICD-10 codes captured divided by the total number of patients. Codes missed by the tool were excluded. The total number, as well as the 7-day running average of discharges per content area, across the sites during the study period were measured. To evaluate for differences in the distribution of content before and after pandemic emergency status, 2 × 16 χ² contingency tables were constructed. To evaluate for changes in the mean relative proportions (%) of each content area, paired t tests were conducted. Confidence intervals were estimated from t distributions.

There were 6,613 patients discharged from all sites (NYU-BK, 2,062; NYU-MN, 2,188; BH, 1,711; VA, 652; Appendix Table). The crosswalk tool captured 6,384 principal discharge ICD-10 codes (96.5%). The five most common content areas during the study period were infectious diseases (ID; n = 2,892), cardiovascular disease (CVD; n = 1,199), gastroenterology (n = 406), pulmonary disease (n = 372), and nephrology and urology (n = 252). These were also the content areas most frequently encountered by residents at baseline (Figure and Table). The distribution of content prior to declaration of pandemic emergency status was significantly different than that after declaration (χ² = 709; df, 15; P <.001). ID diagnoses, driven by COVID-19, rose steeply in the period following declaration, peaked in mid-April, and slowly waned in May (Figure). The mean relative percentage of ID discharges across the sites rose from 26.0% (16.5%-35.4%) at baseline to 58.3% (41.3%-75.3%) in the period after pandemic emergency status was declared (P = .005).

Frequencies of diagnoses mapping to other content areas decreased significantly, reflecting a marked tapering of clinical diversity (Figure and Table). Specifically, decreases were seen in CVD (27.6% [95% CI, 17.9%-37.2%] to 13.9% [95% CI, 5.5%-22.3%]; P = .013); gastroenterology (8.3% [95% CI, 6.2%-10.2%] to 4.6% [95% CI, 2.1%-6.9%]; P = .038); pulmonary disease (8.0% [95% CI, 5.6%-10.2%] to 4.6% [95% CI, 1.6%-7.4%]; P = .040); and nephrology and urology (4.8% [95% CI, 2.6%-6.9%] to 3.1% [95% CI, 1.9%-4.2%]; P = .047) (Table). In late April, diagnoses mapping to these content areas began to repopulate residents’ clinical experiences and by the end of the study period had nearly returned to baseline frequencies. These patterns were similar when discharge diagnoses from each training site were plotted individually (Appendix Figure).

Here, we demonstrate how the clinical educational landscape changed for our residents during the COVID-19 pandemic. We uncover a dramatic deviation in the content to which residents were exposed through patient care activities that disproportionately favored ID at the expense of all other content. We demonstrate that this reduction in clinical diversity persisted for nearly 2 months and was similar at each of our training hospitals, and also provide a trajectory on which other content repopulated residents’ clinical experiences.

These data have served several valuable purposes and support ongoing efforts to map residents’ experiential curriculum at our program and others. Sharing this data with residents, as occurred routinely in town hall forums and noon conferences, has provided them with real-time practice feedback during a time of crisis. This has provided scope for their herculean efforts during the pandemic, served as a blueprint for underrepresented content most ripe for self-study, and offered reassurance of a return to normalcy given the trajectory of clinical content curves. As practice habits feedback is an Accreditation Council for Graduate Medical Education requirement, this strategy has also served as a robust and reproducible means of complying.

Our training program used this characterization of clinical content to help guide teaching in the pandemic era. For example, we preferentially structured case conferences and other didactics around reemerging content areas to capitalize on just-in-time education and harness residents’ eagerness for a respite from COVID-specific education. Residents required to quarantine at home were provided with learning plans centered on content underrepresented in clinical practice.

Given the critical importance of experiential learning in IM residents’ training, our findings quantifying significant changes in clinical exposure could form the basis for predicting poor outcomes in competency-based assessments for residents training in the COVID era, which continues to affect our trainees. For example, our characterization of clinical exposure may predict poor in-training exam or even ABIM certification exam performance in the content areas most drastically affected. Knowledge of this association of clinical exposure and clinical competence could allow training programs like ours to preempt poor performance in competency-based assessments by more aggressively shifting lectures, simulations, and other didactic programs toward content areas underrepresented in the pandemic’s wake.

Limitations of this study include the fact that availability of testing and ICD-10 coding for COVID-19 differed slightly across training sites, potentially contributing to site differences in mapping. Additionally, given our 1:1 mapping of ICD-10 codes to content categories, our strategy attributes COVID-19 to ID alone, and does not capture additional areas germane to this diagnosis, such as pulmonary disease.

We provide a detailed characterization of the evolution of a single IM program’s patient care experiences across four training hospitals during the COVID-19 pandemic. Such characterization can be leveraged to provide effective practice habits feedback and guide teaching efforts, and could form the basis to predict competency-based outcomes for trainees in the COVID era.

The COVID-19 pandemic has disrupted the educational experience of medical trainees around the world, and this has been especially true for those in New York City (NYC), the early epicenter of the global outbreak.¹ The pandemic’s surge required redeployment of trainees away from scheduled rotations, focused didactics around emerging COVID-19 data, and seemingly narrowed trainees’ clinical exposure to a single respiratory infection.

While there is a small body of literature describing the programmatic responses^2,3 and educational adaptations^4-7 that have come about as a result of the pandemic’s disruptive force, a characterization of exactly how trainees’ clinical experiences have been affected is lacking. A detailed understanding of how trainees’ inpatient care activities evolved during the pandemic could provide valuable practice habits feedback, allow for comparisons across training sites, focus content selection for didactic learning and self-study, and potentially help forecast similar clinical changes in the event of a subsequent wave. Perhaps most important, as internal medicine (IM) trainees require broad exposure to diverse clinical conditions to mature toward independent practice, a characterization of exactly how the pandemic has narrowed the diversity of clinical exposure could inform changes in how trainees attain clinical competence.

Profiling IM residents’ clinical experiences in a meaningful way is particularly challenging given the extraordinary breadth of the field. We recently developed a strategy by which resident-attributed International Classification of Diseases, Tenth Revision (ICD-10) principal diagnosis codes are mapped to an educational taxonomy of medical content categories, yielding clinical exposure profiles.⁸ Here, we apply this mapping strategy to all four training hospitals of a large NYC IM residency program to catalogue the evolution of clinical diversity experienced by residents during the COVID-19 pandemic.

Study Population

The NYU IM Residency Program comprises 225 resident physicians rotating at four inpatient training sites: NYU Langone Hospital–Brooklyn (NYU-BK), NYU Langone Hospitals–Manhattan (NYU-MN), Bellevue Hospital (BH), and VA–New York Harbor Healthcare (VA). The study period was defined as February 1, 2020, to May 31, 2020, to capture clinical exposure during baseline, surge, and immediate post-surge periods. The NYU IM residency program declared pandemic emergency status on March 23, 2020, after which all residents were assigned to inpatient acute care and intensive care rotations to augment the inpatient workforce.

Data Source

Clinical data at each training hospital are collected and stored, allowing for asynchronous querying. Given differences in data reporting, strategies for collecting principal ICD-10 codes of patients discharged during the study period differed slightly across sites. Principal ICD-10 codes from patients discharged from NYU-BK and NYU-MN were filtered by nursing unit, allowing selection for resident-staffed units. Principal ICD-10 codes from BH were curated by care team, allowing selection for resident-staffed teams. Principal ICD-10 codes from VA were filtered by both hospital unit and provider service to attribute to resident providers. Dates of each discharge were included, and mortalities were included as discharges. All methods yielded a dataset of principal ICD-10 discharge diagnosis codes attributed primarily to IM residents. Given the rapid changes in hospital staffing to care for increasing patient volumes, in rare circumstances residents and other providers (such as advanced practice providers) shared hospital units. While ICD-10 codes mined from each hospital are attributed primarily to residents, this attribution is not entirely exclusive. Data were analyzed both by training site and in aggregate across the four training sites. No individually identifiable data were analyzed, the primary goal of the project was to improve education, and the data were collected as part of a required aspect of training; as a result, this project met criteria for certification as a quality improvement, and not a human subject, research project.

The Crosswalk Tool

We previously developed a crosswalk tool containing 4,854 ICD-10 diagnoses uniquely mapped to 16 broad medical content areas as defined by the American Board of Internal Medicine (ABIM).⁸ Custom programs (MATLAB, MathWorks, Inc) captured and subsequently mapped resident-attributed ICD-10 discharge codes to content areas if the syntax of the ICD-10 code in question exactly matched or was nested within an ICD-10 code in the crosswalk. This tool allowed us to measure the daily discharge frequency of each content area across the sites.

Analysis

The sensitivity of the crosswalk tool was calculated as the number of ICD-10 codes captured divided by the total number of patients. Codes missed by the tool were excluded. The total number, as well as the 7-day running average of discharges per content area, across the sites during the study period were measured. To evaluate for differences in the distribution of content before and after pandemic emergency status, 2 × 16 χ² contingency tables were constructed. To evaluate for changes in the mean relative proportions (%) of each content area, paired t tests were conducted. Confidence intervals were estimated from t distributions.

There were 6,613 patients discharged from all sites (NYU-BK, 2,062; NYU-MN, 2,188; BH, 1,711; VA, 652; Appendix Table). The crosswalk tool captured 6,384 principal discharge ICD-10 codes (96.5%). The five most common content areas during the study period were infectious diseases (ID; n = 2,892), cardiovascular disease (CVD; n = 1,199), gastroenterology (n = 406), pulmonary disease (n = 372), and nephrology and urology (n = 252). These were also the content areas most frequently encountered by residents at baseline (Figure and Table). The distribution of content prior to declaration of pandemic emergency status was significantly different than that after declaration (χ² = 709; df, 15; P <.001). ID diagnoses, driven by COVID-19, rose steeply in the period following declaration, peaked in mid-April, and slowly waned in May (Figure). The mean relative percentage of ID discharges across the sites rose from 26.0% (16.5%-35.4%) at baseline to 58.3% (41.3%-75.3%) in the period after pandemic emergency status was declared (P = .005).

Frequencies of diagnoses mapping to other content areas decreased significantly, reflecting a marked tapering of clinical diversity (Figure and Table). Specifically, decreases were seen in CVD (27.6% [95% CI, 17.9%-37.2%] to 13.9% [95% CI, 5.5%-22.3%]; P = .013); gastroenterology (8.3% [95% CI, 6.2%-10.2%] to 4.6% [95% CI, 2.1%-6.9%]; P = .038); pulmonary disease (8.0% [95% CI, 5.6%-10.2%] to 4.6% [95% CI, 1.6%-7.4%]; P = .040); and nephrology and urology (4.8% [95% CI, 2.6%-6.9%] to 3.1% [95% CI, 1.9%-4.2%]; P = .047) (Table). In late April, diagnoses mapping to these content areas began to repopulate residents’ clinical experiences and by the end of the study period had nearly returned to baseline frequencies. These patterns were similar when discharge diagnoses from each training site were plotted individually (Appendix Figure).

Here, we demonstrate how the clinical educational landscape changed for our residents during the COVID-19 pandemic. We uncover a dramatic deviation in the content to which residents were exposed through patient care activities that disproportionately favored ID at the expense of all other content. We demonstrate that this reduction in clinical diversity persisted for nearly 2 months and was similar at each of our training hospitals, and also provide a trajectory on which other content repopulated residents’ clinical experiences.

These data have served several valuable purposes and support ongoing efforts to map residents’ experiential curriculum at our program and others. Sharing this data with residents, as occurred routinely in town hall forums and noon conferences, has provided them with real-time practice feedback during a time of crisis. This has provided scope for their herculean efforts during the pandemic, served as a blueprint for underrepresented content most ripe for self-study, and offered reassurance of a return to normalcy given the trajectory of clinical content curves. As practice habits feedback is an Accreditation Council for Graduate Medical Education requirement, this strategy has also served as a robust and reproducible means of complying.

Our training program used this characterization of clinical content to help guide teaching in the pandemic era. For example, we preferentially structured case conferences and other didactics around reemerging content areas to capitalize on just-in-time education and harness residents’ eagerness for a respite from COVID-specific education. Residents required to quarantine at home were provided with learning plans centered on content underrepresented in clinical practice.

Given the critical importance of experiential learning in IM residents’ training, our findings quantifying significant changes in clinical exposure could form the basis for predicting poor outcomes in competency-based assessments for residents training in the COVID era, which continues to affect our trainees. For example, our characterization of clinical exposure may predict poor in-training exam or even ABIM certification exam performance in the content areas most drastically affected. Knowledge of this association of clinical exposure and clinical competence could allow training programs like ours to preempt poor performance in competency-based assessments by more aggressively shifting lectures, simulations, and other didactic programs toward content areas underrepresented in the pandemic’s wake.

Limitations of this study include the fact that availability of testing and ICD-10 coding for COVID-19 differed slightly across training sites, potentially contributing to site differences in mapping. Additionally, given our 1:1 mapping of ICD-10 codes to content categories, our strategy attributes COVID-19 to ID alone, and does not capture additional areas germane to this diagnosis, such as pulmonary disease.

We provide a detailed characterization of the evolution of a single IM program’s patient care experiences across four training hospitals during the COVID-19 pandemic. Such characterization can be leveraged to provide effective practice habits feedback and guide teaching efforts, and could form the basis to predict competency-based outcomes for trainees in the COVID era.

The COVID-19 pandemic has disrupted the educational experience of medical trainees around the world, and this has been especially true for those in New York City (NYC), the early epicenter of the global outbreak.¹ The pandemic’s surge required redeployment of trainees away from scheduled rotations, focused didactics around emerging COVID-19 data, and seemingly narrowed trainees’ clinical exposure to a single respiratory infection.

While there is a small body of literature describing the programmatic responses^2,3 and educational adaptations^4-7 that have come about as a result of the pandemic’s disruptive force, a characterization of exactly how trainees’ clinical experiences have been affected is lacking. A detailed understanding of how trainees’ inpatient care activities evolved during the pandemic could provide valuable practice habits feedback, allow for comparisons across training sites, focus content selection for didactic learning and self-study, and potentially help forecast similar clinical changes in the event of a subsequent wave. Perhaps most important, as internal medicine (IM) trainees require broad exposure to diverse clinical conditions to mature toward independent practice, a characterization of exactly how the pandemic has narrowed the diversity of clinical exposure could inform changes in how trainees attain clinical competence.

Profiling IM residents’ clinical experiences in a meaningful way is particularly challenging given the extraordinary breadth of the field. We recently developed a strategy by which resident-attributed International Classification of Diseases, Tenth Revision (ICD-10) principal diagnosis codes are mapped to an educational taxonomy of medical content categories, yielding clinical exposure profiles.⁸ Here, we apply this mapping strategy to all four training hospitals of a large NYC IM residency program to catalogue the evolution of clinical diversity experienced by residents during the COVID-19 pandemic.

Study Population

The NYU IM Residency Program comprises 225 resident physicians rotating at four inpatient training sites: NYU Langone Hospital–Brooklyn (NYU-BK), NYU Langone Hospitals–Manhattan (NYU-MN), Bellevue Hospital (BH), and VA–New York Harbor Healthcare (VA). The study period was defined as February 1, 2020, to May 31, 2020, to capture clinical exposure during baseline, surge, and immediate post-surge periods. The NYU IM residency program declared pandemic emergency status on March 23, 2020, after which all residents were assigned to inpatient acute care and intensive care rotations to augment the inpatient workforce.

Data Source

Clinical data at each training hospital are collected and stored, allowing for asynchronous querying. Given differences in data reporting, strategies for collecting principal ICD-10 codes of patients discharged during the study period differed slightly across sites. Principal ICD-10 codes from patients discharged from NYU-BK and NYU-MN were filtered by nursing unit, allowing selection for resident-staffed units. Principal ICD-10 codes from BH were curated by care team, allowing selection for resident-staffed teams. Principal ICD-10 codes from VA were filtered by both hospital unit and provider service to attribute to resident providers. Dates of each discharge were included, and mortalities were included as discharges. All methods yielded a dataset of principal ICD-10 discharge diagnosis codes attributed primarily to IM residents. Given the rapid changes in hospital staffing to care for increasing patient volumes, in rare circumstances residents and other providers (such as advanced practice providers) shared hospital units. While ICD-10 codes mined from each hospital are attributed primarily to residents, this attribution is not entirely exclusive. Data were analyzed both by training site and in aggregate across the four training sites. No individually identifiable data were analyzed, the primary goal of the project was to improve education, and the data were collected as part of a required aspect of training; as a result, this project met criteria for certification as a quality improvement, and not a human subject, research project.

The Crosswalk Tool

We previously developed a crosswalk tool containing 4,854 ICD-10 diagnoses uniquely mapped to 16 broad medical content areas as defined by the American Board of Internal Medicine (ABIM).⁸ Custom programs (MATLAB, MathWorks, Inc) captured and subsequently mapped resident-attributed ICD-10 discharge codes to content areas if the syntax of the ICD-10 code in question exactly matched or was nested within an ICD-10 code in the crosswalk. This tool allowed us to measure the daily discharge frequency of each content area across the sites.

Analysis

The sensitivity of the crosswalk tool was calculated as the number of ICD-10 codes captured divided by the total number of patients. Codes missed by the tool were excluded. The total number, as well as the 7-day running average of discharges per content area, across the sites during the study period were measured. To evaluate for differences in the distribution of content before and after pandemic emergency status, 2 × 16 χ² contingency tables were constructed. To evaluate for changes in the mean relative proportions (%) of each content area, paired t tests were conducted. Confidence intervals were estimated from t distributions.

There were 6,613 patients discharged from all sites (NYU-BK, 2,062; NYU-MN, 2,188; BH, 1,711; VA, 652; Appendix Table). The crosswalk tool captured 6,384 principal discharge ICD-10 codes (96.5%). The five most common content areas during the study period were infectious diseases (ID; n = 2,892), cardiovascular disease (CVD; n = 1,199), gastroenterology (n = 406), pulmonary disease (n = 372), and nephrology and urology (n = 252). These were also the content areas most frequently encountered by residents at baseline (Figure and Table). The distribution of content prior to declaration of pandemic emergency status was significantly different than that after declaration (χ² = 709; df, 15; P <.001). ID diagnoses, driven by COVID-19, rose steeply in the period following declaration, peaked in mid-April, and slowly waned in May (Figure). The mean relative percentage of ID discharges across the sites rose from 26.0% (16.5%-35.4%) at baseline to 58.3% (41.3%-75.3%) in the period after pandemic emergency status was declared (P = .005).

Frequencies of diagnoses mapping to other content areas decreased significantly, reflecting a marked tapering of clinical diversity (Figure and Table). Specifically, decreases were seen in CVD (27.6% [95% CI, 17.9%-37.2%] to 13.9% [95% CI, 5.5%-22.3%]; P = .013); gastroenterology (8.3% [95% CI, 6.2%-10.2%] to 4.6% [95% CI, 2.1%-6.9%]; P = .038); pulmonary disease (8.0% [95% CI, 5.6%-10.2%] to 4.6% [95% CI, 1.6%-7.4%]; P = .040); and nephrology and urology (4.8% [95% CI, 2.6%-6.9%] to 3.1% [95% CI, 1.9%-4.2%]; P = .047) (Table). In late April, diagnoses mapping to these content areas began to repopulate residents’ clinical experiences and by the end of the study period had nearly returned to baseline frequencies. These patterns were similar when discharge diagnoses from each training site were plotted individually (Appendix Figure).

Here, we demonstrate how the clinical educational landscape changed for our residents during the COVID-19 pandemic. We uncover a dramatic deviation in the content to which residents were exposed through patient care activities that disproportionately favored ID at the expense of all other content. We demonstrate that this reduction in clinical diversity persisted for nearly 2 months and was similar at each of our training hospitals, and also provide a trajectory on which other content repopulated residents’ clinical experiences.

These data have served several valuable purposes and support ongoing efforts to map residents’ experiential curriculum at our program and others. Sharing this data with residents, as occurred routinely in town hall forums and noon conferences, has provided them with real-time practice feedback during a time of crisis. This has provided scope for their herculean efforts during the pandemic, served as a blueprint for underrepresented content most ripe for self-study, and offered reassurance of a return to normalcy given the trajectory of clinical content curves. As practice habits feedback is an Accreditation Council for Graduate Medical Education requirement, this strategy has also served as a robust and reproducible means of complying.

Our training program used this characterization of clinical content to help guide teaching in the pandemic era. For example, we preferentially structured case conferences and other didactics around reemerging content areas to capitalize on just-in-time education and harness residents’ eagerness for a respite from COVID-specific education. Residents required to quarantine at home were provided with learning plans centered on content underrepresented in clinical practice.

Given the critical importance of experiential learning in IM residents’ training, our findings quantifying significant changes in clinical exposure could form the basis for predicting poor outcomes in competency-based assessments for residents training in the COVID era, which continues to affect our trainees. For example, our characterization of clinical exposure may predict poor in-training exam or even ABIM certification exam performance in the content areas most drastically affected. Knowledge of this association of clinical exposure and clinical competence could allow training programs like ours to preempt poor performance in competency-based assessments by more aggressively shifting lectures, simulations, and other didactic programs toward content areas underrepresented in the pandemic’s wake.

Limitations of this study include the fact that availability of testing and ICD-10 coding for COVID-19 differed slightly across training sites, potentially contributing to site differences in mapping. Additionally, given our 1:1 mapping of ICD-10 codes to content categories, our strategy attributes COVID-19 to ID alone, and does not capture additional areas germane to this diagnosis, such as pulmonary disease.

We provide a detailed characterization of the evolution of a single IM program’s patient care experiences across four training hospitals during the COVID-19 pandemic. Such characterization can be leveraged to provide effective practice habits feedback and guide teaching efforts, and could form the basis to predict competency-based outcomes for trainees in the COVID era.

Peritoneal fluid examination is often recommended for hospitalized patients with ascites.¹ The prevalence of spontaneous bacterial peritonitis (SBP) in these patients ranges from 10% to 30%.^2-6 Bedside paracentesis has clinical outcomes similar to that performed by radiology, with an improved length of stay (LOS) and decreased transfusion requirements.⁷

Internal medicine residency programs are establishing procedure services to address concerns about resident training in procedures and patient safety. Previous studies, which include paracentesis in patients with cirrhosis, have focused on resident comfort with procedures, supervision, procedural complications, and patient satisfaction.^8-12 However, the impact of a procedure service on the time from admission to the procedure has not been studied. In this study, we aimed to examine whether the institution of a hospitalist-run procedure service affected a patient’s LOS in the hospital and the time difference between a patient’s hospital admission and paracentesis (A2P).

An inpatient hospitalist-run procedure service was introduced on July 1, 2016. The service was staffed by a hospitalist and second-year internal medicine residents. The service is available 7:00 am to 5:00 pm all days of the week. To identify patients who underwent paracentesis, we queried our electronic medical records for all peritoneal fluid samples from July 1, 2016, to May 31, 2019. Paracenteses performed in the outpatient clinics, in the radiology suite, or in the emergency department were excluded if the patient was not admitted. We also excluded patients who had paracentesis within 6 hours of presentation, as these patients likely had an urgent clinical indication for paracentesis.

Data on age, gender, race, ethnicity, date and time of hospital admission, and discharge date and time were retrieved. We also retrieved data on the absolute number of polymorphonuclear leukocytes (PMN) in the peritoneal fluid sample; a patient with a count higher than 250/uL was considered to have SBP. The timestamp for the peritoneal fluid results was used to approximate the A2P time. Paracenteses performed by or under direct supervision of procedure service hospitalists were identified through a procedure log maintained by procedure service hospitalists. We generated a binary variable to differentiate patients who were admitted during the day from those admitted during the night, when the procedure service was not available. For all patients, we calculated the model for end-stage liver disease and sodium (MELD-Na) score.¹³ Groups performing paracenteses were categorized into procedure service, residents, and radiology. Primary clinical services were categorized into general medicine, gastroenterology, surgery, and others.

Data were summarized as mean (SD) or median (interquartile range) for continuous variables and as percentages for categorical variables. Patients who had paracenteses by radiology or residents during the study period were considered controls. We used concurrent controls to address secular time trends (eg, measures to decrease LOS or changes in ordering tests in the electronic health record) in outcome measures. Patient characteristics were compared using the Wilcoxon rank-sum test or the χ² test, as appropriate.

Two outcome variables were examined: LOS, and A2P time. Because both outcome variables were right skewed, we used generalized linear models with gamma distribution and log link. The advantage of a generalized linear model approach is that the transformed coefficients are better interpretable than when using the log transformation of the response variable.¹⁴ To account for time trends, we included time in months in the model. Models were adjusted for age, gender, race, whether the admission was during day or night, PMN in peritoneal fluid, MELD-Na score, platelet count on the day of procedure, presence or absence of cirrhosis, diagnosis-related groups weight, primary clinical service, and the group performing paracentesis. To address heterogeneity among patients included in our study and the fact that some patients had multiple paracenteses, we conducted sensitivity analyses by excluding all noncirrhotic patients and including only the first paracentesis. A P value less than .05 was considered significant. All statistical analyses were performed using Stata MP 16.0 for Windows (StataCorp LLC).

Of the 1,321 paracenteses included in our study, 509 (38.5%) were performed by the procedure service, 723 (54.7%) by residents, and 89 (6.7%) by radiology. For comparison, 15.4% of procedures were performed by the radiology service during the 3 years before the start of the procedure service. More than 50% of the first paracenteses were performed within 30 hours of admission. Hospitalists or residents under the direct supervision of a hospitalist performed all paracenteses. Residents performing paracenteses, when not on the procedure service, were on general internal medicine, gastroenterology, hematology and oncology, or surgical services. No failed paracentesis attempts by the procedure service were subsequently performed by radiology. The mean age of the participants was 55.3 (12.2) years, 728 (55%) were White, 502 (38%) were female, and SBP was present in 61 (4.6%) patients. There was no difference by age, gender, time of admission, presence of SBP, or peritoneal fluid PMN in patients who underwent paracentesis by the procedure service versus controls (Table 1). A higher proportion of White patients and patients with cirrhosis underwent paracentesis by the procedure service than by another service. The LOS and A2P time were significantly lower for patients who underwent paracentesis by the procedure service than by another service (Table 1). When examining the adjusted linear secular time trends, LOS decreased by 0.1% per month (95% CI, –0.5% to 0.8%; P = .67) and A2P time by 0.02% per month (95% CI, –1.0% to 1.1%; P = .96).

In unadjusted models but accounting for secular time trends, patients who had paracenteses performed by residents or by radiology had a 50% (95% CI, 22%-83%; P = .002) and 127% (95% CI, 65%-211%; P < .001) longer LOS, respectively, than when paracentesis was performed by the procedure service. After adjusting for potential confounders, the difference in LOS between radiology and the procedure service remained significant; patients who had a paracentesis performed by radiology had a 27% (95% CI, 2%-58%; P = .03) longer LOS than patients who had the procedure performed by the procedure service. This relative LOS translates into 88 (95% CI, 1-174 hours) additional hours in absolute LOS. There was no difference in LOS between the procedure service and residents in the adjusted analysis (Table 2).

Similarly, in unadjusted models for A2P time and accounting for secular time trends, patients who had a paracentesis performed by residents or by radiology had a 52% (95% CI, 23%-88%; P < .001) and 173% (95% CI, 109%-280%; P < .001) longer A2P time, respectively, than patients whose paracentesis was performed by the procedure service. After adjusting for potential confounders, the difference in A2P time between radiology and the procedure service remained significant. Patients who had paracentesis performed by radiology had a 40% (95% CI, 5%-87%; P = .02) longer A2P time than patients who had paracentesis performed by the procedure service. This relative increase translates into 52 (95% CI, 3.3-101 hours) additional hours in absolute A2P time. On the other hand, residents had a significantly shorter A2P time (–19%, 95% CI, –33% to 0.2%; P = .05) (Table 2).

In the sensitivity analysis, excluding noncirrhotic patients and including only the first paracentesis for patients who had multiple procedures performed during admission, the results remained unchanged. In adjusted analysis, patients who had paracentesis performed by radiology had a 47% (95% CI, 3.7%-108%; P = .03) longer LOS and 91% (95% CI, 19%-107%; P = .008) longer A2P time than when paracentesis was performed by the procedure service. There were no differences in LOS or A2P time between the procedure service and residents (Table 2).

In this study, we report that a hospitalist-run procedure service, when compared with a radiology service, is associated with decreased LOS and A2P time independent of studied potential confounders and secular time trends. We also showed that, compared with radiology, the A2P time for nonemergent procedures (those performed 6 hours after admission) was not adversely affected by the procedure service. Residents performing paracenteses independently had shorter A2P time than the procedure service.

Although several institutions have bedside procedure services, data are lacking on benefits. Previously, paracenteses performed by residents have been associated with decreased LOS and need for transfusions when compared with radiology.⁷ Our study extends these findings to show a shortened A2P time. Delays may occur when a patient is referred to radiology because of volume, triaging of higher-acuity procedures, and transportation. Procedure services provide consistent attending supervision, more procedures by upper-level residents, and a lower rate of unsuccessful procedures.^12,15 Current study findings support the importance of continuing bedside procedure training for at least those residents who are interested in hospital medicine.⁷

Our study has several strengths and some potential limitations. The study examined outcomes that are important to patients as well as hospital administrators; it also had a large sample size, spanning 3 years. As it was a retrospective cohort study, there is potential for residual confounding due to unmeasured confounders. We did not examine the potential effect modification of procedure urgency, as such data are difficult to discern. Our method of identification missed patients who received therapeutic paracentesis without laboratory analysis. It is unclear why more White patients were referred to the procedure team; this is an area for further evaluation. Results of this study are likely not generalizable to institutions with a robust radiology service that has built-in redundancy to accommodate urgent procedures and easy availability over the weekends.

We found that a hospitalist-run teaching procedure service is associated with shorter LOS and A2P time. Further research is needed to determine if the benefits of a procedure service extend to lowering morbidity and/or mortality, as well as to determine the cost-effectiveness of a procedure service and whether the significant investment by the institution in establishing a procedure service is mitigated by the gains from better patient outcomes and reduced LOS.

Peritoneal fluid examination is often recommended for hospitalized patients with ascites.¹ The prevalence of spontaneous bacterial peritonitis (SBP) in these patients ranges from 10% to 30%.^2-6 Bedside paracentesis has clinical outcomes similar to that performed by radiology, with an improved length of stay (LOS) and decreased transfusion requirements.⁷

Internal medicine residency programs are establishing procedure services to address concerns about resident training in procedures and patient safety. Previous studies, which include paracentesis in patients with cirrhosis, have focused on resident comfort with procedures, supervision, procedural complications, and patient satisfaction.^8-12 However, the impact of a procedure service on the time from admission to the procedure has not been studied. In this study, we aimed to examine whether the institution of a hospitalist-run procedure service affected a patient’s LOS in the hospital and the time difference between a patient’s hospital admission and paracentesis (A2P).

An inpatient hospitalist-run procedure service was introduced on July 1, 2016. The service was staffed by a hospitalist and second-year internal medicine residents. The service is available 7:00 am to 5:00 pm all days of the week. To identify patients who underwent paracentesis, we queried our electronic medical records for all peritoneal fluid samples from July 1, 2016, to May 31, 2019. Paracenteses performed in the outpatient clinics, in the radiology suite, or in the emergency department were excluded if the patient was not admitted. We also excluded patients who had paracentesis within 6 hours of presentation, as these patients likely had an urgent clinical indication for paracentesis.

Data on age, gender, race, ethnicity, date and time of hospital admission, and discharge date and time were retrieved. We also retrieved data on the absolute number of polymorphonuclear leukocytes (PMN) in the peritoneal fluid sample; a patient with a count higher than 250/uL was considered to have SBP. The timestamp for the peritoneal fluid results was used to approximate the A2P time. Paracenteses performed by or under direct supervision of procedure service hospitalists were identified through a procedure log maintained by procedure service hospitalists. We generated a binary variable to differentiate patients who were admitted during the day from those admitted during the night, when the procedure service was not available. For all patients, we calculated the model for end-stage liver disease and sodium (MELD-Na) score.¹³ Groups performing paracenteses were categorized into procedure service, residents, and radiology. Primary clinical services were categorized into general medicine, gastroenterology, surgery, and others.

Data were summarized as mean (SD) or median (interquartile range) for continuous variables and as percentages for categorical variables. Patients who had paracenteses by radiology or residents during the study period were considered controls. We used concurrent controls to address secular time trends (eg, measures to decrease LOS or changes in ordering tests in the electronic health record) in outcome measures. Patient characteristics were compared using the Wilcoxon rank-sum test or the χ² test, as appropriate.

Two outcome variables were examined: LOS, and A2P time. Because both outcome variables were right skewed, we used generalized linear models with gamma distribution and log link. The advantage of a generalized linear model approach is that the transformed coefficients are better interpretable than when using the log transformation of the response variable.¹⁴ To account for time trends, we included time in months in the model. Models were adjusted for age, gender, race, whether the admission was during day or night, PMN in peritoneal fluid, MELD-Na score, platelet count on the day of procedure, presence or absence of cirrhosis, diagnosis-related groups weight, primary clinical service, and the group performing paracentesis. To address heterogeneity among patients included in our study and the fact that some patients had multiple paracenteses, we conducted sensitivity analyses by excluding all noncirrhotic patients and including only the first paracentesis. A P value less than .05 was considered significant. All statistical analyses were performed using Stata MP 16.0 for Windows (StataCorp LLC).

Of the 1,321 paracenteses included in our study, 509 (38.5%) were performed by the procedure service, 723 (54.7%) by residents, and 89 (6.7%) by radiology. For comparison, 15.4% of procedures were performed by the radiology service during the 3 years before the start of the procedure service. More than 50% of the first paracenteses were performed within 30 hours of admission. Hospitalists or residents under the direct supervision of a hospitalist performed all paracenteses. Residents performing paracenteses, when not on the procedure service, were on general internal medicine, gastroenterology, hematology and oncology, or surgical services. No failed paracentesis attempts by the procedure service were subsequently performed by radiology. The mean age of the participants was 55.3 (12.2) years, 728 (55%) were White, 502 (38%) were female, and SBP was present in 61 (4.6%) patients. There was no difference by age, gender, time of admission, presence of SBP, or peritoneal fluid PMN in patients who underwent paracentesis by the procedure service versus controls (Table 1). A higher proportion of White patients and patients with cirrhosis underwent paracentesis by the procedure service than by another service. The LOS and A2P time were significantly lower for patients who underwent paracentesis by the procedure service than by another service (Table 1). When examining the adjusted linear secular time trends, LOS decreased by 0.1% per month (95% CI, –0.5% to 0.8%; P = .67) and A2P time by 0.02% per month (95% CI, –1.0% to 1.1%; P = .96).

In unadjusted models but accounting for secular time trends, patients who had paracenteses performed by residents or by radiology had a 50% (95% CI, 22%-83%; P = .002) and 127% (95% CI, 65%-211%; P < .001) longer LOS, respectively, than when paracentesis was performed by the procedure service. After adjusting for potential confounders, the difference in LOS between radiology and the procedure service remained significant; patients who had a paracentesis performed by radiology had a 27% (95% CI, 2%-58%; P = .03) longer LOS than patients who had the procedure performed by the procedure service. This relative LOS translates into 88 (95% CI, 1-174 hours) additional hours in absolute LOS. There was no difference in LOS between the procedure service and residents in the adjusted analysis (Table 2).

Similarly, in unadjusted models for A2P time and accounting for secular time trends, patients who had a paracentesis performed by residents or by radiology had a 52% (95% CI, 23%-88%; P < .001) and 173% (95% CI, 109%-280%; P < .001) longer A2P time, respectively, than patients whose paracentesis was performed by the procedure service. After adjusting for potential confounders, the difference in A2P time between radiology and the procedure service remained significant. Patients who had paracentesis performed by radiology had a 40% (95% CI, 5%-87%; P = .02) longer A2P time than patients who had paracentesis performed by the procedure service. This relative increase translates into 52 (95% CI, 3.3-101 hours) additional hours in absolute A2P time. On the other hand, residents had a significantly shorter A2P time (–19%, 95% CI, –33% to 0.2%; P = .05) (Table 2).

In the sensitivity analysis, excluding noncirrhotic patients and including only the first paracentesis for patients who had multiple procedures performed during admission, the results remained unchanged. In adjusted analysis, patients who had paracentesis performed by radiology had a 47% (95% CI, 3.7%-108%; P = .03) longer LOS and 91% (95% CI, 19%-107%; P = .008) longer A2P time than when paracentesis was performed by the procedure service. There were no differences in LOS or A2P time between the procedure service and residents (Table 2).

In this study, we report that a hospitalist-run procedure service, when compared with a radiology service, is associated with decreased LOS and A2P time independent of studied potential confounders and secular time trends. We also showed that, compared with radiology, the A2P time for nonemergent procedures (those performed 6 hours after admission) was not adversely affected by the procedure service. Residents performing paracenteses independently had shorter A2P time than the procedure service.

Although several institutions have bedside procedure services, data are lacking on benefits. Previously, paracenteses performed by residents have been associated with decreased LOS and need for transfusions when compared with radiology.⁷ Our study extends these findings to show a shortened A2P time. Delays may occur when a patient is referred to radiology because of volume, triaging of higher-acuity procedures, and transportation. Procedure services provide consistent attending supervision, more procedures by upper-level residents, and a lower rate of unsuccessful procedures.^12,15 Current study findings support the importance of continuing bedside procedure training for at least those residents who are interested in hospital medicine.⁷

Our study has several strengths and some potential limitations. The study examined outcomes that are important to patients as well as hospital administrators; it also had a large sample size, spanning 3 years. As it was a retrospective cohort study, there is potential for residual confounding due to unmeasured confounders. We did not examine the potential effect modification of procedure urgency, as such data are difficult to discern. Our method of identification missed patients who received therapeutic paracentesis without laboratory analysis. It is unclear why more White patients were referred to the procedure team; this is an area for further evaluation. Results of this study are likely not generalizable to institutions with a robust radiology service that has built-in redundancy to accommodate urgent procedures and easy availability over the weekends.

We found that a hospitalist-run teaching procedure service is associated with shorter LOS and A2P time. Further research is needed to determine if the benefits of a procedure service extend to lowering morbidity and/or mortality, as well as to determine the cost-effectiveness of a procedure service and whether the significant investment by the institution in establishing a procedure service is mitigated by the gains from better patient outcomes and reduced LOS.

Peritoneal fluid examination is often recommended for hospitalized patients with ascites.¹ The prevalence of spontaneous bacterial peritonitis (SBP) in these patients ranges from 10% to 30%.^2-6 Bedside paracentesis has clinical outcomes similar to that performed by radiology, with an improved length of stay (LOS) and decreased transfusion requirements.⁷

Internal medicine residency programs are establishing procedure services to address concerns about resident training in procedures and patient safety. Previous studies, which include paracentesis in patients with cirrhosis, have focused on resident comfort with procedures, supervision, procedural complications, and patient satisfaction.^8-12 However, the impact of a procedure service on the time from admission to the procedure has not been studied. In this study, we aimed to examine whether the institution of a hospitalist-run procedure service affected a patient’s LOS in the hospital and the time difference between a patient’s hospital admission and paracentesis (A2P).

An inpatient hospitalist-run procedure service was introduced on July 1, 2016. The service was staffed by a hospitalist and second-year internal medicine residents. The service is available 7:00 am to 5:00 pm all days of the week. To identify patients who underwent paracentesis, we queried our electronic medical records for all peritoneal fluid samples from July 1, 2016, to May 31, 2019. Paracenteses performed in the outpatient clinics, in the radiology suite, or in the emergency department were excluded if the patient was not admitted. We also excluded patients who had paracentesis within 6 hours of presentation, as these patients likely had an urgent clinical indication for paracentesis.

Data on age, gender, race, ethnicity, date and time of hospital admission, and discharge date and time were retrieved. We also retrieved data on the absolute number of polymorphonuclear leukocytes (PMN) in the peritoneal fluid sample; a patient with a count higher than 250/uL was considered to have SBP. The timestamp for the peritoneal fluid results was used to approximate the A2P time. Paracenteses performed by or under direct supervision of procedure service hospitalists were identified through a procedure log maintained by procedure service hospitalists. We generated a binary variable to differentiate patients who were admitted during the day from those admitted during the night, when the procedure service was not available. For all patients, we calculated the model for end-stage liver disease and sodium (MELD-Na) score.¹³ Groups performing paracenteses were categorized into procedure service, residents, and radiology. Primary clinical services were categorized into general medicine, gastroenterology, surgery, and others.

Data were summarized as mean (SD) or median (interquartile range) for continuous variables and as percentages for categorical variables. Patients who had paracenteses by radiology or residents during the study period were considered controls. We used concurrent controls to address secular time trends (eg, measures to decrease LOS or changes in ordering tests in the electronic health record) in outcome measures. Patient characteristics were compared using the Wilcoxon rank-sum test or the χ² test, as appropriate.

Two outcome variables were examined: LOS, and A2P time. Because both outcome variables were right skewed, we used generalized linear models with gamma distribution and log link. The advantage of a generalized linear model approach is that the transformed coefficients are better interpretable than when using the log transformation of the response variable.¹⁴ To account for time trends, we included time in months in the model. Models were adjusted for age, gender, race, whether the admission was during day or night, PMN in peritoneal fluid, MELD-Na score, platelet count on the day of procedure, presence or absence of cirrhosis, diagnosis-related groups weight, primary clinical service, and the group performing paracentesis. To address heterogeneity among patients included in our study and the fact that some patients had multiple paracenteses, we conducted sensitivity analyses by excluding all noncirrhotic patients and including only the first paracentesis. A P value less than .05 was considered significant. All statistical analyses were performed using Stata MP 16.0 for Windows (StataCorp LLC).

Of the 1,321 paracenteses included in our study, 509 (38.5%) were performed by the procedure service, 723 (54.7%) by residents, and 89 (6.7%) by radiology. For comparison, 15.4% of procedures were performed by the radiology service during the 3 years before the start of the procedure service. More than 50% of the first paracenteses were performed within 30 hours of admission. Hospitalists or residents under the direct supervision of a hospitalist performed all paracenteses. Residents performing paracenteses, when not on the procedure service, were on general internal medicine, gastroenterology, hematology and oncology, or surgical services. No failed paracentesis attempts by the procedure service were subsequently performed by radiology. The mean age of the participants was 55.3 (12.2) years, 728 (55%) were White, 502 (38%) were female, and SBP was present in 61 (4.6%) patients. There was no difference by age, gender, time of admission, presence of SBP, or peritoneal fluid PMN in patients who underwent paracentesis by the procedure service versus controls (Table 1). A higher proportion of White patients and patients with cirrhosis underwent paracentesis by the procedure service than by another service. The LOS and A2P time were significantly lower for patients who underwent paracentesis by the procedure service than by another service (Table 1). When examining the adjusted linear secular time trends, LOS decreased by 0.1% per month (95% CI, –0.5% to 0.8%; P = .67) and A2P time by 0.02% per month (95% CI, –1.0% to 1.1%; P = .96).

In unadjusted models but accounting for secular time trends, patients who had paracenteses performed by residents or by radiology had a 50% (95% CI, 22%-83%; P = .002) and 127% (95% CI, 65%-211%; P < .001) longer LOS, respectively, than when paracentesis was performed by the procedure service. After adjusting for potential confounders, the difference in LOS between radiology and the procedure service remained significant; patients who had a paracentesis performed by radiology had a 27% (95% CI, 2%-58%; P = .03) longer LOS than patients who had the procedure performed by the procedure service. This relative LOS translates into 88 (95% CI, 1-174 hours) additional hours in absolute LOS. There was no difference in LOS between the procedure service and residents in the adjusted analysis (Table 2).

Similarly, in unadjusted models for A2P time and accounting for secular time trends, patients who had a paracentesis performed by residents or by radiology had a 52% (95% CI, 23%-88%; P < .001) and 173% (95% CI, 109%-280%; P < .001) longer A2P time, respectively, than patients whose paracentesis was performed by the procedure service. After adjusting for potential confounders, the difference in A2P time between radiology and the procedure service remained significant. Patients who had paracentesis performed by radiology had a 40% (95% CI, 5%-87%; P = .02) longer A2P time than patients who had paracentesis performed by the procedure service. This relative increase translates into 52 (95% CI, 3.3-101 hours) additional hours in absolute A2P time. On the other hand, residents had a significantly shorter A2P time (–19%, 95% CI, –33% to 0.2%; P = .05) (Table 2).

In the sensitivity analysis, excluding noncirrhotic patients and including only the first paracentesis for patients who had multiple procedures performed during admission, the results remained unchanged. In adjusted analysis, patients who had paracentesis performed by radiology had a 47% (95% CI, 3.7%-108%; P = .03) longer LOS and 91% (95% CI, 19%-107%; P = .008) longer A2P time than when paracentesis was performed by the procedure service. There were no differences in LOS or A2P time between the procedure service and residents (Table 2).

In this study, we report that a hospitalist-run procedure service, when compared with a radiology service, is associated with decreased LOS and A2P time independent of studied potential confounders and secular time trends. We also showed that, compared with radiology, the A2P time for nonemergent procedures (those performed 6 hours after admission) was not adversely affected by the procedure service. Residents performing paracenteses independently had shorter A2P time than the procedure service.

Although several institutions have bedside procedure services, data are lacking on benefits. Previously, paracenteses performed by residents have been associated with decreased LOS and need for transfusions when compared with radiology.⁷ Our study extends these findings to show a shortened A2P time. Delays may occur when a patient is referred to radiology because of volume, triaging of higher-acuity procedures, and transportation. Procedure services provide consistent attending supervision, more procedures by upper-level residents, and a lower rate of unsuccessful procedures.^12,15 Current study findings support the importance of continuing bedside procedure training for at least those residents who are interested in hospital medicine.⁷

Our study has several strengths and some potential limitations. The study examined outcomes that are important to patients as well as hospital administrators; it also had a large sample size, spanning 3 years. As it was a retrospective cohort study, there is potential for residual confounding due to unmeasured confounders. We did not examine the potential effect modification of procedure urgency, as such data are difficult to discern. Our method of identification missed patients who received therapeutic paracentesis without laboratory analysis. It is unclear why more White patients were referred to the procedure team; this is an area for further evaluation. Results of this study are likely not generalizable to institutions with a robust radiology service that has built-in redundancy to accommodate urgent procedures and easy availability over the weekends.

We found that a hospitalist-run teaching procedure service is associated with shorter LOS and A2P time. Further research is needed to determine if the benefits of a procedure service extend to lowering morbidity and/or mortality, as well as to determine the cost-effectiveness of a procedure service and whether the significant investment by the institution in establishing a procedure service is mitigated by the gains from better patient outcomes and reduced LOS.

Hospital medicine research often asks the question whether an intervention, such as a policy or guideline, has improved quality of care and/or whether there were any unintended consequences. Alternatively, investigators may be interested in understanding the impact of an event, such as a natural disaster or a pandemic, on hospital care. The study design that provides the best estimate of the causal effect of the intervention is the randomized controlled trial (RCT). The goal of randomization, which can be implemented at the patient or cluster level (eg, hospitals), is attaining a balance of the known and unknown confounders between study groups.

However, an RCT may not be feasible for several reasons: complexity, insufficient setup time or funding, ethical barriers to randomization, unwillingness of funders or payers to withhold the intervention from patients (ie, the control group), or anticipated contamination of the intervention into the control group (eg, provider practice change interventions). In addition, it may be impossible to conduct an RCT because the investigator does not have control over the design of an intervention or because they are studying an event, such as a pandemic.

In the June 2020 issue of the Journal of Hospital Medicine, Coon et al¹ use a type of quasi-experimental design (QED)—specifically, the interrupted time series (ITS)—to examine the impact of the adoption of ward-based high-flow nasal cannula protocols on intensive care unit (ICU) admission for bronchiolitis at children’s hospitals. In this methodologic progress note, we discuss QEDs for evaluating the impact of healthcare interventions or events and focus on ITS, one of the strongest QEDs.

Quasi-experimental design refers to a broad range of nonrandomized or partially randomized pre- vs postintervention studies.² In order to test a causal hypothesis without randomization, QEDs define a comparison group or a time period in which an intervention has not been implemented, as well as at least one group or time period in which an intervention has been implemented. In a QED, the control may lack similarity with the intervention group or time period because of differences in the patients, sites, or time period (sometimes referred to as having a “nonequivalent control group”). Several design and analytic approaches are available to enhance the extent to which the study is able to make conclusions about the causal impact of the intervention.^2,3 Because randomization is not necessary, QEDs allow for inclusion of a broader population than that which is feasible by RCTs, which increases the applicability and generalizability of the results. Therefore, they are a powerful research design to test the effectiveness of interventions in real-world settings.

The choice of which QED depends on whether the investigators are conducting a prospective evaluation and have control over the study design (ie, the ordering of the intervention, selection of sites or individuals, and/or timing and frequency of the data collection) or whether the investigators do not have control over the intervention, which is also known as a “natural experiment.”^4,5 Some studies may also incorporate two QEDs in tandem.⁶ The Table provides a brief summary of different QEDs, ordered by methodologic strength, and distinguishes those that can be used to study natural experiments. In the study by Coon et al,¹ an ITS is used as opposed to a methodologically stronger QED, such as the stepped-wedge design, because the investigators did not have control over the rollout of heated high-flow nasal canula protocols across hospitals.

Interrupted time series designs use repeated observations of an outcome over time. This method then divides, or “interrupts,” the series of data into two time periods: before the intervention or event and after. Using data from the preintervention period, an underlying trend in the outcome is estimated and assumed to continue forward into the postintervention period to estimate what would have occurred without the intervention. Any significant change in the outcome at the beginning of the postintervention period or change in the trend in the postintervention is then attributed to the intervention.

There are several important methodologic considerations when designing an ITS study, as detailed in other review papers.^2,3,7,8 An ITS design can be retrospective or prospective. It can be of a single center or include multiple sites, as in Coon et al. It can be conducted with or without a control. The inclusion of a control, when appropriately chosen, improves the strength of the study design because it can account for seasonal trends and potential confounders that vary over time. The control can be a different group of hospitals or participants that are similar but did not receive the intervention, or it can be a different outcome in the same group of hospitals or participants that are not expected to be affected by the intervention. The ITS design may also be set up to estimate the individual effects of multicomponent interventions. If the different components are phased in sequentially over time, then it may be possible to interrupt the time series at these points and estimate the impact of each intervention component.

Other examples of ITS studies in hospital medicine include those that evaluated the impact of a readmission-reduction program,⁹ of state sepsis regulations on in-hospital mortality,¹⁰ of resident duty-hour reform on mortality among hospitalized patients,¹¹ of a quality-improvement initiative on early discharge,¹² and of national guidelines on pediatric pneumonia antibiotic selection.¹³ There are several types of ITS analysis, and in this article, we focus on segmented regression without a control group.^7,8

Segmented regression is the statistical model used to measure (a) the immediate change in the outcome (level) at the start of the intervention and (b) the change in the trend of the outcome (slope) in the postintervention period vs that in the preintervention period. Therefore, the intervention effect size is expressed in terms of the level change and the slope change. To function properly, the models require several repeated (eg, monthly) measurements of the outcome before and after the intervention. Some experts suggest a minimum of 4 to 12 observations, depending on a number of factors including the stability of the outcome and seasonal variations.^7,8 If changes before and after more than one intervention are being examined, there should be the minimum number of observations separating them. Unlike typical regression models, time-series models can correct for autocorrelation if it is present in the data. Autocorrelation is the type of correlation that arises when data are collected over time, with those closest in time being more strongly correlated (there are also other types of autocorrelation, such as seasonal patterns). Using available statistical software, autocorrelation can be detected and, if present, it can be controlled for in the segmented regression models.

Coon et al present results of their ITS analysis in a panel of figures detailing each study outcome, ICU admission, ICU length of stay, total length of stay, and rates of mechanical ventilation. Each panel shows the rate of change in the outcome per season across hospitals, before and after adoption of heated high-flow nasal cannula protocols, and the level change at the time of adoption.

To further explain how segmented regression results are presented, in the Figure we detail the structure of a segmented regression figure evaluating the impact of an intervention without a control group. In addition to the regression figure, authors typically provide 95% CIs around the rates, level change, and the difference between the postintervention and preintervention periods, along with P values demonstrating whether the rates, level change, and the differences between period slopes differ significantly from zero.

Segmented regression models assume a linear trend in the outcome. If the outcome follows a nonlinear pattern (eg, exponential spread of a disease during a pandemic), then using different distributions in the modeling or transformations of the data may be necessary. The validity of the comparison between the pre- and postintervention groups relies on the similarity between the populations. When there is imbalance, investigators can consider matching based on important characteristics or applying risk adjustment as necessary. Another important assumption is that the outcome of interest is unchanged in the absence of the intervention. Finally, the analysis assumes that the intervention is fully implemented at the time the postintervention period begins. Often, there is a washout period during which the old approach is stopped and the new approach (the intervention) is being implemented and can easily be taken into account.

There are several strengths of the ITS analysis and segmented regression.^7,8 First, this approach accounts for a possible secular trend in the outcome measure that may have been present prior to the intervention. For example, investigators might conclude that a readmissions program was effective in reducing readmissions if they found that the mean readmission percentage in the period after the intervention was significantly lower than before using a simple pre/post study design. However, what if the readmission rate was already going down prior to the intervention? Using an ITS approach, they may have found that the rate of readmissions simply continued to decrease after the intervention at the same rate that it was decreasing prior to the intervention and, therefore, conclude that the intervention was not effective. Second, because the ITS approach evaluates changes in rates of an outcome at a population level, confounding by individual-level variables will not introduce serious bias unless the confounding occurred at the same time as the intervention. Third, ITS can be used to measure the unintended consequences of interventions or events, and investigators can construct separate time-series analyses for different outcomes. Fourth, ITS can be used to evaluate the impact of the intervention on subpopulations (eg, those grouped by age, sex, race) by conducting stratified analysis. Fifth, ITS provides simple and clear graphical results that can be easily understood by various audiences.

By accounting for preintervention trends, ITS studies permit stronger causal inference than do cross-sectional or simple pre/post QEDs, but they may by prone to confounding by cointerventions or by changes in the population composition. Causal inference based on the ITS analysis is only valid to the extent to which the intervention was the only thing that changed at the point in time between the preintervention and postintervention periods. It is important for investigators to consider this in the design and discuss any coincident interventions. If there are multiple interventions over time, it is possible to account for these changes in the study design by creating multiple points of interruption provided there are sufficient measurements of the outcome between interventions. If the composition of the population changes at the same time as the intervention, this introduces bias. Changes in the ability to measure the outcome or changes to its definition also threaten the validity of the study’s inferences. Finally, it is also important to remember that when the outcome is a population-level measurement, inferences about individual-level outcomes are inappropriate due to ecological fallacies (ie, when inferences about individuals are deduced from inferences about the group to which those individuals belong). For example, Coon et al found that infants with bronchiolitis in the ward-based high-flow nasal cannula protocol group had greater ICU admission rates. It would be inappropriate to conclude that, based on this, an individual infant in a hospital on a ward-based protocol is more likely to be admitted to the ICU.

Studies evaluating interventions and events are important for informing healthcare practice, policy, and public health. While an RCT is the preferred method for such evaluations, investigators must often consider alternative study designs when an RCT is not feasible or when more real-world outcome evaluation is desired. Quasi-experimental designs are employed in studies that do not use randomization to study the impact of interventions in real-world settings, and an interrupted time series is a strong QED for the evaluation of interventions and natural experiments.

Hospital medicine research often asks the question whether an intervention, such as a policy or guideline, has improved quality of care and/or whether there were any unintended consequences. Alternatively, investigators may be interested in understanding the impact of an event, such as a natural disaster or a pandemic, on hospital care. The study design that provides the best estimate of the causal effect of the intervention is the randomized controlled trial (RCT). The goal of randomization, which can be implemented at the patient or cluster level (eg, hospitals), is attaining a balance of the known and unknown confounders between study groups.

However, an RCT may not be feasible for several reasons: complexity, insufficient setup time or funding, ethical barriers to randomization, unwillingness of funders or payers to withhold the intervention from patients (ie, the control group), or anticipated contamination of the intervention into the control group (eg, provider practice change interventions). In addition, it may be impossible to conduct an RCT because the investigator does not have control over the design of an intervention or because they are studying an event, such as a pandemic.

In the June 2020 issue of the Journal of Hospital Medicine, Coon et al¹ use a type of quasi-experimental design (QED)—specifically, the interrupted time series (ITS)—to examine the impact of the adoption of ward-based high-flow nasal cannula protocols on intensive care unit (ICU) admission for bronchiolitis at children’s hospitals. In this methodologic progress note, we discuss QEDs for evaluating the impact of healthcare interventions or events and focus on ITS, one of the strongest QEDs.

Quasi-experimental design refers to a broad range of nonrandomized or partially randomized pre- vs postintervention studies.² In order to test a causal hypothesis without randomization, QEDs define a comparison group or a time period in which an intervention has not been implemented, as well as at least one group or time period in which an intervention has been implemented. In a QED, the control may lack similarity with the intervention group or time period because of differences in the patients, sites, or time period (sometimes referred to as having a “nonequivalent control group”). Several design and analytic approaches are available to enhance the extent to which the study is able to make conclusions about the causal impact of the intervention.^2,3 Because randomization is not necessary, QEDs allow for inclusion of a broader population than that which is feasible by RCTs, which increases the applicability and generalizability of the results. Therefore, they are a powerful research design to test the effectiveness of interventions in real-world settings.

The choice of which QED depends on whether the investigators are conducting a prospective evaluation and have control over the study design (ie, the ordering of the intervention, selection of sites or individuals, and/or timing and frequency of the data collection) or whether the investigators do not have control over the intervention, which is also known as a “natural experiment.”^4,5 Some studies may also incorporate two QEDs in tandem.⁶ The Table provides a brief summary of different QEDs, ordered by methodologic strength, and distinguishes those that can be used to study natural experiments. In the study by Coon et al,¹ an ITS is used as opposed to a methodologically stronger QED, such as the stepped-wedge design, because the investigators did not have control over the rollout of heated high-flow nasal canula protocols across hospitals.

Interrupted time series designs use repeated observations of an outcome over time. This method then divides, or “interrupts,” the series of data into two time periods: before the intervention or event and after. Using data from the preintervention period, an underlying trend in the outcome is estimated and assumed to continue forward into the postintervention period to estimate what would have occurred without the intervention. Any significant change in the outcome at the beginning of the postintervention period or change in the trend in the postintervention is then attributed to the intervention.

There are several important methodologic considerations when designing an ITS study, as detailed in other review papers.^2,3,7,8 An ITS design can be retrospective or prospective. It can be of a single center or include multiple sites, as in Coon et al. It can be conducted with or without a control. The inclusion of a control, when appropriately chosen, improves the strength of the study design because it can account for seasonal trends and potential confounders that vary over time. The control can be a different group of hospitals or participants that are similar but did not receive the intervention, or it can be a different outcome in the same group of hospitals or participants that are not expected to be affected by the intervention. The ITS design may also be set up to estimate the individual effects of multicomponent interventions. If the different components are phased in sequentially over time, then it may be possible to interrupt the time series at these points and estimate the impact of each intervention component.

Other examples of ITS studies in hospital medicine include those that evaluated the impact of a readmission-reduction program,⁹ of state sepsis regulations on in-hospital mortality,¹⁰ of resident duty-hour reform on mortality among hospitalized patients,¹¹ of a quality-improvement initiative on early discharge,¹² and of national guidelines on pediatric pneumonia antibiotic selection.¹³ There are several types of ITS analysis, and in this article, we focus on segmented regression without a control group.^7,8

Segmented regression is the statistical model used to measure (a) the immediate change in the outcome (level) at the start of the intervention and (b) the change in the trend of the outcome (slope) in the postintervention period vs that in the preintervention period. Therefore, the intervention effect size is expressed in terms of the level change and the slope change. To function properly, the models require several repeated (eg, monthly) measurements of the outcome before and after the intervention. Some experts suggest a minimum of 4 to 12 observations, depending on a number of factors including the stability of the outcome and seasonal variations.^7,8 If changes before and after more than one intervention are being examined, there should be the minimum number of observations separating them. Unlike typical regression models, time-series models can correct for autocorrelation if it is present in the data. Autocorrelation is the type of correlation that arises when data are collected over time, with those closest in time being more strongly correlated (there are also other types of autocorrelation, such as seasonal patterns). Using available statistical software, autocorrelation can be detected and, if present, it can be controlled for in the segmented regression models.

Coon et al present results of their ITS analysis in a panel of figures detailing each study outcome, ICU admission, ICU length of stay, total length of stay, and rates of mechanical ventilation. Each panel shows the rate of change in the outcome per season across hospitals, before and after adoption of heated high-flow nasal cannula protocols, and the level change at the time of adoption.

To further explain how segmented regression results are presented, in the Figure we detail the structure of a segmented regression figure evaluating the impact of an intervention without a control group. In addition to the regression figure, authors typically provide 95% CIs around the rates, level change, and the difference between the postintervention and preintervention periods, along with P values demonstrating whether the rates, level change, and the differences between period slopes differ significantly from zero.

Segmented regression models assume a linear trend in the outcome. If the outcome follows a nonlinear pattern (eg, exponential spread of a disease during a pandemic), then using different distributions in the modeling or transformations of the data may be necessary. The validity of the comparison between the pre- and postintervention groups relies on the similarity between the populations. When there is imbalance, investigators can consider matching based on important characteristics or applying risk adjustment as necessary. Another important assumption is that the outcome of interest is unchanged in the absence of the intervention. Finally, the analysis assumes that the intervention is fully implemented at the time the postintervention period begins. Often, there is a washout period during which the old approach is stopped and the new approach (the intervention) is being implemented and can easily be taken into account.

There are several strengths of the ITS analysis and segmented regression.^7,8 First, this approach accounts for a possible secular trend in the outcome measure that may have been present prior to the intervention. For example, investigators might conclude that a readmissions program was effective in reducing readmissions if they found that the mean readmission percentage in the period after the intervention was significantly lower than before using a simple pre/post study design. However, what if the readmission rate was already going down prior to the intervention? Using an ITS approach, they may have found that the rate of readmissions simply continued to decrease after the intervention at the same rate that it was decreasing prior to the intervention and, therefore, conclude that the intervention was not effective. Second, because the ITS approach evaluates changes in rates of an outcome at a population level, confounding by individual-level variables will not introduce serious bias unless the confounding occurred at the same time as the intervention. Third, ITS can be used to measure the unintended consequences of interventions or events, and investigators can construct separate time-series analyses for different outcomes. Fourth, ITS can be used to evaluate the impact of the intervention on subpopulations (eg, those grouped by age, sex, race) by conducting stratified analysis. Fifth, ITS provides simple and clear graphical results that can be easily understood by various audiences.

By accounting for preintervention trends, ITS studies permit stronger causal inference than do cross-sectional or simple pre/post QEDs, but they may by prone to confounding by cointerventions or by changes in the population composition. Causal inference based on the ITS analysis is only valid to the extent to which the intervention was the only thing that changed at the point in time between the preintervention and postintervention periods. It is important for investigators to consider this in the design and discuss any coincident interventions. If there are multiple interventions over time, it is possible to account for these changes in the study design by creating multiple points of interruption provided there are sufficient measurements of the outcome between interventions. If the composition of the population changes at the same time as the intervention, this introduces bias. Changes in the ability to measure the outcome or changes to its definition also threaten the validity of the study’s inferences. Finally, it is also important to remember that when the outcome is a population-level measurement, inferences about individual-level outcomes are inappropriate due to ecological fallacies (ie, when inferences about individuals are deduced from inferences about the group to which those individuals belong). For example, Coon et al found that infants with bronchiolitis in the ward-based high-flow nasal cannula protocol group had greater ICU admission rates. It would be inappropriate to conclude that, based on this, an individual infant in a hospital on a ward-based protocol is more likely to be admitted to the ICU.

Studies evaluating interventions and events are important for informing healthcare practice, policy, and public health. While an RCT is the preferred method for such evaluations, investigators must often consider alternative study designs when an RCT is not feasible or when more real-world outcome evaluation is desired. Quasi-experimental designs are employed in studies that do not use randomization to study the impact of interventions in real-world settings, and an interrupted time series is a strong QED for the evaluation of interventions and natural experiments.

Hospital medicine research often asks the question whether an intervention, such as a policy or guideline, has improved quality of care and/or whether there were any unintended consequences. Alternatively, investigators may be interested in understanding the impact of an event, such as a natural disaster or a pandemic, on hospital care. The study design that provides the best estimate of the causal effect of the intervention is the randomized controlled trial (RCT). The goal of randomization, which can be implemented at the patient or cluster level (eg, hospitals), is attaining a balance of the known and unknown confounders between study groups.

However, an RCT may not be feasible for several reasons: complexity, insufficient setup time or funding, ethical barriers to randomization, unwillingness of funders or payers to withhold the intervention from patients (ie, the control group), or anticipated contamination of the intervention into the control group (eg, provider practice change interventions). In addition, it may be impossible to conduct an RCT because the investigator does not have control over the design of an intervention or because they are studying an event, such as a pandemic.

In the June 2020 issue of the Journal of Hospital Medicine, Coon et al¹ use a type of quasi-experimental design (QED)—specifically, the interrupted time series (ITS)—to examine the impact of the adoption of ward-based high-flow nasal cannula protocols on intensive care unit (ICU) admission for bronchiolitis at children’s hospitals. In this methodologic progress note, we discuss QEDs for evaluating the impact of healthcare interventions or events and focus on ITS, one of the strongest QEDs.

Quasi-experimental design refers to a broad range of nonrandomized or partially randomized pre- vs postintervention studies.² In order to test a causal hypothesis without randomization, QEDs define a comparison group or a time period in which an intervention has not been implemented, as well as at least one group or time period in which an intervention has been implemented. In a QED, the control may lack similarity with the intervention group or time period because of differences in the patients, sites, or time period (sometimes referred to as having a “nonequivalent control group”). Several design and analytic approaches are available to enhance the extent to which the study is able to make conclusions about the causal impact of the intervention.^2,3 Because randomization is not necessary, QEDs allow for inclusion of a broader population than that which is feasible by RCTs, which increases the applicability and generalizability of the results. Therefore, they are a powerful research design to test the effectiveness of interventions in real-world settings.

The choice of which QED depends on whether the investigators are conducting a prospective evaluation and have control over the study design (ie, the ordering of the intervention, selection of sites or individuals, and/or timing and frequency of the data collection) or whether the investigators do not have control over the intervention, which is also known as a “natural experiment.”^4,5 Some studies may also incorporate two QEDs in tandem.⁶ The Table provides a brief summary of different QEDs, ordered by methodologic strength, and distinguishes those that can be used to study natural experiments. In the study by Coon et al,¹ an ITS is used as opposed to a methodologically stronger QED, such as the stepped-wedge design, because the investigators did not have control over the rollout of heated high-flow nasal canula protocols across hospitals.

Interrupted time series designs use repeated observations of an outcome over time. This method then divides, or “interrupts,” the series of data into two time periods: before the intervention or event and after. Using data from the preintervention period, an underlying trend in the outcome is estimated and assumed to continue forward into the postintervention period to estimate what would have occurred without the intervention. Any significant change in the outcome at the beginning of the postintervention period or change in the trend in the postintervention is then attributed to the intervention.

There are several important methodologic considerations when designing an ITS study, as detailed in other review papers.^2,3,7,8 An ITS design can be retrospective or prospective. It can be of a single center or include multiple sites, as in Coon et al. It can be conducted with or without a control. The inclusion of a control, when appropriately chosen, improves the strength of the study design because it can account for seasonal trends and potential confounders that vary over time. The control can be a different group of hospitals or participants that are similar but did not receive the intervention, or it can be a different outcome in the same group of hospitals or participants that are not expected to be affected by the intervention. The ITS design may also be set up to estimate the individual effects of multicomponent interventions. If the different components are phased in sequentially over time, then it may be possible to interrupt the time series at these points and estimate the impact of each intervention component.

Other examples of ITS studies in hospital medicine include those that evaluated the impact of a readmission-reduction program,⁹ of state sepsis regulations on in-hospital mortality,¹⁰ of resident duty-hour reform on mortality among hospitalized patients,¹¹ of a quality-improvement initiative on early discharge,¹² and of national guidelines on pediatric pneumonia antibiotic selection.¹³ There are several types of ITS analysis, and in this article, we focus on segmented regression without a control group.^7,8

Segmented regression is the statistical model used to measure (a) the immediate change in the outcome (level) at the start of the intervention and (b) the change in the trend of the outcome (slope) in the postintervention period vs that in the preintervention period. Therefore, the intervention effect size is expressed in terms of the level change and the slope change. To function properly, the models require several repeated (eg, monthly) measurements of the outcome before and after the intervention. Some experts suggest a minimum of 4 to 12 observations, depending on a number of factors including the stability of the outcome and seasonal variations.^7,8 If changes before and after more than one intervention are being examined, there should be the minimum number of observations separating them. Unlike typical regression models, time-series models can correct for autocorrelation if it is present in the data. Autocorrelation is the type of correlation that arises when data are collected over time, with those closest in time being more strongly correlated (there are also other types of autocorrelation, such as seasonal patterns). Using available statistical software, autocorrelation can be detected and, if present, it can be controlled for in the segmented regression models.

Coon et al present results of their ITS analysis in a panel of figures detailing each study outcome, ICU admission, ICU length of stay, total length of stay, and rates of mechanical ventilation. Each panel shows the rate of change in the outcome per season across hospitals, before and after adoption of heated high-flow nasal cannula protocols, and the level change at the time of adoption.

To further explain how segmented regression results are presented, in the Figure we detail the structure of a segmented regression figure evaluating the impact of an intervention without a control group. In addition to the regression figure, authors typically provide 95% CIs around the rates, level change, and the difference between the postintervention and preintervention periods, along with P values demonstrating whether the rates, level change, and the differences between period slopes differ significantly from zero.

Segmented regression models assume a linear trend in the outcome. If the outcome follows a nonlinear pattern (eg, exponential spread of a disease during a pandemic), then using different distributions in the modeling or transformations of the data may be necessary. The validity of the comparison between the pre- and postintervention groups relies on the similarity between the populations. When there is imbalance, investigators can consider matching based on important characteristics or applying risk adjustment as necessary. Another important assumption is that the outcome of interest is unchanged in the absence of the intervention. Finally, the analysis assumes that the intervention is fully implemented at the time the postintervention period begins. Often, there is a washout period during which the old approach is stopped and the new approach (the intervention) is being implemented and can easily be taken into account.

There are several strengths of the ITS analysis and segmented regression.^7,8 First, this approach accounts for a possible secular trend in the outcome measure that may have been present prior to the intervention. For example, investigators might conclude that a readmissions program was effective in reducing readmissions if they found that the mean readmission percentage in the period after the intervention was significantly lower than before using a simple pre/post study design. However, what if the readmission rate was already going down prior to the intervention? Using an ITS approach, they may have found that the rate of readmissions simply continued to decrease after the intervention at the same rate that it was decreasing prior to the intervention and, therefore, conclude that the intervention was not effective. Second, because the ITS approach evaluates changes in rates of an outcome at a population level, confounding by individual-level variables will not introduce serious bias unless the confounding occurred at the same time as the intervention. Third, ITS can be used to measure the unintended consequences of interventions or events, and investigators can construct separate time-series analyses for different outcomes. Fourth, ITS can be used to evaluate the impact of the intervention on subpopulations (eg, those grouped by age, sex, race) by conducting stratified analysis. Fifth, ITS provides simple and clear graphical results that can be easily understood by various audiences.

By accounting for preintervention trends, ITS studies permit stronger causal inference than do cross-sectional or simple pre/post QEDs, but they may by prone to confounding by cointerventions or by changes in the population composition. Causal inference based on the ITS analysis is only valid to the extent to which the intervention was the only thing that changed at the point in time between the preintervention and postintervention periods. It is important for investigators to consider this in the design and discuss any coincident interventions. If there are multiple interventions over time, it is possible to account for these changes in the study design by creating multiple points of interruption provided there are sufficient measurements of the outcome between interventions. If the composition of the population changes at the same time as the intervention, this introduces bias. Changes in the ability to measure the outcome or changes to its definition also threaten the validity of the study’s inferences. Finally, it is also important to remember that when the outcome is a population-level measurement, inferences about individual-level outcomes are inappropriate due to ecological fallacies (ie, when inferences about individuals are deduced from inferences about the group to which those individuals belong). For example, Coon et al found that infants with bronchiolitis in the ward-based high-flow nasal cannula protocol group had greater ICU admission rates. It would be inappropriate to conclude that, based on this, an individual infant in a hospital on a ward-based protocol is more likely to be admitted to the ICU.

Studies evaluating interventions and events are important for informing healthcare practice, policy, and public health. While an RCT is the preferred method for such evaluations, investigators must often consider alternative study designs when an RCT is not feasible or when more real-world outcome evaluation is desired. Quasi-experimental designs are employed in studies that do not use randomization to study the impact of interventions in real-world settings, and an interrupted time series is a strong QED for the evaluation of interventions and natural experiments.

Inspired by the ABIM Foundation’s Choosing Wisel y ^® campaign, the “Things We Do for No Reason ^™ ” (TWDFNR) series reviews practices that have become common parts of hospital care but may provide little value to our patients. Practices reviewed in the TWDFNR series do not represent clear-cut conclusions or clinical practice standards but are meant as a starting place for research and active discussions among hospitalists and patients. We invite you to be part of that discussion.

The hospitalist admits an 18-year-old man for newly diagnosed granulomatosis with polyangiitis to receive expedited pulse-dose steroids and plasma exchange. After consulting interventional radiology for catheter placement the following day, the hospitalist places a “strict” nil per os (nothing by mouth, NPO) after midnight order. During rounds the following morning, the patient reports that he wants to eat. At 9 am, interventional radiology informs the nurse that the line placement will take place at 3 pm. Due to emergencies and other unplanned delays, the catheter placement occurs at 5 pm. The patient and family express their displeasure about the prolonged fasting and ask why this happened.

Hospitalists commonly order “NPO after midnight” diets in anticipation of procedures requiring sedation or general anesthesia. Typically, NPO refers to no food or drink, but in some instances, NPO includes no oral medications. Up to half of medical patients experience some time of fasting while hospitalized.¹ However, NPO practices vary widely across institutions.^2,3 A study from 2014 notes that, on average, patients fast preprocedure for approximately 13.5 hours for solids and 9.6 hours for liquids.² Prolonged fasting times offer little benefit to patients and may lead to frequent patient dissatisfaction and complaints.

In 1883, Sir Joseph Lister described 19th century NPO practices distinguishing solids from liquids, allowing patients “tea or beef tea” until 2 to 3 hours prior to surgery.⁴ However, in 1946, Mendelson published an influential account of 66 pregnant women who aspirated during delivery under general anesthesia.⁵ Two of the 66 patients, both of whom had eaten a full meal 6 to 8 hours prior to general anesthesia, died. The study not only increased awareness of the risk of aspiration with general anesthesia in pregnancy, but it influenced the care for the nonpregnant population of patients as well. By the 1960s, anesthesia texts recommended “NPO after midnight” for both liquids and solids in all patients, regardless of pregnancy status.⁴ To minimize the risk to patients, we have continued to pass down the practice of NPO after midnight to subsequent generations.

Additionally, medical centers and hospitals feel pressure to provide efficient, patient-centered, high-value care. Given the complexity of procedural scheduling and the penalties associated with delays, keeping patients NPO ensures their availability for the next open procedural slot. NPO after midnight orders aim to prevent potential delays in treatment that occur when inadvertent ingestion of food and drink leads to cancellation of procedures.

Recent studies have led to a more sophisticated understanding of gastric emptying and the risks of aspiration during sedation and intubation. Gastric emptying studies routinely show that transit of clear liquids out of the stomach is virtually complete within two hours of drinking.⁶ Age, body mass index, and alcohol have no effect on gastric emptying time, and almost all patients return to preingestion gastric residual volumes within 2 hours of clear liquid consumption.^6,7 While morbidly obese patients tend to have higher gastric fluid volumes after 9 hours of fasting, their stomachs empty at rates similar to nonobese individuals.⁶ Note that, regardless of fasting times, morbid obesity predisposes patients to a higher overall gastric volume and lower pH of gastric contents, which may increase risk of aspiration.⁸ A Cochrane review found no statistical difference in gastric volumes or stomach pH in patients on a standard fast vs shortened (<180 minutes) liquid fast.⁹ The review included nine studies that found patients who consumed a clear liquid beverage had reduced gastric volumes, compared with patients in a fasting state (P < .001).⁹

In a pediatric retrospective study of pulmonary aspiration events, the researchers demonstrated that clinically significant aspiration (presence of bilious secretions in the tracheobronchial airways) occurred at a rate of 0.04% with emergency surgery.¹⁰ Bowel obstruction or ileus accounted for approximately 54% of those cases. Importantly, the reported aspiration rate approximates the rate of pregnant patients from the 1946 Mendelson study of 0.14% (66 out of 44,016), which originally prompted the use of the prolonged NPO status. Based on the Cochrane review of perioperative fasting recommendations for those older than 18 years, consuming fluids more than 90 minutes preoperatively confers a negligible (0 adverse events reported in 9 studies) risk for aspiration or regurgitation events.⁹

In 1998, as a result of these and other similar studies, the American Society of Anesthesiologists (ASA) along with global anesthesia partners adopted guidelines that allowed clear liquids up until 2 hours prior to anesthesia or sedation in low-aspiration-risk patients undergoing elective cases.¹¹ The guidelines allowed for other beverages and food based on their standard transit times (Table). The ASA guidelines do not define low-aspiration-risk patients. Anesthesiologists generally exclude from the low-risk category patients who may have delayed gastric emptying from medical or iatrogenic causes. The updated 2017 ASA guidelines remain unchanged regarding fasting guidelines.¹² Studies suggest that approximately 10% to 20% of NPO after midnight orders are avoidable.^1,3 For those instances, procedures are often deemed not necessary or do not require NPO status.¹

In a study evaluating the reasons that necessary procedures are canceled, only 0.5% of inpatient procedures are cancelled due to the inappropriate ingestion of food or drink.³ In addition, NPO status creates risk. Patients with prolonged NPO status report greater hunger, thirst, tiredness, and weakness prior to surgery when compared with patients receiving a carbohydrate-rich drink 2 hours prior to procedures.^9,13,14 In fact, multiple studies have suggested that preoperative carbohydrate-rich drinks 2 hours before surgery can be associated with decreased insulin resistance in the perioperative period, decreased length of stay, and improvement in perioperative metabolic, cardiac, and psychosomatic status.^9,13-15 These types of studies have informed the enhanced recovery after surgery program, which recommends a carbohydrate beverage 2 to 3 hours prior to surgery.

Prescribe the minimum recommended fasting times only for low-aspiration-risk patients undergoing elective procedures. Risk for regurgitation or aspiration increases for patients with conditions resulting in decreased gastric emptying, gastric or bowel obstruction, or lower esophageal sphincter incompetence. Those patients may require longer NPO time periods.⁸ Higher-risk diagnoses and clinical conditions include gastroparesis, trauma, and pregnancy.^5,8,16 Specific risk factors for aspiration in children may include trauma, bowel obstruction, depressed consciousness, shock, or ileus.¹⁰ For surgical emergencies, balance the risk of surgical delay vs perceived aspiration risk.

Use evidence-based guidelines to assess periprocedural aspiration risk. The ASA guidelines suggest that healthy, nonpregnant patients should fast for 8 hours after heavy meals, 6 hours after a light, nonfatty meal, and 2 hours after clear liquids (eg, water, fruit juices without pulp, carbonated beverages, black coffee).¹² Focus on the type of food or drink rather than the volume ingested.¹² Additionally, patients should ingest, with small amounts of clear fluids, appropriate home medications for acute and chronic conditions regardless of NPO status.

While procedure delays or cancellations for any reason upset patients and families and can disrupt the flow of the operating room and procedural suite, we can achieve the delicate balance between efficiency and patient safety and comfort. Since complex inpatient procedural scheduling may not allow for liberalization of solids requiring 6 to 8 hours of fasting time, focus on liberalizing liquids 2 hours prior to anesthesia. This allows staff to minimize the time low-risk patients fast while still maintaining flexibility for operating room case scheduling. We must promote communication between operating room and floor staff to anticipate timing of procedures each day. Healthcare facilities should aim to achieve time-based preprocedural NPO status as opposed to an arbitrary starting time like midnight.⁴

• Risk stratify patients for anesthesia-related aspiration with the aim of identifying those at low aspiration risk.
• For low-risk patients, adhere to recommended fasting times: 2 hours for a clear carbohydrate beverage, 4 hours for breast milk, 6 hours for a light meal or formula, and 8 hours for a fatty meal.
• For patients not deemed low risk, determine the appropriate length of preprocedural fasting by consulting with the anesthesia and surgical teams.

NPO after midnight represents a low-value and arbitrary practice that leaves patients fasting longer than necessary.^2,3,12 In addition to the 2017 ASA guidelines, newer studies and protocols are improving patient satisfaction, minimizing patient dehydration and electrolyte disturbances, and incorporating enhanced recovery after surgery factors into a better patient experience. Returning to the clinical scenario, the hospitalist team can increase patient satisfaction by focusing on liberalizing clear fluids with a carbohydrate beverage up to 2 hours prior to elective surgery while still allowing for schedule flexibility. For this patient, a 3 pm procedure time would have allowed him to have a light breakfast and carbohydrate beverages until 2 hours prior to anesthesia. Dispose of the antiquated practice of NPO after midnight by maximizing clear fluid intake in accordance with current guidelines prior to sedation and general anesthesia. This change in practice will help to achieve normophysiology and increase patient satisfaction.

Do you think this is a low-value practice? Is this truly a “Thing We Do for No Reason™”? Share what you do in your practice and join in the conversation online by retweeting it on Twitter (#TWDFNR) and liking it on Facebook. We invite you to propose ideas for other “Things We Do for No Reason™” topics by emailing [email protected].

Disclaimer: The opinions expressed in this article are those of the authors alone and do not reflect the views of the Department of Veterans Affairs. The Veterans Affairs Quality Scholars Program is supported by the Veterans Affairs Office of Academic Affiliations, Washington, DC.

Inspired by the ABIM Foundation’s Choosing Wisel y ^® campaign, the “Things We Do for No Reason ^™ ” (TWDFNR) series reviews practices that have become common parts of hospital care but may provide little value to our patients. Practices reviewed in the TWDFNR series do not represent clear-cut conclusions or clinical practice standards but are meant as a starting place for research and active discussions among hospitalists and patients. We invite you to be part of that discussion.

The hospitalist admits an 18-year-old man for newly diagnosed granulomatosis with polyangiitis to receive expedited pulse-dose steroids and plasma exchange. After consulting interventional radiology for catheter placement the following day, the hospitalist places a “strict” nil per os (nothing by mouth, NPO) after midnight order. During rounds the following morning, the patient reports that he wants to eat. At 9 am, interventional radiology informs the nurse that the line placement will take place at 3 pm. Due to emergencies and other unplanned delays, the catheter placement occurs at 5 pm. The patient and family express their displeasure about the prolonged fasting and ask why this happened.

Hospitalists commonly order “NPO after midnight” diets in anticipation of procedures requiring sedation or general anesthesia. Typically, NPO refers to no food or drink, but in some instances, NPO includes no oral medications. Up to half of medical patients experience some time of fasting while hospitalized.¹ However, NPO practices vary widely across institutions.^2,3 A study from 2014 notes that, on average, patients fast preprocedure for approximately 13.5 hours for solids and 9.6 hours for liquids.² Prolonged fasting times offer little benefit to patients and may lead to frequent patient dissatisfaction and complaints.

In 1883, Sir Joseph Lister described 19th century NPO practices distinguishing solids from liquids, allowing patients “tea or beef tea” until 2 to 3 hours prior to surgery.⁴ However, in 1946, Mendelson published an influential account of 66 pregnant women who aspirated during delivery under general anesthesia.⁵ Two of the 66 patients, both of whom had eaten a full meal 6 to 8 hours prior to general anesthesia, died. The study not only increased awareness of the risk of aspiration with general anesthesia in pregnancy, but it influenced the care for the nonpregnant population of patients as well. By the 1960s, anesthesia texts recommended “NPO after midnight” for both liquids and solids in all patients, regardless of pregnancy status.⁴ To minimize the risk to patients, we have continued to pass down the practice of NPO after midnight to subsequent generations.

Additionally, medical centers and hospitals feel pressure to provide efficient, patient-centered, high-value care. Given the complexity of procedural scheduling and the penalties associated with delays, keeping patients NPO ensures their availability for the next open procedural slot. NPO after midnight orders aim to prevent potential delays in treatment that occur when inadvertent ingestion of food and drink leads to cancellation of procedures.

Recent studies have led to a more sophisticated understanding of gastric emptying and the risks of aspiration during sedation and intubation. Gastric emptying studies routinely show that transit of clear liquids out of the stomach is virtually complete within two hours of drinking.⁶ Age, body mass index, and alcohol have no effect on gastric emptying time, and almost all patients return to preingestion gastric residual volumes within 2 hours of clear liquid consumption.^6,7 While morbidly obese patients tend to have higher gastric fluid volumes after 9 hours of fasting, their stomachs empty at rates similar to nonobese individuals.⁶ Note that, regardless of fasting times, morbid obesity predisposes patients to a higher overall gastric volume and lower pH of gastric contents, which may increase risk of aspiration.⁸ A Cochrane review found no statistical difference in gastric volumes or stomach pH in patients on a standard fast vs shortened (<180 minutes) liquid fast.⁹ The review included nine studies that found patients who consumed a clear liquid beverage had reduced gastric volumes, compared with patients in a fasting state (P < .001).⁹

In a pediatric retrospective study of pulmonary aspiration events, the researchers demonstrated that clinically significant aspiration (presence of bilious secretions in the tracheobronchial airways) occurred at a rate of 0.04% with emergency surgery.¹⁰ Bowel obstruction or ileus accounted for approximately 54% of those cases. Importantly, the reported aspiration rate approximates the rate of pregnant patients from the 1946 Mendelson study of 0.14% (66 out of 44,016), which originally prompted the use of the prolonged NPO status. Based on the Cochrane review of perioperative fasting recommendations for those older than 18 years, consuming fluids more than 90 minutes preoperatively confers a negligible (0 adverse events reported in 9 studies) risk for aspiration or regurgitation events.⁹

In 1998, as a result of these and other similar studies, the American Society of Anesthesiologists (ASA) along with global anesthesia partners adopted guidelines that allowed clear liquids up until 2 hours prior to anesthesia or sedation in low-aspiration-risk patients undergoing elective cases.¹¹ The guidelines allowed for other beverages and food based on their standard transit times (Table). The ASA guidelines do not define low-aspiration-risk patients. Anesthesiologists generally exclude from the low-risk category patients who may have delayed gastric emptying from medical or iatrogenic causes. The updated 2017 ASA guidelines remain unchanged regarding fasting guidelines.¹² Studies suggest that approximately 10% to 20% of NPO after midnight orders are avoidable.^1,3 For those instances, procedures are often deemed not necessary or do not require NPO status.¹

In a study evaluating the reasons that necessary procedures are canceled, only 0.5% of inpatient procedures are cancelled due to the inappropriate ingestion of food or drink.³ In addition, NPO status creates risk. Patients with prolonged NPO status report greater hunger, thirst, tiredness, and weakness prior to surgery when compared with patients receiving a carbohydrate-rich drink 2 hours prior to procedures.^9,13,14 In fact, multiple studies have suggested that preoperative carbohydrate-rich drinks 2 hours before surgery can be associated with decreased insulin resistance in the perioperative period, decreased length of stay, and improvement in perioperative metabolic, cardiac, and psychosomatic status.^9,13-15 These types of studies have informed the enhanced recovery after surgery program, which recommends a carbohydrate beverage 2 to 3 hours prior to surgery.

Prescribe the minimum recommended fasting times only for low-aspiration-risk patients undergoing elective procedures. Risk for regurgitation or aspiration increases for patients with conditions resulting in decreased gastric emptying, gastric or bowel obstruction, or lower esophageal sphincter incompetence. Those patients may require longer NPO time periods.⁸ Higher-risk diagnoses and clinical conditions include gastroparesis, trauma, and pregnancy.^5,8,16 Specific risk factors for aspiration in children may include trauma, bowel obstruction, depressed consciousness, shock, or ileus.¹⁰ For surgical emergencies, balance the risk of surgical delay vs perceived aspiration risk.

Use evidence-based guidelines to assess periprocedural aspiration risk. The ASA guidelines suggest that healthy, nonpregnant patients should fast for 8 hours after heavy meals, 6 hours after a light, nonfatty meal, and 2 hours after clear liquids (eg, water, fruit juices without pulp, carbonated beverages, black coffee).¹² Focus on the type of food or drink rather than the volume ingested.¹² Additionally, patients should ingest, with small amounts of clear fluids, appropriate home medications for acute and chronic conditions regardless of NPO status.

While procedure delays or cancellations for any reason upset patients and families and can disrupt the flow of the operating room and procedural suite, we can achieve the delicate balance between efficiency and patient safety and comfort. Since complex inpatient procedural scheduling may not allow for liberalization of solids requiring 6 to 8 hours of fasting time, focus on liberalizing liquids 2 hours prior to anesthesia. This allows staff to minimize the time low-risk patients fast while still maintaining flexibility for operating room case scheduling. We must promote communication between operating room and floor staff to anticipate timing of procedures each day. Healthcare facilities should aim to achieve time-based preprocedural NPO status as opposed to an arbitrary starting time like midnight.⁴

• Risk stratify patients for anesthesia-related aspiration with the aim of identifying those at low aspiration risk.
• For low-risk patients, adhere to recommended fasting times: 2 hours for a clear carbohydrate beverage, 4 hours for breast milk, 6 hours for a light meal or formula, and 8 hours for a fatty meal.
• For patients not deemed low risk, determine the appropriate length of preprocedural fasting by consulting with the anesthesia and surgical teams.

NPO after midnight represents a low-value and arbitrary practice that leaves patients fasting longer than necessary.^2,3,12 In addition to the 2017 ASA guidelines, newer studies and protocols are improving patient satisfaction, minimizing patient dehydration and electrolyte disturbances, and incorporating enhanced recovery after surgery factors into a better patient experience. Returning to the clinical scenario, the hospitalist team can increase patient satisfaction by focusing on liberalizing clear fluids with a carbohydrate beverage up to 2 hours prior to elective surgery while still allowing for schedule flexibility. For this patient, a 3 pm procedure time would have allowed him to have a light breakfast and carbohydrate beverages until 2 hours prior to anesthesia. Dispose of the antiquated practice of NPO after midnight by maximizing clear fluid intake in accordance with current guidelines prior to sedation and general anesthesia. This change in practice will help to achieve normophysiology and increase patient satisfaction.

Do you think this is a low-value practice? Is this truly a “Thing We Do for No Reason™”? Share what you do in your practice and join in the conversation online by retweeting it on Twitter (#TWDFNR) and liking it on Facebook. We invite you to propose ideas for other “Things We Do for No Reason™” topics by emailing [email protected].

Disclaimer: The opinions expressed in this article are those of the authors alone and do not reflect the views of the Department of Veterans Affairs. The Veterans Affairs Quality Scholars Program is supported by the Veterans Affairs Office of Academic Affiliations, Washington, DC.

Inspired by the ABIM Foundation’s Choosing Wisel y ^® campaign, the “Things We Do for No Reason ^™ ” (TWDFNR) series reviews practices that have become common parts of hospital care but may provide little value to our patients. Practices reviewed in the TWDFNR series do not represent clear-cut conclusions or clinical practice standards but are meant as a starting place for research and active discussions among hospitalists and patients. We invite you to be part of that discussion.

The hospitalist admits an 18-year-old man for newly diagnosed granulomatosis with polyangiitis to receive expedited pulse-dose steroids and plasma exchange. After consulting interventional radiology for catheter placement the following day, the hospitalist places a “strict” nil per os (nothing by mouth, NPO) after midnight order. During rounds the following morning, the patient reports that he wants to eat. At 9 am, interventional radiology informs the nurse that the line placement will take place at 3 pm. Due to emergencies and other unplanned delays, the catheter placement occurs at 5 pm. The patient and family express their displeasure about the prolonged fasting and ask why this happened.

Hospitalists commonly order “NPO after midnight” diets in anticipation of procedures requiring sedation or general anesthesia. Typically, NPO refers to no food or drink, but in some instances, NPO includes no oral medications. Up to half of medical patients experience some time of fasting while hospitalized.¹ However, NPO practices vary widely across institutions.^2,3 A study from 2014 notes that, on average, patients fast preprocedure for approximately 13.5 hours for solids and 9.6 hours for liquids.² Prolonged fasting times offer little benefit to patients and may lead to frequent patient dissatisfaction and complaints.

In 1883, Sir Joseph Lister described 19th century NPO practices distinguishing solids from liquids, allowing patients “tea or beef tea” until 2 to 3 hours prior to surgery.⁴ However, in 1946, Mendelson published an influential account of 66 pregnant women who aspirated during delivery under general anesthesia.⁵ Two of the 66 patients, both of whom had eaten a full meal 6 to 8 hours prior to general anesthesia, died. The study not only increased awareness of the risk of aspiration with general anesthesia in pregnancy, but it influenced the care for the nonpregnant population of patients as well. By the 1960s, anesthesia texts recommended “NPO after midnight” for both liquids and solids in all patients, regardless of pregnancy status.⁴ To minimize the risk to patients, we have continued to pass down the practice of NPO after midnight to subsequent generations.

Additionally, medical centers and hospitals feel pressure to provide efficient, patient-centered, high-value care. Given the complexity of procedural scheduling and the penalties associated with delays, keeping patients NPO ensures their availability for the next open procedural slot. NPO after midnight orders aim to prevent potential delays in treatment that occur when inadvertent ingestion of food and drink leads to cancellation of procedures.

Recent studies have led to a more sophisticated understanding of gastric emptying and the risks of aspiration during sedation and intubation. Gastric emptying studies routinely show that transit of clear liquids out of the stomach is virtually complete within two hours of drinking.⁶ Age, body mass index, and alcohol have no effect on gastric emptying time, and almost all patients return to preingestion gastric residual volumes within 2 hours of clear liquid consumption.^6,7 While morbidly obese patients tend to have higher gastric fluid volumes after 9 hours of fasting, their stomachs empty at rates similar to nonobese individuals.⁶ Note that, regardless of fasting times, morbid obesity predisposes patients to a higher overall gastric volume and lower pH of gastric contents, which may increase risk of aspiration.⁸ A Cochrane review found no statistical difference in gastric volumes or stomach pH in patients on a standard fast vs shortened (<180 minutes) liquid fast.⁹ The review included nine studies that found patients who consumed a clear liquid beverage had reduced gastric volumes, compared with patients in a fasting state (P < .001).⁹

In a pediatric retrospective study of pulmonary aspiration events, the researchers demonstrated that clinically significant aspiration (presence of bilious secretions in the tracheobronchial airways) occurred at a rate of 0.04% with emergency surgery.¹⁰ Bowel obstruction or ileus accounted for approximately 54% of those cases. Importantly, the reported aspiration rate approximates the rate of pregnant patients from the 1946 Mendelson study of 0.14% (66 out of 44,016), which originally prompted the use of the prolonged NPO status. Based on the Cochrane review of perioperative fasting recommendations for those older than 18 years, consuming fluids more than 90 minutes preoperatively confers a negligible (0 adverse events reported in 9 studies) risk for aspiration or regurgitation events.⁹

In 1998, as a result of these and other similar studies, the American Society of Anesthesiologists (ASA) along with global anesthesia partners adopted guidelines that allowed clear liquids up until 2 hours prior to anesthesia or sedation in low-aspiration-risk patients undergoing elective cases.¹¹ The guidelines allowed for other beverages and food based on their standard transit times (Table). The ASA guidelines do not define low-aspiration-risk patients. Anesthesiologists generally exclude from the low-risk category patients who may have delayed gastric emptying from medical or iatrogenic causes. The updated 2017 ASA guidelines remain unchanged regarding fasting guidelines.¹² Studies suggest that approximately 10% to 20% of NPO after midnight orders are avoidable.^1,3 For those instances, procedures are often deemed not necessary or do not require NPO status.¹

In a study evaluating the reasons that necessary procedures are canceled, only 0.5% of inpatient procedures are cancelled due to the inappropriate ingestion of food or drink.³ In addition, NPO status creates risk. Patients with prolonged NPO status report greater hunger, thirst, tiredness, and weakness prior to surgery when compared with patients receiving a carbohydrate-rich drink 2 hours prior to procedures.^9,13,14 In fact, multiple studies have suggested that preoperative carbohydrate-rich drinks 2 hours before surgery can be associated with decreased insulin resistance in the perioperative period, decreased length of stay, and improvement in perioperative metabolic, cardiac, and psychosomatic status.^9,13-15 These types of studies have informed the enhanced recovery after surgery program, which recommends a carbohydrate beverage 2 to 3 hours prior to surgery.

Prescribe the minimum recommended fasting times only for low-aspiration-risk patients undergoing elective procedures. Risk for regurgitation or aspiration increases for patients with conditions resulting in decreased gastric emptying, gastric or bowel obstruction, or lower esophageal sphincter incompetence. Those patients may require longer NPO time periods.⁸ Higher-risk diagnoses and clinical conditions include gastroparesis, trauma, and pregnancy.^5,8,16 Specific risk factors for aspiration in children may include trauma, bowel obstruction, depressed consciousness, shock, or ileus.¹⁰ For surgical emergencies, balance the risk of surgical delay vs perceived aspiration risk.

Use evidence-based guidelines to assess periprocedural aspiration risk. The ASA guidelines suggest that healthy, nonpregnant patients should fast for 8 hours after heavy meals, 6 hours after a light, nonfatty meal, and 2 hours after clear liquids (eg, water, fruit juices without pulp, carbonated beverages, black coffee).¹² Focus on the type of food or drink rather than the volume ingested.¹² Additionally, patients should ingest, with small amounts of clear fluids, appropriate home medications for acute and chronic conditions regardless of NPO status.

While procedure delays or cancellations for any reason upset patients and families and can disrupt the flow of the operating room and procedural suite, we can achieve the delicate balance between efficiency and patient safety and comfort. Since complex inpatient procedural scheduling may not allow for liberalization of solids requiring 6 to 8 hours of fasting time, focus on liberalizing liquids 2 hours prior to anesthesia. This allows staff to minimize the time low-risk patients fast while still maintaining flexibility for operating room case scheduling. We must promote communication between operating room and floor staff to anticipate timing of procedures each day. Healthcare facilities should aim to achieve time-based preprocedural NPO status as opposed to an arbitrary starting time like midnight.⁴

• Risk stratify patients for anesthesia-related aspiration with the aim of identifying those at low aspiration risk.
• For low-risk patients, adhere to recommended fasting times: 2 hours for a clear carbohydrate beverage, 4 hours for breast milk, 6 hours for a light meal or formula, and 8 hours for a fatty meal.
• For patients not deemed low risk, determine the appropriate length of preprocedural fasting by consulting with the anesthesia and surgical teams.

NPO after midnight represents a low-value and arbitrary practice that leaves patients fasting longer than necessary.^2,3,12 In addition to the 2017 ASA guidelines, newer studies and protocols are improving patient satisfaction, minimizing patient dehydration and electrolyte disturbances, and incorporating enhanced recovery after surgery factors into a better patient experience. Returning to the clinical scenario, the hospitalist team can increase patient satisfaction by focusing on liberalizing clear fluids with a carbohydrate beverage up to 2 hours prior to elective surgery while still allowing for schedule flexibility. For this patient, a 3 pm procedure time would have allowed him to have a light breakfast and carbohydrate beverages until 2 hours prior to anesthesia. Dispose of the antiquated practice of NPO after midnight by maximizing clear fluid intake in accordance with current guidelines prior to sedation and general anesthesia. This change in practice will help to achieve normophysiology and increase patient satisfaction.

Do you think this is a low-value practice? Is this truly a “Thing We Do for No Reason™”? Share what you do in your practice and join in the conversation online by retweeting it on Twitter (#TWDFNR) and liking it on Facebook. We invite you to propose ideas for other “Things We Do for No Reason™” topics by emailing [email protected].

Disclaimer: The opinions expressed in this article are those of the authors alone and do not reflect the views of the Department of Veterans Affairs. The Veterans Affairs Quality Scholars Program is supported by the Veterans Affairs Office of Academic Affiliations, Washington, DC.

This icon represents the patient’s case. Each paragraph that follows represents the discussant’s thoughts.

An 81-year-old woman with a remote history of left proximal femoral fracture (status post–open reduction and internal fixation) acutely developed severe pain in her left lateral thigh while at her home. A few days prior to her left thigh pain, the patient had routine blood work done. Her lab results (prior to the onset of her symptoms) revealed that her hemoglobin decreased from 10 g/dL, noted 9 months earlier, to 6.6 g/dL. Her primary care physician, who was planning to see the patient for her next regularly scheduled follow-up, was made aware of the patient’s decline in hemoglobin prior to the planned visit. The primary care physician called the patient to inform her about her concerning lab findings and coincidentally became aware of the acute, new-onset left thigh pain. The primary care physician requested that the patient be taken by her daughter to the emergency department (ED) for further evaluation.

The acute decrease in hemoglobin carries a broad differential and may or may not be related to the subsequent development of thigh pain. The presentation of an acute onset of pain in the thigh within the context of this patient’s age and gender suggests a femur fracture; this can be osteoporosis-related or a pathologic fracture associated with malignancy. Several malignancies are plausible, including multiple myeloma (given the anemia) or breast cancer. The proximal part of long bones is the most common site of pathologic fractures, and the femur accounts for half of these cases. Plain radiographs would be appropriate initial imaging and may be followed by either a computed tomography (CT) scan or magnetic resonance imaging (MRI).

In the ED, she denied any recent trauma, hemoptysis, recent dark or bloody stools, vaginal bleeding, abdominal pain, or history of gastric ulcers. She had not experienced any similar episodes of thigh pain in the past. She had a history of atrial fibrillation, hypertension, diabetes mellitus type 2 with diabetic retinopathy and peripheral neuropathy, osteoporosis, nonalcoholic fatty liver disease (NAFLD), and internal hemorrhoids. Her medications included apixaban, metoprolol succinate, metformin, losartan, sitagliptin, calcium, vitamin D, alendronate, and fish oil. She had mild tenderness to palpation of her thigh, but her exam was otherwise normal. Radiography of the left hip and pelvis showed no acute fracture (Figure 1). An upper and lower endoscopy 3 years prior to her presentation revealed internal hemorrhoids.

The patient is taking apixaban, a direct factor Xa inhibitor. The absence of other obvious sources of bleeding suggests that the cause of anemia and pain is most likely bleeding into the anterior thigh compartment, exacerbated by the underlying anticoagulation. Since there was no trauma preceding this episode, the differential diagnosis must be expanded to include other, less common sources of bleeding, including a vascular anomaly such as a pseudoaneurysm or arteriovenous malformation. While the radiographs were normal, a CT scan or MRI may allow for identification of a fracture, other bone lesion, and/or hematoma.

A complete blood count revealed a hemoglobin of 6.6 g/dL (normal, 11.5-14.1 g/dL) with a mean corpuscular volume of 62 fL (normal, 79-96 fL). A CT scan of the abdomen and pelvis with intravenous contrast (Figure 2) was obtained to evaluate for intra-abdominal hemorrhage and retroperitoneal hematoma; it showed mild abdominal and pelvic ascites, a small right pleural effusion with compressive atelectasis, and generalized anasarca, but no evidence of bleeding. She was administered 2 units of packed red blood cells. Apixaban was held and 40 mg intravenous pantoprazole twice daily was started. Her iron level was 12 µg/dL (normal, 50-170 µg/dL); total iron-binding capacity (TIBC) was 431 µg/dL (normal, 179-378 µg/dL); and ferritin level was 19 ng/mL (normal, 10-204 ng/mL). Her basic metabolic panel, liver enzymes, international normalized ratio, partial thromboplastin time, and folate were normal. Serum vitamin B₁₂ level was 277 pg/mL (normal, 213-816 pg/mL), and the reticulocyte count was 1.7%.

Myopathies can be hereditary or acquired. Hereditary myopathies include congenital myopathies, muscular dystrophies, channelopathies, primary metabolic myopathies, and mitochondrial myopathies. Acquired myopathies include infectious myopathies, inflammatory myopathies, endocrine myopathies, secondary metabolic myopathies, and drug-induced and toxic myopathies. The findings of hemorrhagic myositis and skin edema are very intriguing, especially given their localized features. An overt femur fracture was previously ruled out, and an anterior thigh compartment syndrome was considered less likely after orthopedic surgery consultation. There is no description of the patient taking medications that could cause myopathy (such as statins), and there are also no clinical features suggestive of primary inflammatory myopathy, such as dermatomyositis. Increased suspicion of a focal inflammatory process such as localized scleroderma with regional inflammatory myopathy or another focal myopathy must be considered. The next diagnostic steps would include measuring the creatine kinase level, as well as obtaining an MRI of the leg to assess the nature and extent of the myopathy.

Multidisciplinary involvement, including hematology, rheumatology, and surgery, aided in narrowing the differential diagnosis. On hospital day 10, an MRI of the left thigh was performed for suspicion of diabetic myonecrosis (Figure 4). The MRI revealed a 10 cm × 3.6 cm × 22 cm intramuscular hematoma in the belly of the vastus lateralis muscle with associated soft tissue swelling, overlying subcutaneous edema, and skin thickening that was suggestive of hemorrhagic diabetic myonecrosis with some atypical features. A rheumatology consult was requested to evaluate for possible vasculitis in the left lower extremity, and vasculitis was not considered likely. The diagnosis of diabetic myonecrosis with associated intramuscular hemorrhage secondary to apixaban was made after careful reconsideration of the clinical presentation, imaging and laboratory data, and overall picture. Based on the clinical findings, imaging results, and exclusion of alternative causative pathologies of thigh swelling, no biopsy was performed, as it was not considered necessary to make the diagnosis of diabetic myonecrosis. The patient was discharged on hospital day 11 and was doing well. She followed up with her primary care doctor and has regained normal function of her leg.

Diabetic myonecrosis, or diabetic muscle infarction, is an uncommon nontraumatic myopathy that occurs in patients with diabetes who develop acute, focal muscle pain without recent trauma. In this case, the muscle infarction was further complicated by hemorrhagic transformation. Diabetic myonecrosis is relatively uncommon and a diagnosis made by combining history, examination, and laboratory findings and excluding other alternative conditions.

A clear schema for approaching the patient with acute, nontraumatic myopathies is important in avoiding diagnostic error. One effective schema is to divide myopathy into infectious and noninfectious categories. Causes of infectious myopathy include bacterial infections (eg, pyomyositis), inflammatory damage to muscles associated with viruses (eg, influenza), as well as rarer causes. Bacterial processes tend to be relatively focal and affect a specific muscle group or anatomic compartment, while viral causes are often more diffuse and occur in the context of a systemic viral syndrome. Bacterial causes range in severity, and life-threatening conditions, such as necrotizing soft tissue infection, must be considered. In this case, bacterial causes were less likely given the patient’s lack of fever, leukocytosis, and systemic signs of infection.^1,2 However, these findings are not uniformly sensitive, and clinicians should not exclude potentially life- or limb-threatening infections without thorough evaluation. For example, pyomyositis may present without fever in the subacute stage, without leukocytosis if the patient is immunocompromised, and without overt pus if the infection is not in the suppurative stage.³ Viral causes were made less likely in this patient given the lack of a current or recent systemic viral syndrome.

Once infectious etiologies are deemed unlikely, noninfectious etiologies for nontraumatic myopathies should be considered. Some causes of noninfectious myopathy present with the muscle symptoms as a predominant feature, while others present in the context of another illness such as cancer, metabolic disorders, or other systemic disorders. Many noninfectious causes of myopathy associated with systemic illnesses have diffuse or relatively diffuse symptoms, with pain and/or weakness in multiple muscle groups, often in a bilateral distribution. Such examples include dermatomyositis and polymyositis as well as myositis associated with other rheumatologic conditions. Nontraumatic rhabdomyolysis is diffuse and can occur in association with medications and/or genetic conditions.

Angervall and Stener⁴ first described diabetic myonecrosis in 1965 as tumoriform focal muscular degeneration due to diabetic microangiopathy. The most commonly affected muscle groups in diabetic myonecrosis are the anterior thigh, calf, and posterior thigh, followed by muscles in the upper extremities.⁵ Patients with diabetic myonecrosis have an overall mean age at presentation of 44.6 years; affected patients with type 1 diabetes mellitus present at a mean age nearly 20 years younger than those with type 2 diabetes mellitus (35.9 years vs 52.2 years, respectively).⁶ Patients tend to have a long (often >15 years) history of diabetes with microvascular complications such as retinopathy (reported in 71%), nephropathy (reported in 57%), and/or neuropathy (reported in 55%).⁷

The mainstay of the diagnosis of diabetic myonecrosis is a thorough history and physical examination and imaging. Routine laboratory evaluation is relatively unhelpful in diagnosing diabetic myonecrosis, but appropriate imaging can provide valuable supportive information. A CT scan and MRI are both helpful in excluding other etiologies as well as identifying features consistent with diabetic myonecrosis. A CT scan can help exclude a localized abscess, tumor, or bone destruction and, in affected patients, may show increased subcutaneous attenuation and increased muscle size with decreased attenuation secondary to edema.² However, a CT scan may not give optimal assessment of muscle tissue, and therefore MRI may need to be considered. MRI T2 images have a sensitivity nearing 90% for detecting myonecrosis.¹ The diagnostic value of MRI often obviates the need for muscle biopsy.

Spontaneous infarction with hemorrhagic features seen on imaging can be explained by a combination of damage from atherosclerotic or microvascular disease, an activated coagulation cascade, and an impaired fibrinolytic pathway.⁸ Hemorrhagic conversion in diabetic myonecrosis appears to be uncommon.⁹ In our case, we suspect that it developed because of the combination of bleeding risk from apixaban and the underlying mechanisms of diabetic myonecrosis.

The treatment of diabetic myonecrosis is mainly supportive, with an emphasis on rest, nonsteroidal anti-inflammatory agents, antiplatelet agents, and strict glycemic control.¹⁰ There is conflicting information about the value of limb immobilization versus active physical therapy as appropriate treatment modalities.¹¹ Patients who present with clinical concern for sepsis or compartment syndrome require consultation for consideration of acute surgical intervention.¹⁰ The short-term prognosis is promising with supportive therapy, but the condition may recur.¹² The recurrence rate may be as high as 40%, with a 2-year mortality of 10%.¹³ Ultimately, patients need to be followed closely in the outpatient setting to reduce the risk of recurrence.

In this patient, the simultaneous occurrence of focal pain and acute blood loss anemia led to a diagnosis of diabetic myonecrosis that was complicated by hemorrhagic conversion, a truly painful coincidence. The patient underwent a thorough evaluation for acute blood loss before the diagnosis was ultimately made. Clinicians should consider diabetic myonecrosis in patients with diabetes who present with acute muscle pain but no evidence of infection.

• Diabetic myonecrosis is an underrecognized entity and should be included in the differential diagnosis for patients with diabetes who present with acute muscle pain and no history of trauma.
• Imaging with CT and/or MRI of the affected region is the mainstay of diagnosis; treatment is predicated on severity and risk factors and can range from conservative therapy to operative intervention.
• Although the prognosis is good in these patients, careful outpatient follow-up is necessary to oversee their recovery to help reduce the risk of recurrence.

The authors thank Dr Vijay Singh for his radiology input on image selection for this manuscript.

This icon represents the patient’s case. Each paragraph that follows represents the discussant’s thoughts.

An 81-year-old woman with a remote history of left proximal femoral fracture (status post–open reduction and internal fixation) acutely developed severe pain in her left lateral thigh while at her home. A few days prior to her left thigh pain, the patient had routine blood work done. Her lab results (prior to the onset of her symptoms) revealed that her hemoglobin decreased from 10 g/dL, noted 9 months earlier, to 6.6 g/dL. Her primary care physician, who was planning to see the patient for her next regularly scheduled follow-up, was made aware of the patient’s decline in hemoglobin prior to the planned visit. The primary care physician called the patient to inform her about her concerning lab findings and coincidentally became aware of the acute, new-onset left thigh pain. The primary care physician requested that the patient be taken by her daughter to the emergency department (ED) for further evaluation.

The acute decrease in hemoglobin carries a broad differential and may or may not be related to the subsequent development of thigh pain. The presentation of an acute onset of pain in the thigh within the context of this patient’s age and gender suggests a femur fracture; this can be osteoporosis-related or a pathologic fracture associated with malignancy. Several malignancies are plausible, including multiple myeloma (given the anemia) or breast cancer. The proximal part of long bones is the most common site of pathologic fractures, and the femur accounts for half of these cases. Plain radiographs would be appropriate initial imaging and may be followed by either a computed tomography (CT) scan or magnetic resonance imaging (MRI).

In the ED, she denied any recent trauma, hemoptysis, recent dark or bloody stools, vaginal bleeding, abdominal pain, or history of gastric ulcers. She had not experienced any similar episodes of thigh pain in the past. She had a history of atrial fibrillation, hypertension, diabetes mellitus type 2 with diabetic retinopathy and peripheral neuropathy, osteoporosis, nonalcoholic fatty liver disease (NAFLD), and internal hemorrhoids. Her medications included apixaban, metoprolol succinate, metformin, losartan, sitagliptin, calcium, vitamin D, alendronate, and fish oil. She had mild tenderness to palpation of her thigh, but her exam was otherwise normal. Radiography of the left hip and pelvis showed no acute fracture (Figure 1). An upper and lower endoscopy 3 years prior to her presentation revealed internal hemorrhoids.

The patient is taking apixaban, a direct factor Xa inhibitor. The absence of other obvious sources of bleeding suggests that the cause of anemia and pain is most likely bleeding into the anterior thigh compartment, exacerbated by the underlying anticoagulation. Since there was no trauma preceding this episode, the differential diagnosis must be expanded to include other, less common sources of bleeding, including a vascular anomaly such as a pseudoaneurysm or arteriovenous malformation. While the radiographs were normal, a CT scan or MRI may allow for identification of a fracture, other bone lesion, and/or hematoma.

A complete blood count revealed a hemoglobin of 6.6 g/dL (normal, 11.5-14.1 g/dL) with a mean corpuscular volume of 62 fL (normal, 79-96 fL). A CT scan of the abdomen and pelvis with intravenous contrast (Figure 2) was obtained to evaluate for intra-abdominal hemorrhage and retroperitoneal hematoma; it showed mild abdominal and pelvic ascites, a small right pleural effusion with compressive atelectasis, and generalized anasarca, but no evidence of bleeding. She was administered 2 units of packed red blood cells. Apixaban was held and 40 mg intravenous pantoprazole twice daily was started. Her iron level was 12 µg/dL (normal, 50-170 µg/dL); total iron-binding capacity (TIBC) was 431 µg/dL (normal, 179-378 µg/dL); and ferritin level was 19 ng/mL (normal, 10-204 ng/mL). Her basic metabolic panel, liver enzymes, international normalized ratio, partial thromboplastin time, and folate were normal. Serum vitamin B₁₂ level was 277 pg/mL (normal, 213-816 pg/mL), and the reticulocyte count was 1.7%.

Myopathies can be hereditary or acquired. Hereditary myopathies include congenital myopathies, muscular dystrophies, channelopathies, primary metabolic myopathies, and mitochondrial myopathies. Acquired myopathies include infectious myopathies, inflammatory myopathies, endocrine myopathies, secondary metabolic myopathies, and drug-induced and toxic myopathies. The findings of hemorrhagic myositis and skin edema are very intriguing, especially given their localized features. An overt femur fracture was previously ruled out, and an anterior thigh compartment syndrome was considered less likely after orthopedic surgery consultation. There is no description of the patient taking medications that could cause myopathy (such as statins), and there are also no clinical features suggestive of primary inflammatory myopathy, such as dermatomyositis. Increased suspicion of a focal inflammatory process such as localized scleroderma with regional inflammatory myopathy or another focal myopathy must be considered. The next diagnostic steps would include measuring the creatine kinase level, as well as obtaining an MRI of the leg to assess the nature and extent of the myopathy.

Multidisciplinary involvement, including hematology, rheumatology, and surgery, aided in narrowing the differential diagnosis. On hospital day 10, an MRI of the left thigh was performed for suspicion of diabetic myonecrosis (Figure 4). The MRI revealed a 10 cm × 3.6 cm × 22 cm intramuscular hematoma in the belly of the vastus lateralis muscle with associated soft tissue swelling, overlying subcutaneous edema, and skin thickening that was suggestive of hemorrhagic diabetic myonecrosis with some atypical features. A rheumatology consult was requested to evaluate for possible vasculitis in the left lower extremity, and vasculitis was not considered likely. The diagnosis of diabetic myonecrosis with associated intramuscular hemorrhage secondary to apixaban was made after careful reconsideration of the clinical presentation, imaging and laboratory data, and overall picture. Based on the clinical findings, imaging results, and exclusion of alternative causative pathologies of thigh swelling, no biopsy was performed, as it was not considered necessary to make the diagnosis of diabetic myonecrosis. The patient was discharged on hospital day 11 and was doing well. She followed up with her primary care doctor and has regained normal function of her leg.

Diabetic myonecrosis, or diabetic muscle infarction, is an uncommon nontraumatic myopathy that occurs in patients with diabetes who develop acute, focal muscle pain without recent trauma. In this case, the muscle infarction was further complicated by hemorrhagic transformation. Diabetic myonecrosis is relatively uncommon and a diagnosis made by combining history, examination, and laboratory findings and excluding other alternative conditions.

A clear schema for approaching the patient with acute, nontraumatic myopathies is important in avoiding diagnostic error. One effective schema is to divide myopathy into infectious and noninfectious categories. Causes of infectious myopathy include bacterial infections (eg, pyomyositis), inflammatory damage to muscles associated with viruses (eg, influenza), as well as rarer causes. Bacterial processes tend to be relatively focal and affect a specific muscle group or anatomic compartment, while viral causes are often more diffuse and occur in the context of a systemic viral syndrome. Bacterial causes range in severity, and life-threatening conditions, such as necrotizing soft tissue infection, must be considered. In this case, bacterial causes were less likely given the patient’s lack of fever, leukocytosis, and systemic signs of infection.^1,2 However, these findings are not uniformly sensitive, and clinicians should not exclude potentially life- or limb-threatening infections without thorough evaluation. For example, pyomyositis may present without fever in the subacute stage, without leukocytosis if the patient is immunocompromised, and without overt pus if the infection is not in the suppurative stage.³ Viral causes were made less likely in this patient given the lack of a current or recent systemic viral syndrome.

Once infectious etiologies are deemed unlikely, noninfectious etiologies for nontraumatic myopathies should be considered. Some causes of noninfectious myopathy present with the muscle symptoms as a predominant feature, while others present in the context of another illness such as cancer, metabolic disorders, or other systemic disorders. Many noninfectious causes of myopathy associated with systemic illnesses have diffuse or relatively diffuse symptoms, with pain and/or weakness in multiple muscle groups, often in a bilateral distribution. Such examples include dermatomyositis and polymyositis as well as myositis associated with other rheumatologic conditions. Nontraumatic rhabdomyolysis is diffuse and can occur in association with medications and/or genetic conditions.

Angervall and Stener⁴ first described diabetic myonecrosis in 1965 as tumoriform focal muscular degeneration due to diabetic microangiopathy. The most commonly affected muscle groups in diabetic myonecrosis are the anterior thigh, calf, and posterior thigh, followed by muscles in the upper extremities.⁵ Patients with diabetic myonecrosis have an overall mean age at presentation of 44.6 years; affected patients with type 1 diabetes mellitus present at a mean age nearly 20 years younger than those with type 2 diabetes mellitus (35.9 years vs 52.2 years, respectively).⁶ Patients tend to have a long (often >15 years) history of diabetes with microvascular complications such as retinopathy (reported in 71%), nephropathy (reported in 57%), and/or neuropathy (reported in 55%).⁷

The mainstay of the diagnosis of diabetic myonecrosis is a thorough history and physical examination and imaging. Routine laboratory evaluation is relatively unhelpful in diagnosing diabetic myonecrosis, but appropriate imaging can provide valuable supportive information. A CT scan and MRI are both helpful in excluding other etiologies as well as identifying features consistent with diabetic myonecrosis. A CT scan can help exclude a localized abscess, tumor, or bone destruction and, in affected patients, may show increased subcutaneous attenuation and increased muscle size with decreased attenuation secondary to edema.² However, a CT scan may not give optimal assessment of muscle tissue, and therefore MRI may need to be considered. MRI T2 images have a sensitivity nearing 90% for detecting myonecrosis.¹ The diagnostic value of MRI often obviates the need for muscle biopsy.

Spontaneous infarction with hemorrhagic features seen on imaging can be explained by a combination of damage from atherosclerotic or microvascular disease, an activated coagulation cascade, and an impaired fibrinolytic pathway.⁸ Hemorrhagic conversion in diabetic myonecrosis appears to be uncommon.⁹ In our case, we suspect that it developed because of the combination of bleeding risk from apixaban and the underlying mechanisms of diabetic myonecrosis.

The treatment of diabetic myonecrosis is mainly supportive, with an emphasis on rest, nonsteroidal anti-inflammatory agents, antiplatelet agents, and strict glycemic control.¹⁰ There is conflicting information about the value of limb immobilization versus active physical therapy as appropriate treatment modalities.¹¹ Patients who present with clinical concern for sepsis or compartment syndrome require consultation for consideration of acute surgical intervention.¹⁰ The short-term prognosis is promising with supportive therapy, but the condition may recur.¹² The recurrence rate may be as high as 40%, with a 2-year mortality of 10%.¹³ Ultimately, patients need to be followed closely in the outpatient setting to reduce the risk of recurrence.

In this patient, the simultaneous occurrence of focal pain and acute blood loss anemia led to a diagnosis of diabetic myonecrosis that was complicated by hemorrhagic conversion, a truly painful coincidence. The patient underwent a thorough evaluation for acute blood loss before the diagnosis was ultimately made. Clinicians should consider diabetic myonecrosis in patients with diabetes who present with acute muscle pain but no evidence of infection.

• Diabetic myonecrosis is an underrecognized entity and should be included in the differential diagnosis for patients with diabetes who present with acute muscle pain and no history of trauma.
• Imaging with CT and/or MRI of the affected region is the mainstay of diagnosis; treatment is predicated on severity and risk factors and can range from conservative therapy to operative intervention.
• Although the prognosis is good in these patients, careful outpatient follow-up is necessary to oversee their recovery to help reduce the risk of recurrence.

The authors thank Dr Vijay Singh for his radiology input on image selection for this manuscript.

This icon represents the patient’s case. Each paragraph that follows represents the discussant’s thoughts.

An 81-year-old woman with a remote history of left proximal femoral fracture (status post–open reduction and internal fixation) acutely developed severe pain in her left lateral thigh while at her home. A few days prior to her left thigh pain, the patient had routine blood work done. Her lab results (prior to the onset of her symptoms) revealed that her hemoglobin decreased from 10 g/dL, noted 9 months earlier, to 6.6 g/dL. Her primary care physician, who was planning to see the patient for her next regularly scheduled follow-up, was made aware of the patient’s decline in hemoglobin prior to the planned visit. The primary care physician called the patient to inform her about her concerning lab findings and coincidentally became aware of the acute, new-onset left thigh pain. The primary care physician requested that the patient be taken by her daughter to the emergency department (ED) for further evaluation.

The acute decrease in hemoglobin carries a broad differential and may or may not be related to the subsequent development of thigh pain. The presentation of an acute onset of pain in the thigh within the context of this patient’s age and gender suggests a femur fracture; this can be osteoporosis-related or a pathologic fracture associated with malignancy. Several malignancies are plausible, including multiple myeloma (given the anemia) or breast cancer. The proximal part of long bones is the most common site of pathologic fractures, and the femur accounts for half of these cases. Plain radiographs would be appropriate initial imaging and may be followed by either a computed tomography (CT) scan or magnetic resonance imaging (MRI).

In the ED, she denied any recent trauma, hemoptysis, recent dark or bloody stools, vaginal bleeding, abdominal pain, or history of gastric ulcers. She had not experienced any similar episodes of thigh pain in the past. She had a history of atrial fibrillation, hypertension, diabetes mellitus type 2 with diabetic retinopathy and peripheral neuropathy, osteoporosis, nonalcoholic fatty liver disease (NAFLD), and internal hemorrhoids. Her medications included apixaban, metoprolol succinate, metformin, losartan, sitagliptin, calcium, vitamin D, alendronate, and fish oil. She had mild tenderness to palpation of her thigh, but her exam was otherwise normal. Radiography of the left hip and pelvis showed no acute fracture (Figure 1). An upper and lower endoscopy 3 years prior to her presentation revealed internal hemorrhoids.

The patient is taking apixaban, a direct factor Xa inhibitor. The absence of other obvious sources of bleeding suggests that the cause of anemia and pain is most likely bleeding into the anterior thigh compartment, exacerbated by the underlying anticoagulation. Since there was no trauma preceding this episode, the differential diagnosis must be expanded to include other, less common sources of bleeding, including a vascular anomaly such as a pseudoaneurysm or arteriovenous malformation. While the radiographs were normal, a CT scan or MRI may allow for identification of a fracture, other bone lesion, and/or hematoma.

A complete blood count revealed a hemoglobin of 6.6 g/dL (normal, 11.5-14.1 g/dL) with a mean corpuscular volume of 62 fL (normal, 79-96 fL). A CT scan of the abdomen and pelvis with intravenous contrast (Figure 2) was obtained to evaluate for intra-abdominal hemorrhage and retroperitoneal hematoma; it showed mild abdominal and pelvic ascites, a small right pleural effusion with compressive atelectasis, and generalized anasarca, but no evidence of bleeding. She was administered 2 units of packed red blood cells. Apixaban was held and 40 mg intravenous pantoprazole twice daily was started. Her iron level was 12 µg/dL (normal, 50-170 µg/dL); total iron-binding capacity (TIBC) was 431 µg/dL (normal, 179-378 µg/dL); and ferritin level was 19 ng/mL (normal, 10-204 ng/mL). Her basic metabolic panel, liver enzymes, international normalized ratio, partial thromboplastin time, and folate were normal. Serum vitamin B₁₂ level was 277 pg/mL (normal, 213-816 pg/mL), and the reticulocyte count was 1.7%.

Myopathies can be hereditary or acquired. Hereditary myopathies include congenital myopathies, muscular dystrophies, channelopathies, primary metabolic myopathies, and mitochondrial myopathies. Acquired myopathies include infectious myopathies, inflammatory myopathies, endocrine myopathies, secondary metabolic myopathies, and drug-induced and toxic myopathies. The findings of hemorrhagic myositis and skin edema are very intriguing, especially given their localized features. An overt femur fracture was previously ruled out, and an anterior thigh compartment syndrome was considered less likely after orthopedic surgery consultation. There is no description of the patient taking medications that could cause myopathy (such as statins), and there are also no clinical features suggestive of primary inflammatory myopathy, such as dermatomyositis. Increased suspicion of a focal inflammatory process such as localized scleroderma with regional inflammatory myopathy or another focal myopathy must be considered. The next diagnostic steps would include measuring the creatine kinase level, as well as obtaining an MRI of the leg to assess the nature and extent of the myopathy.

Multidisciplinary involvement, including hematology, rheumatology, and surgery, aided in narrowing the differential diagnosis. On hospital day 10, an MRI of the left thigh was performed for suspicion of diabetic myonecrosis (Figure 4). The MRI revealed a 10 cm × 3.6 cm × 22 cm intramuscular hematoma in the belly of the vastus lateralis muscle with associated soft tissue swelling, overlying subcutaneous edema, and skin thickening that was suggestive of hemorrhagic diabetic myonecrosis with some atypical features. A rheumatology consult was requested to evaluate for possible vasculitis in the left lower extremity, and vasculitis was not considered likely. The diagnosis of diabetic myonecrosis with associated intramuscular hemorrhage secondary to apixaban was made after careful reconsideration of the clinical presentation, imaging and laboratory data, and overall picture. Based on the clinical findings, imaging results, and exclusion of alternative causative pathologies of thigh swelling, no biopsy was performed, as it was not considered necessary to make the diagnosis of diabetic myonecrosis. The patient was discharged on hospital day 11 and was doing well. She followed up with her primary care doctor and has regained normal function of her leg.

Diabetic myonecrosis, or diabetic muscle infarction, is an uncommon nontraumatic myopathy that occurs in patients with diabetes who develop acute, focal muscle pain without recent trauma. In this case, the muscle infarction was further complicated by hemorrhagic transformation. Diabetic myonecrosis is relatively uncommon and a diagnosis made by combining history, examination, and laboratory findings and excluding other alternative conditions.

A clear schema for approaching the patient with acute, nontraumatic myopathies is important in avoiding diagnostic error. One effective schema is to divide myopathy into infectious and noninfectious categories. Causes of infectious myopathy include bacterial infections (eg, pyomyositis), inflammatory damage to muscles associated with viruses (eg, influenza), as well as rarer causes. Bacterial processes tend to be relatively focal and affect a specific muscle group or anatomic compartment, while viral causes are often more diffuse and occur in the context of a systemic viral syndrome. Bacterial causes range in severity, and life-threatening conditions, such as necrotizing soft tissue infection, must be considered. In this case, bacterial causes were less likely given the patient’s lack of fever, leukocytosis, and systemic signs of infection.^1,2 However, these findings are not uniformly sensitive, and clinicians should not exclude potentially life- or limb-threatening infections without thorough evaluation. For example, pyomyositis may present without fever in the subacute stage, without leukocytosis if the patient is immunocompromised, and without overt pus if the infection is not in the suppurative stage.³ Viral causes were made less likely in this patient given the lack of a current or recent systemic viral syndrome.

Once infectious etiologies are deemed unlikely, noninfectious etiologies for nontraumatic myopathies should be considered. Some causes of noninfectious myopathy present with the muscle symptoms as a predominant feature, while others present in the context of another illness such as cancer, metabolic disorders, or other systemic disorders. Many noninfectious causes of myopathy associated with systemic illnesses have diffuse or relatively diffuse symptoms, with pain and/or weakness in multiple muscle groups, often in a bilateral distribution. Such examples include dermatomyositis and polymyositis as well as myositis associated with other rheumatologic conditions. Nontraumatic rhabdomyolysis is diffuse and can occur in association with medications and/or genetic conditions.

Angervall and Stener⁴ first described diabetic myonecrosis in 1965 as tumoriform focal muscular degeneration due to diabetic microangiopathy. The most commonly affected muscle groups in diabetic myonecrosis are the anterior thigh, calf, and posterior thigh, followed by muscles in the upper extremities.⁵ Patients with diabetic myonecrosis have an overall mean age at presentation of 44.6 years; affected patients with type 1 diabetes mellitus present at a mean age nearly 20 years younger than those with type 2 diabetes mellitus (35.9 years vs 52.2 years, respectively).⁶ Patients tend to have a long (often >15 years) history of diabetes with microvascular complications such as retinopathy (reported in 71%), nephropathy (reported in 57%), and/or neuropathy (reported in 55%).⁷

The mainstay of the diagnosis of diabetic myonecrosis is a thorough history and physical examination and imaging. Routine laboratory evaluation is relatively unhelpful in diagnosing diabetic myonecrosis, but appropriate imaging can provide valuable supportive information. A CT scan and MRI are both helpful in excluding other etiologies as well as identifying features consistent with diabetic myonecrosis. A CT scan can help exclude a localized abscess, tumor, or bone destruction and, in affected patients, may show increased subcutaneous attenuation and increased muscle size with decreased attenuation secondary to edema.² However, a CT scan may not give optimal assessment of muscle tissue, and therefore MRI may need to be considered. MRI T2 images have a sensitivity nearing 90% for detecting myonecrosis.¹ The diagnostic value of MRI often obviates the need for muscle biopsy.

Spontaneous infarction with hemorrhagic features seen on imaging can be explained by a combination of damage from atherosclerotic or microvascular disease, an activated coagulation cascade, and an impaired fibrinolytic pathway.⁸ Hemorrhagic conversion in diabetic myonecrosis appears to be uncommon.⁹ In our case, we suspect that it developed because of the combination of bleeding risk from apixaban and the underlying mechanisms of diabetic myonecrosis.

The treatment of diabetic myonecrosis is mainly supportive, with an emphasis on rest, nonsteroidal anti-inflammatory agents, antiplatelet agents, and strict glycemic control.¹⁰ There is conflicting information about the value of limb immobilization versus active physical therapy as appropriate treatment modalities.¹¹ Patients who present with clinical concern for sepsis or compartment syndrome require consultation for consideration of acute surgical intervention.¹⁰ The short-term prognosis is promising with supportive therapy, but the condition may recur.¹² The recurrence rate may be as high as 40%, with a 2-year mortality of 10%.¹³ Ultimately, patients need to be followed closely in the outpatient setting to reduce the risk of recurrence.

In this patient, the simultaneous occurrence of focal pain and acute blood loss anemia led to a diagnosis of diabetic myonecrosis that was complicated by hemorrhagic conversion, a truly painful coincidence. The patient underwent a thorough evaluation for acute blood loss before the diagnosis was ultimately made. Clinicians should consider diabetic myonecrosis in patients with diabetes who present with acute muscle pain but no evidence of infection.

• Diabetic myonecrosis is an underrecognized entity and should be included in the differential diagnosis for patients with diabetes who present with acute muscle pain and no history of trauma.
• Imaging with CT and/or MRI of the affected region is the mainstay of diagnosis; treatment is predicated on severity and risk factors and can range from conservative therapy to operative intervention.
• Although the prognosis is good in these patients, careful outpatient follow-up is necessary to oversee their recovery to help reduce the risk of recurrence.

The authors thank Dr Vijay Singh for his radiology input on image selection for this manuscript.