Care of hospitalized patients requires effective teamwork within groups composed of physicians (eg, residents, hospitalists, specialists), advanced practice providers, nurses, patient‐care technicians, pharmacists, social workers, and therapists. Sadly, hospital‐based team members often fail to communicate. For example, 2 studies found that nurses and physicians communicated with one another on only 50% to 60% of their patients' hospital days, resulting in a lack of a mutual understanding of the plan of care.[1, 2]

Failure to communicate effectively may be because the hospital setting poses important challenges to teamwork, including the use of large teams with membership that changes frequently because of the need to provide care around the clock. Furthermore, individual team members often have high workloads, care for multiple patients simultaneously, and are seldom in the same place at the same time.

Interdisciplinary rounds (IDR) are a microsystem‐level solution with the goal to share information, achieve mutual understanding, and collaboratively revise the plan of care within care teams. Though common, IDR look very different across hospitals, making studies that evaluate novel strategies to improve IDR and measure their impact of great interest to hospital medicine.

In this issue of the Journal of Hospital Medicine, Bhamidipati and colleagues present a systematic review of published studies evaluating the effect of IDR on patient outcomes.[3] The systematic review included 22 studies, including 12 experimental/quasiexperimental and 10 observational studies. Overall, 13 studies were of low to medium quality, and 9 were high quality. Importantly, relatively few studies reported the degree to which IDR were implemented as planned. The investigators found evidence that IDR had a positive effect on length of stay (LOS) and staff satisfaction, but little evidence to support an effect on patient safety or satisfaction. Furthermore, the investigators found significant variability in IDR design and team composition. Some of this variation is to be expected, as IDR, like other interventions to improve quality and safety of patient care in complex settings, should be implemented with an expectation that the team may need to make adaptations based on local contextual factors such as workload (eg, daily census), environment (eg, open vs closed intensive care unit), local politics (eg, uniquely strong support for/against the intervention), and prior experience (eg, prior failed, similar interventions).[4, 5] Moreover, objectives for IDR may differ across settings. Some hospitals may have room (and a need) to improve LOS, whereas others may prioritize improving patient safety or patient experience metrics.

Bhamidipati and colleagues explain that their review did not reveal a causal pathway between IDR design and outcomes. We believe this lack of association is because most of the included studies did not propose a causal pathway between the IDR components implemented and the outcomes assessed. That is, few studies referred to conceptual models that explain how components of the IDR intervention might influence downstream patient outcomes.

IDR have the potential to influence a number of patient outcomes, including those reflecting efficiency (eg, length of stay), patient safety (eg, adverse events), and patient centeredness (eg, patient satisfaction). However, these outcomes are influenced by many factors, including patient characteristics and other efforts to improve care. As explained by the investigators, the results of many of the included studies may have been confounded due to relatively weak study designs and statistical analyses. Importantly, few of the studies included in this review report the more proximal measure of teamwork. If we hypothesize that IDR improve patient outcomes, they do so by improving teamwork. After all, the purpose of IDR is to assemble team members so they can communicate about and coordinate care. Measuring teamwork behaviors is difficult, especially on medical services. Measuring teamwork climate, the measurable aspects of team culture, is relatively easy. A recent systematic review of teamwork climate assessments in internal medicine identified the Safety Attitudes Questionnaire and the Team Climate Inventory as having substantial validity evidence and association with improved patient outcomes.[6]

Bhamidipati and colleagues proposed a definition for IDR and taxonomy for IDR design and reporting based on their systematic review. Although very useful, the IDR definition may be too limiting as evidenced by the fact that very few studies would be included in a systematic review using this definition as the inclusion criteria. Their proposed taxonomy should serve as a useful framework for future research efforts and appropriately recommends reporting of site characteristics, components of IDR design, and outcomes.

The systematic review by Bhamidipati et al. must also be interpreted in conjunction with another recently published systematic review by Pannick and colleagues assessing the effect of interdisciplinary team care interventions on general medical wards.[7] Contrary to the findings of the Bhamidipati et al. study, Pannick and colleagues found that most interdisciplinary team care interventions had no effect on LOS, but that half of the studies found an improvement in complications of care. Importantly, Pannick and colleagues included only experimental and quasiexperimental studies in their systematic review (ie, no observational studies).

There is clearly more work to be done in researching IDR and other interventions to improve teamwork in general medical settings. Larger studies are needed to provide sufficient power to detect improvement in outcomes. Future studies need to report the degree to which interventions are implemented as planned and need to use stronger study designs (eg, cluster randomized control or interrupted time series) to avoid the influence of confounders. Qualitative methods should be used to assess the influence of contextual factors on the success of interventions.[4] Most importantly, future studies should be based on conceptual models that explain how components of the intervention influence proximal measures of teamwork and downstream patient outcomes.

In the meantime, what is a hospital leader to do? We believe efforts to improve IDR are warranted, but that IDR program leaders need to first specify their primary objective(s). For example, in some hospitals, there may be little room to further reduce LOS, so another goalreducing preventable readmissions or reducing adverse eventsmight be specified as the key performance indicator. This crucial first step of creating a shared goal informs the design, implementation, and evaluation of IDR. We also believe that geographic localization of physicians to specific units is foundational to improving IDR. Physicians cannot feasibly attend IDR if their patients are spread across multiple units (or buildings). Finally, hospital leaders also need to view IDR as part of a larger set of interventions to improve teamwork. Leaders need to assess the adequacy of staffing levels, workflow, and team composition.[8] Unit‐based interdisciplinary leadership models should be used to help link efforts at various levels within a larger system.[9] These models designate a unit medical director and nurse manager who are jointly responsible for unit performance.

In conclusion, IDR play an important role in improving patient outcomes, but only do so by improving teamwork. In redesigning IDR, leaders need to be thoughtful about what outcomes IDR can affect, how IDR affect them, and how IDR fit into larger‐scale efforts to improve performance.

Care of hospitalized patients requires effective teamwork within groups composed of physicians (eg, residents, hospitalists, specialists), advanced practice providers, nurses, patient‐care technicians, pharmacists, social workers, and therapists. Sadly, hospital‐based team members often fail to communicate. For example, 2 studies found that nurses and physicians communicated with one another on only 50% to 60% of their patients' hospital days, resulting in a lack of a mutual understanding of the plan of care.[1, 2]

Failure to communicate effectively may be because the hospital setting poses important challenges to teamwork, including the use of large teams with membership that changes frequently because of the need to provide care around the clock. Furthermore, individual team members often have high workloads, care for multiple patients simultaneously, and are seldom in the same place at the same time.

Interdisciplinary rounds (IDR) are a microsystem‐level solution with the goal to share information, achieve mutual understanding, and collaboratively revise the plan of care within care teams. Though common, IDR look very different across hospitals, making studies that evaluate novel strategies to improve IDR and measure their impact of great interest to hospital medicine.

In this issue of the Journal of Hospital Medicine, Bhamidipati and colleagues present a systematic review of published studies evaluating the effect of IDR on patient outcomes.[3] The systematic review included 22 studies, including 12 experimental/quasiexperimental and 10 observational studies. Overall, 13 studies were of low to medium quality, and 9 were high quality. Importantly, relatively few studies reported the degree to which IDR were implemented as planned. The investigators found evidence that IDR had a positive effect on length of stay (LOS) and staff satisfaction, but little evidence to support an effect on patient safety or satisfaction. Furthermore, the investigators found significant variability in IDR design and team composition. Some of this variation is to be expected, as IDR, like other interventions to improve quality and safety of patient care in complex settings, should be implemented with an expectation that the team may need to make adaptations based on local contextual factors such as workload (eg, daily census), environment (eg, open vs closed intensive care unit), local politics (eg, uniquely strong support for/against the intervention), and prior experience (eg, prior failed, similar interventions).[4, 5] Moreover, objectives for IDR may differ across settings. Some hospitals may have room (and a need) to improve LOS, whereas others may prioritize improving patient safety or patient experience metrics.

Bhamidipati and colleagues explain that their review did not reveal a causal pathway between IDR design and outcomes. We believe this lack of association is because most of the included studies did not propose a causal pathway between the IDR components implemented and the outcomes assessed. That is, few studies referred to conceptual models that explain how components of the IDR intervention might influence downstream patient outcomes.

IDR have the potential to influence a number of patient outcomes, including those reflecting efficiency (eg, length of stay), patient safety (eg, adverse events), and patient centeredness (eg, patient satisfaction). However, these outcomes are influenced by many factors, including patient characteristics and other efforts to improve care. As explained by the investigators, the results of many of the included studies may have been confounded due to relatively weak study designs and statistical analyses. Importantly, few of the studies included in this review report the more proximal measure of teamwork. If we hypothesize that IDR improve patient outcomes, they do so by improving teamwork. After all, the purpose of IDR is to assemble team members so they can communicate about and coordinate care. Measuring teamwork behaviors is difficult, especially on medical services. Measuring teamwork climate, the measurable aspects of team culture, is relatively easy. A recent systematic review of teamwork climate assessments in internal medicine identified the Safety Attitudes Questionnaire and the Team Climate Inventory as having substantial validity evidence and association with improved patient outcomes.[6]

Bhamidipati and colleagues proposed a definition for IDR and taxonomy for IDR design and reporting based on their systematic review. Although very useful, the IDR definition may be too limiting as evidenced by the fact that very few studies would be included in a systematic review using this definition as the inclusion criteria. Their proposed taxonomy should serve as a useful framework for future research efforts and appropriately recommends reporting of site characteristics, components of IDR design, and outcomes.

The systematic review by Bhamidipati et al. must also be interpreted in conjunction with another recently published systematic review by Pannick and colleagues assessing the effect of interdisciplinary team care interventions on general medical wards.[7] Contrary to the findings of the Bhamidipati et al. study, Pannick and colleagues found that most interdisciplinary team care interventions had no effect on LOS, but that half of the studies found an improvement in complications of care. Importantly, Pannick and colleagues included only experimental and quasiexperimental studies in their systematic review (ie, no observational studies).

There is clearly more work to be done in researching IDR and other interventions to improve teamwork in general medical settings. Larger studies are needed to provide sufficient power to detect improvement in outcomes. Future studies need to report the degree to which interventions are implemented as planned and need to use stronger study designs (eg, cluster randomized control or interrupted time series) to avoid the influence of confounders. Qualitative methods should be used to assess the influence of contextual factors on the success of interventions.[4] Most importantly, future studies should be based on conceptual models that explain how components of the intervention influence proximal measures of teamwork and downstream patient outcomes.

In the meantime, what is a hospital leader to do? We believe efforts to improve IDR are warranted, but that IDR program leaders need to first specify their primary objective(s). For example, in some hospitals, there may be little room to further reduce LOS, so another goalreducing preventable readmissions or reducing adverse eventsmight be specified as the key performance indicator. This crucial first step of creating a shared goal informs the design, implementation, and evaluation of IDR. We also believe that geographic localization of physicians to specific units is foundational to improving IDR. Physicians cannot feasibly attend IDR if their patients are spread across multiple units (or buildings). Finally, hospital leaders also need to view IDR as part of a larger set of interventions to improve teamwork. Leaders need to assess the adequacy of staffing levels, workflow, and team composition.[8] Unit‐based interdisciplinary leadership models should be used to help link efforts at various levels within a larger system.[9] These models designate a unit medical director and nurse manager who are jointly responsible for unit performance.

In conclusion, IDR play an important role in improving patient outcomes, but only do so by improving teamwork. In redesigning IDR, leaders need to be thoughtful about what outcomes IDR can affect, how IDR affect them, and how IDR fit into larger‐scale efforts to improve performance.

Care of hospitalized patients requires effective teamwork within groups composed of physicians (eg, residents, hospitalists, specialists), advanced practice providers, nurses, patient‐care technicians, pharmacists, social workers, and therapists. Sadly, hospital‐based team members often fail to communicate. For example, 2 studies found that nurses and physicians communicated with one another on only 50% to 60% of their patients' hospital days, resulting in a lack of a mutual understanding of the plan of care.[1, 2]

Failure to communicate effectively may be because the hospital setting poses important challenges to teamwork, including the use of large teams with membership that changes frequently because of the need to provide care around the clock. Furthermore, individual team members often have high workloads, care for multiple patients simultaneously, and are seldom in the same place at the same time.

Interdisciplinary rounds (IDR) are a microsystem‐level solution with the goal to share information, achieve mutual understanding, and collaboratively revise the plan of care within care teams. Though common, IDR look very different across hospitals, making studies that evaluate novel strategies to improve IDR and measure their impact of great interest to hospital medicine.

In this issue of the Journal of Hospital Medicine, Bhamidipati and colleagues present a systematic review of published studies evaluating the effect of IDR on patient outcomes.[3] The systematic review included 22 studies, including 12 experimental/quasiexperimental and 10 observational studies. Overall, 13 studies were of low to medium quality, and 9 were high quality. Importantly, relatively few studies reported the degree to which IDR were implemented as planned. The investigators found evidence that IDR had a positive effect on length of stay (LOS) and staff satisfaction, but little evidence to support an effect on patient safety or satisfaction. Furthermore, the investigators found significant variability in IDR design and team composition. Some of this variation is to be expected, as IDR, like other interventions to improve quality and safety of patient care in complex settings, should be implemented with an expectation that the team may need to make adaptations based on local contextual factors such as workload (eg, daily census), environment (eg, open vs closed intensive care unit), local politics (eg, uniquely strong support for/against the intervention), and prior experience (eg, prior failed, similar interventions).[4, 5] Moreover, objectives for IDR may differ across settings. Some hospitals may have room (and a need) to improve LOS, whereas others may prioritize improving patient safety or patient experience metrics.

Bhamidipati and colleagues explain that their review did not reveal a causal pathway between IDR design and outcomes. We believe this lack of association is because most of the included studies did not propose a causal pathway between the IDR components implemented and the outcomes assessed. That is, few studies referred to conceptual models that explain how components of the IDR intervention might influence downstream patient outcomes.

IDR have the potential to influence a number of patient outcomes, including those reflecting efficiency (eg, length of stay), patient safety (eg, adverse events), and patient centeredness (eg, patient satisfaction). However, these outcomes are influenced by many factors, including patient characteristics and other efforts to improve care. As explained by the investigators, the results of many of the included studies may have been confounded due to relatively weak study designs and statistical analyses. Importantly, few of the studies included in this review report the more proximal measure of teamwork. If we hypothesize that IDR improve patient outcomes, they do so by improving teamwork. After all, the purpose of IDR is to assemble team members so they can communicate about and coordinate care. Measuring teamwork behaviors is difficult, especially on medical services. Measuring teamwork climate, the measurable aspects of team culture, is relatively easy. A recent systematic review of teamwork climate assessments in internal medicine identified the Safety Attitudes Questionnaire and the Team Climate Inventory as having substantial validity evidence and association with improved patient outcomes.[6]

Bhamidipati and colleagues proposed a definition for IDR and taxonomy for IDR design and reporting based on their systematic review. Although very useful, the IDR definition may be too limiting as evidenced by the fact that very few studies would be included in a systematic review using this definition as the inclusion criteria. Their proposed taxonomy should serve as a useful framework for future research efforts and appropriately recommends reporting of site characteristics, components of IDR design, and outcomes.

The systematic review by Bhamidipati et al. must also be interpreted in conjunction with another recently published systematic review by Pannick and colleagues assessing the effect of interdisciplinary team care interventions on general medical wards.[7] Contrary to the findings of the Bhamidipati et al. study, Pannick and colleagues found that most interdisciplinary team care interventions had no effect on LOS, but that half of the studies found an improvement in complications of care. Importantly, Pannick and colleagues included only experimental and quasiexperimental studies in their systematic review (ie, no observational studies).

There is clearly more work to be done in researching IDR and other interventions to improve teamwork in general medical settings. Larger studies are needed to provide sufficient power to detect improvement in outcomes. Future studies need to report the degree to which interventions are implemented as planned and need to use stronger study designs (eg, cluster randomized control or interrupted time series) to avoid the influence of confounders. Qualitative methods should be used to assess the influence of contextual factors on the success of interventions.[4] Most importantly, future studies should be based on conceptual models that explain how components of the intervention influence proximal measures of teamwork and downstream patient outcomes.

In the meantime, what is a hospital leader to do? We believe efforts to improve IDR are warranted, but that IDR program leaders need to first specify their primary objective(s). For example, in some hospitals, there may be little room to further reduce LOS, so another goalreducing preventable readmissions or reducing adverse eventsmight be specified as the key performance indicator. This crucial first step of creating a shared goal informs the design, implementation, and evaluation of IDR. We also believe that geographic localization of physicians to specific units is foundational to improving IDR. Physicians cannot feasibly attend IDR if their patients are spread across multiple units (or buildings). Finally, hospital leaders also need to view IDR as part of a larger set of interventions to improve teamwork. Leaders need to assess the adequacy of staffing levels, workflow, and team composition.[8] Unit‐based interdisciplinary leadership models should be used to help link efforts at various levels within a larger system.[9] These models designate a unit medical director and nurse manager who are jointly responsible for unit performance.

In conclusion, IDR play an important role in improving patient outcomes, but only do so by improving teamwork. In redesigning IDR, leaders need to be thoughtful about what outcomes IDR can affect, how IDR affect them, and how IDR fit into larger‐scale efforts to improve performance.

Up to 10% of the acutely ill patients in hospitals today are not admitted, but cared for under outpatient observation.[1, 2] Hospitals use observation services to replace inpatient services for patients who do not meet inpatient illness standards. Use of this technique for Medicare beneficiaries has grown in recent years, and future policy changes may further increase the percentage of hospital stays classified as observation.[3, 4]

Increased use of observation services has cost‐sharing implications for Medicare beneficiaries. Because hospital observation stays are paid through Medicare's outpatient (Part B) rather than inpatient (Part A) benefit, beneficiaries do not have an out‐of‐pocket maximum per hospital stay.[3] Although prior analyses documented mean out‐of‐pocket costs of $400 to $600 per stay, out‐of‐pocket costs may exceed the Part A deductible and leave beneficiaries responsible for hospital costs that may have been reimbursed by Part A in an inpatient stay.[5, 6, 7] Approximately 6% to 10% of Medicare observation stays result in out‐of‐pocket costs exceeding the Part A deductible nationally, and recent work in this journal documented that 26.6% of beneficiaries with repeat observation stays within a 60‐day period may pay more than the 1‐time deductible‐based payment that beneficiaries are responsible for under Medicare Part A.[5, 6, 7]

However, prior analyses of beneficiary out‐of‐pocket costs did not account for supplemental insurance payments. Approximately 80% to 90% of fee‐for‐service Medicare beneficiaries who use either a private (Medigap or employer based) or state‐based (Medicaid or other plans) supplemental insurance plan.[8, 9, 10] The effect of supplemental insurance on out‐of‐pocket costs has been documented for inpatient stays, yet has not been explored for observation stays.[9] We sought to describe Medicare beneficiaries' out‐of‐pocket costs by accounting for payments from all insurers.

METHODS

We obtained payment data from 2 affiliated hospitals for all Medicare observation hospital stays between April 2013 and March 2014. Stays insured by Medicare Advantage plans (Part C) were excluded, and charges from skilled nursing facility (SNF) stays were not available. Although the exact origin of each stay was not available, the majority of stays at these hospitals came from emergency visits, direct placement from outpatient providers, and postoutpatient procedural monitoring. The dataset included charges, insurance adjustments and payments from all sources, diagnoses, and demographics. Insurance adjustments represented the reductions applied to total stay charges by each insurer in congruence with their contract with the health system. Our dataset included charges for all hospital materials and services usually billed under Part A, including medications. Out‐of‐pocket costs were calculated by subtracting insurance adjustments and payments from the total charges for each stay. To identify potential cost‐shifting from Medicare to beneficiaries, we compared out‐of‐pocket costs to the Part A deductible ($1184 for stays in 2013 and $1216 for 2014). Household income data were estimated using zip code and Internal Revenue Service data for 2013.[11]

The University of California Los Angeles institutional review board approved this study. Statistical significance was calculated using a 2‐tailed unpaired t test for means and ² for percentages.

RESULTS

There were 2029 total observation stays during the study period, representing 5.0% of all discharges from both hospitals. Medicare beneficiaries accounted for 722 of those observation stays. Among the 498 finalized Medicare observation stays, the median patient age was 73 years, and median household income was $50,591. The median length of stay was 25 hours, with 1.8% of stays lasting longer than 2 midnights. Seventy percent of beneficiaries had private supplemental insurance, whereas 6% had state‐based supplemental plans. Table 1 presents detailed costs. Out‐of‐pocket costs ranged from $0 to $16,196. Stays without supplemental insurance had mean and median out of pocket costs of $537 and $286, respectively. The mean out‐of‐pocket costs for stays with private supplemental insurance decreased to $45 (P < 0.01) and $168 (P = 0.21) for stays with state‐based supplemental insurance, with a median below $1 for both. On average, beneficiaries without supplemental insurance were responsible for $654 less than the Part A deductible. Thirteen beneficiaries had multiple finalized stays within 60 days, with a mean out‐of‐pocket cost of $119, median of $20, and 1 stay produced an out‐of‐pocket cost exceeding the Part A deductible. An additional 224 Medicare stays not finalized because of missing supplemental payments had a mean out‐of‐pocket cost of $125 (P < 0.01) and median of $5 after applying supplemental insurer adjustments (not shown in Table 1).

A minority of observation stays produced out‐of‐pocket costs exceeding the Part A deductible. Those percentages were 7.6% for stays without a supplemental insurer, 3.5% (P < 0.01) for stays with a state‐based supplemental insurer, and 0.3% (P < 0.01) for stays with a private supplemental insurer. Of the 224 nonfinalized Medicare stays, 1.3% (P < 0.01) exceeded the Part A deductible after supplemental insurer adjustments.

DISCUSSION

This study demonstrates that supplemental insurance can dramatically reduce Medicare beneficiaries' out‐of‐pocket costs in observation services. Mean out‐of‐pocket costs of $45 and $168 for stays with private and state‐based supplemental insurer plans are significantly lower than prior estimates calculated without supplemental insurance information, as are the percentages of stays with out‐of‐pocket costs exceeding the Part A deductible, at 0.3% and 3.5%, respectively.[5, 6, 7] Because the majority of Medicare beneficiaries use supplemental insurance, excessive out‐of‐pockets in observation services may occur less frequently than previously reported.[9, 10] Clinicians concerned about excessive out‐of‐pocket costs for Medicare beneficiaries can be reassured they are usually modest for beneficiaries with supplemental insurance.

This study's mean out‐of‐pocket cost and percentage of stays with out‐of‐pocket costs exceeding the Part A deductible for beneficiaries without supplemental insurance are similar to results from prior national analyses performed without supplemental insurer information.[5, 6, 7] But this study was limited by a small sample size from 2 affiliated hospitals, with few repeat observation stays within a 60‐day period. In addition, posthospitalization SNF fees were not included, which traditionally have been a significant source of out‐of‐pocket costs in observation services.[3, 7] Populations with supplemental insurance treated elsewhere may incur hospital out‐of‐pocket costs differing from these results due to dissimilarities in the presence and quality of supplemental insurance.

However, most Medigap plans are federally regulated to cover the majority of out‐of‐pockets unpaid by Medicare.[12] Medicaid plans usually place limits on out‐of‐pocket costs, and any other state or employer‐based supplemental plans will also reduce out‐of‐pocket costs.[13] Thus, it is likely accurate to assume mean observation services out‐of‐pocket costs for hospital fees are lower than previously reported by national analyses performed without supplemental insurance information. Attempts at estimating beneficiary out‐of‐pocket costs in the future should account for supplemental insurance adjustments and payments.

Up to 10% of the acutely ill patients in hospitals today are not admitted, but cared for under outpatient observation.[1, 2] Hospitals use observation services to replace inpatient services for patients who do not meet inpatient illness standards. Use of this technique for Medicare beneficiaries has grown in recent years, and future policy changes may further increase the percentage of hospital stays classified as observation.[3, 4]

Increased use of observation services has cost‐sharing implications for Medicare beneficiaries. Because hospital observation stays are paid through Medicare's outpatient (Part B) rather than inpatient (Part A) benefit, beneficiaries do not have an out‐of‐pocket maximum per hospital stay.[3] Although prior analyses documented mean out‐of‐pocket costs of $400 to $600 per stay, out‐of‐pocket costs may exceed the Part A deductible and leave beneficiaries responsible for hospital costs that may have been reimbursed by Part A in an inpatient stay.[5, 6, 7] Approximately 6% to 10% of Medicare observation stays result in out‐of‐pocket costs exceeding the Part A deductible nationally, and recent work in this journal documented that 26.6% of beneficiaries with repeat observation stays within a 60‐day period may pay more than the 1‐time deductible‐based payment that beneficiaries are responsible for under Medicare Part A.[5, 6, 7]

However, prior analyses of beneficiary out‐of‐pocket costs did not account for supplemental insurance payments. Approximately 80% to 90% of fee‐for‐service Medicare beneficiaries who use either a private (Medigap or employer based) or state‐based (Medicaid or other plans) supplemental insurance plan.[8, 9, 10] The effect of supplemental insurance on out‐of‐pocket costs has been documented for inpatient stays, yet has not been explored for observation stays.[9] We sought to describe Medicare beneficiaries' out‐of‐pocket costs by accounting for payments from all insurers.

METHODS

We obtained payment data from 2 affiliated hospitals for all Medicare observation hospital stays between April 2013 and March 2014. Stays insured by Medicare Advantage plans (Part C) were excluded, and charges from skilled nursing facility (SNF) stays were not available. Although the exact origin of each stay was not available, the majority of stays at these hospitals came from emergency visits, direct placement from outpatient providers, and postoutpatient procedural monitoring. The dataset included charges, insurance adjustments and payments from all sources, diagnoses, and demographics. Insurance adjustments represented the reductions applied to total stay charges by each insurer in congruence with their contract with the health system. Our dataset included charges for all hospital materials and services usually billed under Part A, including medications. Out‐of‐pocket costs were calculated by subtracting insurance adjustments and payments from the total charges for each stay. To identify potential cost‐shifting from Medicare to beneficiaries, we compared out‐of‐pocket costs to the Part A deductible ($1184 for stays in 2013 and $1216 for 2014). Household income data were estimated using zip code and Internal Revenue Service data for 2013.[11]

The University of California Los Angeles institutional review board approved this study. Statistical significance was calculated using a 2‐tailed unpaired t test for means and ² for percentages.

RESULTS

There were 2029 total observation stays during the study period, representing 5.0% of all discharges from both hospitals. Medicare beneficiaries accounted for 722 of those observation stays. Among the 498 finalized Medicare observation stays, the median patient age was 73 years, and median household income was $50,591. The median length of stay was 25 hours, with 1.8% of stays lasting longer than 2 midnights. Seventy percent of beneficiaries had private supplemental insurance, whereas 6% had state‐based supplemental plans. Table 1 presents detailed costs. Out‐of‐pocket costs ranged from $0 to $16,196. Stays without supplemental insurance had mean and median out of pocket costs of $537 and $286, respectively. The mean out‐of‐pocket costs for stays with private supplemental insurance decreased to $45 (P < 0.01) and $168 (P = 0.21) for stays with state‐based supplemental insurance, with a median below $1 for both. On average, beneficiaries without supplemental insurance were responsible for $654 less than the Part A deductible. Thirteen beneficiaries had multiple finalized stays within 60 days, with a mean out‐of‐pocket cost of $119, median of $20, and 1 stay produced an out‐of‐pocket cost exceeding the Part A deductible. An additional 224 Medicare stays not finalized because of missing supplemental payments had a mean out‐of‐pocket cost of $125 (P < 0.01) and median of $5 after applying supplemental insurer adjustments (not shown in Table 1).

A minority of observation stays produced out‐of‐pocket costs exceeding the Part A deductible. Those percentages were 7.6% for stays without a supplemental insurer, 3.5% (P < 0.01) for stays with a state‐based supplemental insurer, and 0.3% (P < 0.01) for stays with a private supplemental insurer. Of the 224 nonfinalized Medicare stays, 1.3% (P < 0.01) exceeded the Part A deductible after supplemental insurer adjustments.

DISCUSSION

This study demonstrates that supplemental insurance can dramatically reduce Medicare beneficiaries' out‐of‐pocket costs in observation services. Mean out‐of‐pocket costs of $45 and $168 for stays with private and state‐based supplemental insurer plans are significantly lower than prior estimates calculated without supplemental insurance information, as are the percentages of stays with out‐of‐pocket costs exceeding the Part A deductible, at 0.3% and 3.5%, respectively.[5, 6, 7] Because the majority of Medicare beneficiaries use supplemental insurance, excessive out‐of‐pockets in observation services may occur less frequently than previously reported.[9, 10] Clinicians concerned about excessive out‐of‐pocket costs for Medicare beneficiaries can be reassured they are usually modest for beneficiaries with supplemental insurance.

This study's mean out‐of‐pocket cost and percentage of stays with out‐of‐pocket costs exceeding the Part A deductible for beneficiaries without supplemental insurance are similar to results from prior national analyses performed without supplemental insurer information.[5, 6, 7] But this study was limited by a small sample size from 2 affiliated hospitals, with few repeat observation stays within a 60‐day period. In addition, posthospitalization SNF fees were not included, which traditionally have been a significant source of out‐of‐pocket costs in observation services.[3, 7] Populations with supplemental insurance treated elsewhere may incur hospital out‐of‐pocket costs differing from these results due to dissimilarities in the presence and quality of supplemental insurance.

However, most Medigap plans are federally regulated to cover the majority of out‐of‐pockets unpaid by Medicare.[12] Medicaid plans usually place limits on out‐of‐pocket costs, and any other state or employer‐based supplemental plans will also reduce out‐of‐pocket costs.[13] Thus, it is likely accurate to assume mean observation services out‐of‐pocket costs for hospital fees are lower than previously reported by national analyses performed without supplemental insurance information. Attempts at estimating beneficiary out‐of‐pocket costs in the future should account for supplemental insurance adjustments and payments.

Up to 10% of the acutely ill patients in hospitals today are not admitted, but cared for under outpatient observation.[1, 2] Hospitals use observation services to replace inpatient services for patients who do not meet inpatient illness standards. Use of this technique for Medicare beneficiaries has grown in recent years, and future policy changes may further increase the percentage of hospital stays classified as observation.[3, 4]

Increased use of observation services has cost‐sharing implications for Medicare beneficiaries. Because hospital observation stays are paid through Medicare's outpatient (Part B) rather than inpatient (Part A) benefit, beneficiaries do not have an out‐of‐pocket maximum per hospital stay.[3] Although prior analyses documented mean out‐of‐pocket costs of $400 to $600 per stay, out‐of‐pocket costs may exceed the Part A deductible and leave beneficiaries responsible for hospital costs that may have been reimbursed by Part A in an inpatient stay.[5, 6, 7] Approximately 6% to 10% of Medicare observation stays result in out‐of‐pocket costs exceeding the Part A deductible nationally, and recent work in this journal documented that 26.6% of beneficiaries with repeat observation stays within a 60‐day period may pay more than the 1‐time deductible‐based payment that beneficiaries are responsible for under Medicare Part A.[5, 6, 7]

However, prior analyses of beneficiary out‐of‐pocket costs did not account for supplemental insurance payments. Approximately 80% to 90% of fee‐for‐service Medicare beneficiaries who use either a private (Medigap or employer based) or state‐based (Medicaid or other plans) supplemental insurance plan.[8, 9, 10] The effect of supplemental insurance on out‐of‐pocket costs has been documented for inpatient stays, yet has not been explored for observation stays.[9] We sought to describe Medicare beneficiaries' out‐of‐pocket costs by accounting for payments from all insurers.

METHODS

We obtained payment data from 2 affiliated hospitals for all Medicare observation hospital stays between April 2013 and March 2014. Stays insured by Medicare Advantage plans (Part C) were excluded, and charges from skilled nursing facility (SNF) stays were not available. Although the exact origin of each stay was not available, the majority of stays at these hospitals came from emergency visits, direct placement from outpatient providers, and postoutpatient procedural monitoring. The dataset included charges, insurance adjustments and payments from all sources, diagnoses, and demographics. Insurance adjustments represented the reductions applied to total stay charges by each insurer in congruence with their contract with the health system. Our dataset included charges for all hospital materials and services usually billed under Part A, including medications. Out‐of‐pocket costs were calculated by subtracting insurance adjustments and payments from the total charges for each stay. To identify potential cost‐shifting from Medicare to beneficiaries, we compared out‐of‐pocket costs to the Part A deductible ($1184 for stays in 2013 and $1216 for 2014). Household income data were estimated using zip code and Internal Revenue Service data for 2013.[11]

The University of California Los Angeles institutional review board approved this study. Statistical significance was calculated using a 2‐tailed unpaired t test for means and ² for percentages.

RESULTS

There were 2029 total observation stays during the study period, representing 5.0% of all discharges from both hospitals. Medicare beneficiaries accounted for 722 of those observation stays. Among the 498 finalized Medicare observation stays, the median patient age was 73 years, and median household income was $50,591. The median length of stay was 25 hours, with 1.8% of stays lasting longer than 2 midnights. Seventy percent of beneficiaries had private supplemental insurance, whereas 6% had state‐based supplemental plans. Table 1 presents detailed costs. Out‐of‐pocket costs ranged from $0 to $16,196. Stays without supplemental insurance had mean and median out of pocket costs of $537 and $286, respectively. The mean out‐of‐pocket costs for stays with private supplemental insurance decreased to $45 (P < 0.01) and $168 (P = 0.21) for stays with state‐based supplemental insurance, with a median below $1 for both. On average, beneficiaries without supplemental insurance were responsible for $654 less than the Part A deductible. Thirteen beneficiaries had multiple finalized stays within 60 days, with a mean out‐of‐pocket cost of $119, median of $20, and 1 stay produced an out‐of‐pocket cost exceeding the Part A deductible. An additional 224 Medicare stays not finalized because of missing supplemental payments had a mean out‐of‐pocket cost of $125 (P < 0.01) and median of $5 after applying supplemental insurer adjustments (not shown in Table 1).

A minority of observation stays produced out‐of‐pocket costs exceeding the Part A deductible. Those percentages were 7.6% for stays without a supplemental insurer, 3.5% (P < 0.01) for stays with a state‐based supplemental insurer, and 0.3% (P < 0.01) for stays with a private supplemental insurer. Of the 224 nonfinalized Medicare stays, 1.3% (P < 0.01) exceeded the Part A deductible after supplemental insurer adjustments.

DISCUSSION

This study demonstrates that supplemental insurance can dramatically reduce Medicare beneficiaries' out‐of‐pocket costs in observation services. Mean out‐of‐pocket costs of $45 and $168 for stays with private and state‐based supplemental insurer plans are significantly lower than prior estimates calculated without supplemental insurance information, as are the percentages of stays with out‐of‐pocket costs exceeding the Part A deductible, at 0.3% and 3.5%, respectively.[5, 6, 7] Because the majority of Medicare beneficiaries use supplemental insurance, excessive out‐of‐pockets in observation services may occur less frequently than previously reported.[9, 10] Clinicians concerned about excessive out‐of‐pocket costs for Medicare beneficiaries can be reassured they are usually modest for beneficiaries with supplemental insurance.

This study's mean out‐of‐pocket cost and percentage of stays with out‐of‐pocket costs exceeding the Part A deductible for beneficiaries without supplemental insurance are similar to results from prior national analyses performed without supplemental insurer information.[5, 6, 7] But this study was limited by a small sample size from 2 affiliated hospitals, with few repeat observation stays within a 60‐day period. In addition, posthospitalization SNF fees were not included, which traditionally have been a significant source of out‐of‐pocket costs in observation services.[3, 7] Populations with supplemental insurance treated elsewhere may incur hospital out‐of‐pocket costs differing from these results due to dissimilarities in the presence and quality of supplemental insurance.

However, most Medigap plans are federally regulated to cover the majority of out‐of‐pockets unpaid by Medicare.[12] Medicaid plans usually place limits on out‐of‐pocket costs, and any other state or employer‐based supplemental plans will also reduce out‐of‐pocket costs.[13] Thus, it is likely accurate to assume mean observation services out‐of‐pocket costs for hospital fees are lower than previously reported by national analyses performed without supplemental insurance information. Attempts at estimating beneficiary out‐of‐pocket costs in the future should account for supplemental insurance adjustments and payments.

Delirium, a clinical syndrome characterized by inattention and acute cognitive dysfunction, is very common in older hospitalized patients, with a reported incidence of 18% to 35% at time of admission and overall occurrence rates of 29% to 64%.[1] Previous studies have reported that a diagnosis of delirium is not benign and is associated with other adverse outcomes including prolonged hospitalization, institutionalization, increased cost, and mortality. These outcomes occurred independent of age, prior cognitive functioning, and comorbidities.[2] Guidelines recommend that management of inpatient delirium should be focused on addressing the underlying etiology and managed with nonpharmacological interventions whenever possible.[3, 4, 5] However, implementing these recommendations can prove to be very challenging in hospital settings. Providers frequently have to resort to medical therapies, including antipsychotics (APs). Although these medications are commonly used to treat delirium in elderly patients, there is limited evidence to support their efficacy, and there are currently no proven pharmacological alternatives to these medications.[6] Furthermore, previous studies have demonstrated an increased risk of stroke, infection, cognitive impairment, and mortality in elders with dementia who receive long‐term AP therapy.[7, 8, 9] Yet as many as 48% of hospitalized elders who were newly started on APs had these drugs continued at time of discharge.[10]

There have been few studies describing the long‐term outcomes of elderly patient who are started on APs in the hospital. Most information on outcomes comes from patients with dementia. Therefore, we studied the 1‐year outcomes of a cohort of patients with and without dementia who were started on APs in the hospital and then discharged on these medications. In this cohort, we aimed to describe the number of readmissions, reasons for readmissions, duration of AP therapy, use of other sedating medications such as anxiolytics, hypnotics, and antihistamines as well as the incidence of readmission and death 1 year after the index hospital discharge.

METHODS

We previously described a retrospective cohort of 300 elders (65 years old) admitted to a tertiary care hospital between October 1, 2012 and September 31, 2013 who were newly prescribed APs while hospitalized.[10] Of patients alive at the time of discharge (260), 56% (146 patients) were discharged on APs. Two investigators extracted these 148 patient charts independently to identify and quantify the number of readmissions to the index hospital. We then limited the sample to only the first readmission per patient following the index admission and extracted this readmission for each patient. We first determined if APs were present on the admission medication reconciliation. If APs were not present on admission, we examined whether they were resumed during the hospitalization using the electronic medication administration summary. If they were present on admission, we looked to see if they were discontinued during the readmission and if additional new APs were started during the hospitalizations. We documented the circumstances around APs use and identified patients who died during their hospitalizations. We identified delirium using the same terms that were described in our prior study on the same cohort of patients.[10] We determined if patients were delirious using a predetermined algorithm (Figure 1). Briefly, we first determined delirium was documented. We then examined whether there was a Confusion Assessment Method (CAM) instrument included in the record. If a CAM instrument was not documented, we then looked for documentation using specific terms (eg, disorientation, confusions). We identified patients with dementia by determining whether dementia was documented along with other admission medical comorbidities. If it was not, we determined whether dementia was newly diagnosed during the hospital stay using progress notes or consultation notes. We did not objectively define criteria for diagnosis of dementia. We used the National Death Index (NDI) to determine mortality for all patients 1 year after discharge from the index hospitalization. The NDI is a national database of death records maintained by the National Center for Health Statistics. It has shown consistently high sensitivity and specificity for detection of death.[11]

Figure 1

We used descriptive statistics (means, standard deviations, range, and percents as appropriate to the scale of measurement) to describe the patient sample. We then used multiple logistic regression to identify significant predictors of death within 1 year of discharge.[12] Univariate analysis was used to select candidates for the logistic model (t tests for continuous factors and ² for discrete factors). All factors with a significance level <0.2 on univariate analysis were included in the logistic regression, in addition to age and sex (regardless of significance). A maximum likelihood procedure was used to calculate the regression coefficients for the logistic model. The likelihood ratio criterion was used to determine the significance of individual factors in the regression model.[13] Factors with a significance level of 0.15 or less were retained in the final model, in addition to age and sex.

RESULTS

The 260 patients discharged alive from their index admissions had a 1‐year mortality rate of 29% (75/260). Of the 146/260 patients discharged on APs, 60 (41%) patients experienced at least 1 readmission (mean = 2 readmissions per patient; range, 18, with 111 total readmissions for 60 patients) within 1 year from discharge (Figure 2). Most common diagnoses at the time of readmissions were related neurological and psychiatric disorders (14%), cardiovascular and circulation disorders (13%), renal injury and electrolyte disorders (11%), and infections (6%). Among patients with at least 1 readmission, the mean age was 81.3 (range, 65.599.7), 60% were male, and 45% were admitted from a skilled nursing facility or rehabilitation facility (Table 1). Median time to readmission was 43.5 days (range, 1343 days), and 79% were readmitted to a medical service. The remaining 20% were admitted to a surgical service. Inpatient mortality during first readmissions was 8% (5/60). At the time of first readmission, 39/60 (65%) of patients were still on the same APs on which they had been discharged, and the APs were continued during the hospitalization in 79% of the patients (61% quetiapine, 19% olanzapine, and 13% risperidone). About half of patients whose APs were discontinued prior to readmission received a new AP during their hospital stays (9/20; 45%). One patient had been started on quetiapine in the outpatient setting. No patients were found to have new benzodiazepines, nonbenzodiazepine hypnotic, or antihistamines on their admission medication list.

Figure 2

Eighteen patients received 1 or more new APs during the readmission hospitalizations. These included haloperidol (89%) and quetiapine (39%). Delirium was the main reported indication for starting APs (78%), but in 17% of cases no indication was documented. An electrocardiogram (ECG) was performed in 94% prior to APs administration and for 22% after APs administration. Corrected QT interval (QTc) of >500 ms was present in 18% of patients in pretreatment ECG and 50% of patients in post‐AP ECG. Of patients who survived readmission, 58% (32/55) were discharged to postacute facilities. Of the 39 patients who were on the same APs from index admission, 27 (69%) patients were eventually discharged on the same APs or new APs started during the readmission.

In the multivariable model (Table 2), predictors of death at 1 year included discharge to postacute facilities after index admission (odds ratio [OR]: 2.28; 95% confidence interval [CI]: 1.10‐4.73, P = 0.03) and QTc prolongation >500 ms during index admission (OR: 3.41; 95% CI: 1.34‐8.67, P = 0.01). Age and gender were not associated with 1‐year mortality.

DISCUSSION

In a cohort of elderly patients who were discharged on APs, nearly one‐third (29%) died within 1 year of the hospitalization in which APs were initiated. Nearly half of the survivors from the index admission (41%) experienced at least 1 admission within 1 year from discharge. Of readmitted patients, two‐thirds were taking the same APs that had been started during the index hospitalization. Half of the patients not on APs on readmission were started on an AP during the hospitalization, most often because they became delirious on return to the acute care setting. Compared to patients discharged home after an index admission, patients who were discharged to postacute facilities were almost 4 times as likely to die during the year subsequent to the admission. These data suggest that once patients are started on APs, most are continued on them until the next admission or are restarted during that readmission. Moreover, hospitalized elders who require an AP are at high risk for mortality in the coming year.

Prior studies have reported that patients with delirium have elevated 1‐year mortality rates.[14, 15, 16, 17, 18, 19] A secondary analysis of the Delirium Prevention Trial, which included 437 hospitalized older patients, revealed a 1‐year mortality rate of 20% in those who were never delirious during hospitalization, compared to 26% to 38% in patients with delirium.[19] Additionally, 1‐year mortality in hospitalized older patients with delirium (36%) was shown to be higher than patients with dementia (29%) or depression (26%).[17] Unlike these studies, not all of the patients in our study had documented delirium, but all received an AP. Still, it is notable that the 1‐year mortality rate for delirium in general is similar to what we found in this study.

The literature has also reported that long‐term AP use is associated with excess mortality in elder patients, especially those with dementia.[20, 21, 22] In a retrospective cohort study, older patients with dementia who were taking antipsychotics had significantly higher 1‐year mortality rates (23%29%) than patients not taking antipsychotic medications (15%). In a large Canadian propensity score‐matched cohort study that included over 13,000 demented older adults, the mortality was higher in the community‐dwelling elders who received atypical APs compared to no APs, with a difference of 1.1% in 180‐day mortality rate after initiation of APs.[21] The absolute mortality rate was 2.6% higher in patients who received typical compared to atypical APs. Unlike these studies, not every patient in our cohort had a diagnosis of dementia, but again, mortality rates in these studies appear similar to our cohort.

In contrast, other observational studies have not found an increased risk associated with receipt of APs. For example, a prospective study that enrolled approximately 950 patients with probable dementia showed that AP use was not associated with time to death after adjustment for comorbidities, demographic and cognitive variables.[23] These conflicting results highlight the difficulties of attributing outcomes in high‐risk populations. Although the excess mortality observed in patients taking APs may be related to the risks of APs, it is quite possible that patients who require APs (most often for delirium or agitated dementia) are at higher risk of death. This confounding by indication may be nearly impossible to adjust for retrospectively, even using techniques such as propensity matching.

Our report adds to the literature; we know of no studies to date describing a cohort of patients, most with delirium, who were started on APs in the hospital. We also attempted to identify the reasons that patients were started on APs, which have been infrequently reported. As noted above, our 1‐year mortality rate of 29% among older patients prescribed APs in the hospital was quite similar to mortality rates both for patients with delirium who were not necessarily treated with APs and patients with dementia who were treated with APs. This finding further supports the argument that risk factors for mortality, including dementia, delirium, and AP use are very difficult to tease apart. It is possible that the reasons that APs are prescribed (agitated delirium or dementia) have as much to do with the excess mortality reported in observational studies of APs as the use of APs themselves.

The high rate of continued AP use we observed (two‐thirds of readmitted patients) may reflect limited pharmacological alternatives to these medications with little evidence to support treating the symptoms of delirium with other drug classes, along with suboptimal environmental and behavioral modifications in postacute facilities and hospitals. This is unfortunate given that delirium is often preventable. Systematic implementation of well‐documented strategies to decrease delirium in hospitals and postacute facilities would likely reduce the prescription of APs and has the potential to slow the decline in this vulnerable population. A meta‐analysis incorporating both randomized and nonrandomized trials of medical and surgical patients showed that multicomponent nonpharmacologic interventions decreased delirium by 50%.[24] Thus, simple interventions such as reorientation, early mobilization, optimizing vision and hearing, sleepwake cycle preservation, and hydration might avoid roughly 1 million cases of delirium in hospitalized older adults annually.[24] The Hospital Elder Life Program and Acute Care for Elders units are examples of programs that have been shown to decrease the incidence of delirium.[25, 26]

Despite vigorous efforts to prevent delirium, a subgroup of patients still will become delirious. These patients are at high risk for death. Our mortality prediction model revealed that patients who were discharged to postacute facilities were 4 times more likely to die during the subsequent year compared to patients who were discharged home. Patients discharged to postacute facilities are likely to have a higher burden of disease, greater functional and cognitive impairment, and more frailty than those who are able to return to the community. Very ill and/or frail patients receiving APs in the hospital and requiring APs on discharge to postacute care facilities have limited survival and may benefit from expedited palliative care interventions to clarify prognosis and goals, and relieve suffering. At a minimum, our study identifies a need for further study to identify this very high‐risk group of elders. It is notable that 50% of patients were found to have a post‐treatment ECG with a QTc of >500 ms, a finding that has not been previously described. This would put these patients at higher risk of mortality, and as such we suggest that current guidelines should continue to emphasize the importance of post‐treatment ECGs and set clear criteria for discontinuation in elderly patients.

Our study is limited by its retrospective, single‐center design and small sample size, therefore limiting the interpretation and generalizability of the results to other hospitals. Quetiapine was the most common AP medication used in our hospital; therefore, our findings cannot be generalized to hospitals that utilize other AP agents. Future studies should examine antipsychotic use across hospitals to determine variation in prescribing patterns and outcomes. Nevertheless, the care of these patients were transitioned to a large number of geriatricians and primary care and nursing home physicians after discharge, and the reflected practice patterns extended beyond our hospital. Additionally, we were unable to determine when and why APs were discontinued or started in the outpatient setting. We were only able to detect readmissions to the 3 hospitals within our health system and therefore may have missed some readmissions to other institutions, although the majority of patients in our region tend to return to the same hospital. For patients who were not readmitted, we were also unable to identify whether they remained on the APs initiated during their index hospitalizations. Any retrospective study is limited by the difficulty of distinguishing delirium from the behavioral and psychiatric symptoms of dementia, but we identified delirium using standard terms described in previous literature.[10] We were unable to determine the types of delirium (hyperactive vs hypoactive) given that the documentations on behavioral symptoms were largely missing from the charts. The number of patients with preexisting diagnosis of dementia was likely underestimated, as we were only able to verify the diagnosis from the medical history. Additionally, the retrospective design based on chart review limited the factors that we could detect and grade accurately for inclusion in our mortality prediction model. Of note, our model did not contain objective measures of cognition, agitation, function, and markers for frailty such as walking speed, weak grip strength, weight loss, and low physical activity.

CONCLUSION

Initiating an AP (eg, haloperidol, quetiapine, olanzapine, and risperidone) in the hospital is likely to result in long‐term use of these medications despite the fact that AP use has been associated with multiple risks including falls, fractures, stroke, cardiovascular disease, and increased mortality in those with underlying dementia.[27] When possible, behavioral interventions to prevent delirium and slow the trajectory of decline should be implemented to reduce AP use. If patients with delirium are started on antipsychotics, it is important to monitor for prolonged QTc given the associated risk of mortality. In a subgroup of patients at high risk for death in the upcoming year, occurrence of delirium or use of APs during a hospitalization should both be considered triggers for early advance care planning and possibly palliative care and end‐of‐life discussions, with an emphasis on quality of life.

Disclosures: The research was supported by the Department of Medicine, Baystate Medical Center/Tufts University School of Medicine. Dr. Lagu is supported by the National Heart, Lung, and Blood Institute of the National Institutes of Health under award number K01HL114745. Drs. Lagu and Loh had full access to all of the data in the study. They take responsibility for the integrity of the data and the accuracy of the analysis. Drs. Loh, Brennan, Lindenauer, and Lagu conceived of the study. Drs. Loh, Ramdass, and Ms. Garb acquired the data. Ms Garb analyzed and interpreted the data. Drs. Loh, Ramdass, and Thim drafted the manuscript. Drs. Brennan, Lindenauer, and Lagu and Ms. Garb critically reviewed the manuscript for important intellectual content. Dr. Lagu has received consulting fees from the Institute for Healthcare Improvement, under contract to the Centers for Medicare and Medicaid Services, for her work on a project to help health systems achieve disability competence. Dr. Brennan is supported by a Geriatric Work Force Enhancement Grant from the US Department of Health and Human Services award number 1 U1QHP287020100. The authors report no conflicts of interest.

Delirium, a clinical syndrome characterized by inattention and acute cognitive dysfunction, is very common in older hospitalized patients, with a reported incidence of 18% to 35% at time of admission and overall occurrence rates of 29% to 64%.[1] Previous studies have reported that a diagnosis of delirium is not benign and is associated with other adverse outcomes including prolonged hospitalization, institutionalization, increased cost, and mortality. These outcomes occurred independent of age, prior cognitive functioning, and comorbidities.[2] Guidelines recommend that management of inpatient delirium should be focused on addressing the underlying etiology and managed with nonpharmacological interventions whenever possible.[3, 4, 5] However, implementing these recommendations can prove to be very challenging in hospital settings. Providers frequently have to resort to medical therapies, including antipsychotics (APs). Although these medications are commonly used to treat delirium in elderly patients, there is limited evidence to support their efficacy, and there are currently no proven pharmacological alternatives to these medications.[6] Furthermore, previous studies have demonstrated an increased risk of stroke, infection, cognitive impairment, and mortality in elders with dementia who receive long‐term AP therapy.[7, 8, 9] Yet as many as 48% of hospitalized elders who were newly started on APs had these drugs continued at time of discharge.[10]

There have been few studies describing the long‐term outcomes of elderly patient who are started on APs in the hospital. Most information on outcomes comes from patients with dementia. Therefore, we studied the 1‐year outcomes of a cohort of patients with and without dementia who were started on APs in the hospital and then discharged on these medications. In this cohort, we aimed to describe the number of readmissions, reasons for readmissions, duration of AP therapy, use of other sedating medications such as anxiolytics, hypnotics, and antihistamines as well as the incidence of readmission and death 1 year after the index hospital discharge.

METHODS

We previously described a retrospective cohort of 300 elders (65 years old) admitted to a tertiary care hospital between October 1, 2012 and September 31, 2013 who were newly prescribed APs while hospitalized.[10] Of patients alive at the time of discharge (260), 56% (146 patients) were discharged on APs. Two investigators extracted these 148 patient charts independently to identify and quantify the number of readmissions to the index hospital. We then limited the sample to only the first readmission per patient following the index admission and extracted this readmission for each patient. We first determined if APs were present on the admission medication reconciliation. If APs were not present on admission, we examined whether they were resumed during the hospitalization using the electronic medication administration summary. If they were present on admission, we looked to see if they were discontinued during the readmission and if additional new APs were started during the hospitalizations. We documented the circumstances around APs use and identified patients who died during their hospitalizations. We identified delirium using the same terms that were described in our prior study on the same cohort of patients.[10] We determined if patients were delirious using a predetermined algorithm (Figure 1). Briefly, we first determined delirium was documented. We then examined whether there was a Confusion Assessment Method (CAM) instrument included in the record. If a CAM instrument was not documented, we then looked for documentation using specific terms (eg, disorientation, confusions). We identified patients with dementia by determining whether dementia was documented along with other admission medical comorbidities. If it was not, we determined whether dementia was newly diagnosed during the hospital stay using progress notes or consultation notes. We did not objectively define criteria for diagnosis of dementia. We used the National Death Index (NDI) to determine mortality for all patients 1 year after discharge from the index hospitalization. The NDI is a national database of death records maintained by the National Center for Health Statistics. It has shown consistently high sensitivity and specificity for detection of death.[11]

Figure 1

We used descriptive statistics (means, standard deviations, range, and percents as appropriate to the scale of measurement) to describe the patient sample. We then used multiple logistic regression to identify significant predictors of death within 1 year of discharge.[12] Univariate analysis was used to select candidates for the logistic model (t tests for continuous factors and ² for discrete factors). All factors with a significance level <0.2 on univariate analysis were included in the logistic regression, in addition to age and sex (regardless of significance). A maximum likelihood procedure was used to calculate the regression coefficients for the logistic model. The likelihood ratio criterion was used to determine the significance of individual factors in the regression model.[13] Factors with a significance level of 0.15 or less were retained in the final model, in addition to age and sex.

RESULTS

The 260 patients discharged alive from their index admissions had a 1‐year mortality rate of 29% (75/260). Of the 146/260 patients discharged on APs, 60 (41%) patients experienced at least 1 readmission (mean = 2 readmissions per patient; range, 18, with 111 total readmissions for 60 patients) within 1 year from discharge (Figure 2). Most common diagnoses at the time of readmissions were related neurological and psychiatric disorders (14%), cardiovascular and circulation disorders (13%), renal injury and electrolyte disorders (11%), and infections (6%). Among patients with at least 1 readmission, the mean age was 81.3 (range, 65.599.7), 60% were male, and 45% were admitted from a skilled nursing facility or rehabilitation facility (Table 1). Median time to readmission was 43.5 days (range, 1343 days), and 79% were readmitted to a medical service. The remaining 20% were admitted to a surgical service. Inpatient mortality during first readmissions was 8% (5/60). At the time of first readmission, 39/60 (65%) of patients were still on the same APs on which they had been discharged, and the APs were continued during the hospitalization in 79% of the patients (61% quetiapine, 19% olanzapine, and 13% risperidone). About half of patients whose APs were discontinued prior to readmission received a new AP during their hospital stays (9/20; 45%). One patient had been started on quetiapine in the outpatient setting. No patients were found to have new benzodiazepines, nonbenzodiazepine hypnotic, or antihistamines on their admission medication list.

Figure 2

Eighteen patients received 1 or more new APs during the readmission hospitalizations. These included haloperidol (89%) and quetiapine (39%). Delirium was the main reported indication for starting APs (78%), but in 17% of cases no indication was documented. An electrocardiogram (ECG) was performed in 94% prior to APs administration and for 22% after APs administration. Corrected QT interval (QTc) of >500 ms was present in 18% of patients in pretreatment ECG and 50% of patients in post‐AP ECG. Of patients who survived readmission, 58% (32/55) were discharged to postacute facilities. Of the 39 patients who were on the same APs from index admission, 27 (69%) patients were eventually discharged on the same APs or new APs started during the readmission.

In the multivariable model (Table 2), predictors of death at 1 year included discharge to postacute facilities after index admission (odds ratio [OR]: 2.28; 95% confidence interval [CI]: 1.10‐4.73, P = 0.03) and QTc prolongation >500 ms during index admission (OR: 3.41; 95% CI: 1.34‐8.67, P = 0.01). Age and gender were not associated with 1‐year mortality.

DISCUSSION

In a cohort of elderly patients who were discharged on APs, nearly one‐third (29%) died within 1 year of the hospitalization in which APs were initiated. Nearly half of the survivors from the index admission (41%) experienced at least 1 admission within 1 year from discharge. Of readmitted patients, two‐thirds were taking the same APs that had been started during the index hospitalization. Half of the patients not on APs on readmission were started on an AP during the hospitalization, most often because they became delirious on return to the acute care setting. Compared to patients discharged home after an index admission, patients who were discharged to postacute facilities were almost 4 times as likely to die during the year subsequent to the admission. These data suggest that once patients are started on APs, most are continued on them until the next admission or are restarted during that readmission. Moreover, hospitalized elders who require an AP are at high risk for mortality in the coming year.

Prior studies have reported that patients with delirium have elevated 1‐year mortality rates.[14, 15, 16, 17, 18, 19] A secondary analysis of the Delirium Prevention Trial, which included 437 hospitalized older patients, revealed a 1‐year mortality rate of 20% in those who were never delirious during hospitalization, compared to 26% to 38% in patients with delirium.[19] Additionally, 1‐year mortality in hospitalized older patients with delirium (36%) was shown to be higher than patients with dementia (29%) or depression (26%).[17] Unlike these studies, not all of the patients in our study had documented delirium, but all received an AP. Still, it is notable that the 1‐year mortality rate for delirium in general is similar to what we found in this study.

The literature has also reported that long‐term AP use is associated with excess mortality in elder patients, especially those with dementia.[20, 21, 22] In a retrospective cohort study, older patients with dementia who were taking antipsychotics had significantly higher 1‐year mortality rates (23%29%) than patients not taking antipsychotic medications (15%). In a large Canadian propensity score‐matched cohort study that included over 13,000 demented older adults, the mortality was higher in the community‐dwelling elders who received atypical APs compared to no APs, with a difference of 1.1% in 180‐day mortality rate after initiation of APs.[21] The absolute mortality rate was 2.6% higher in patients who received typical compared to atypical APs. Unlike these studies, not every patient in our cohort had a diagnosis of dementia, but again, mortality rates in these studies appear similar to our cohort.

In contrast, other observational studies have not found an increased risk associated with receipt of APs. For example, a prospective study that enrolled approximately 950 patients with probable dementia showed that AP use was not associated with time to death after adjustment for comorbidities, demographic and cognitive variables.[23] These conflicting results highlight the difficulties of attributing outcomes in high‐risk populations. Although the excess mortality observed in patients taking APs may be related to the risks of APs, it is quite possible that patients who require APs (most often for delirium or agitated dementia) are at higher risk of death. This confounding by indication may be nearly impossible to adjust for retrospectively, even using techniques such as propensity matching.

Our report adds to the literature; we know of no studies to date describing a cohort of patients, most with delirium, who were started on APs in the hospital. We also attempted to identify the reasons that patients were started on APs, which have been infrequently reported. As noted above, our 1‐year mortality rate of 29% among older patients prescribed APs in the hospital was quite similar to mortality rates both for patients with delirium who were not necessarily treated with APs and patients with dementia who were treated with APs. This finding further supports the argument that risk factors for mortality, including dementia, delirium, and AP use are very difficult to tease apart. It is possible that the reasons that APs are prescribed (agitated delirium or dementia) have as much to do with the excess mortality reported in observational studies of APs as the use of APs themselves.

The high rate of continued AP use we observed (two‐thirds of readmitted patients) may reflect limited pharmacological alternatives to these medications with little evidence to support treating the symptoms of delirium with other drug classes, along with suboptimal environmental and behavioral modifications in postacute facilities and hospitals. This is unfortunate given that delirium is often preventable. Systematic implementation of well‐documented strategies to decrease delirium in hospitals and postacute facilities would likely reduce the prescription of APs and has the potential to slow the decline in this vulnerable population. A meta‐analysis incorporating both randomized and nonrandomized trials of medical and surgical patients showed that multicomponent nonpharmacologic interventions decreased delirium by 50%.[24] Thus, simple interventions such as reorientation, early mobilization, optimizing vision and hearing, sleepwake cycle preservation, and hydration might avoid roughly 1 million cases of delirium in hospitalized older adults annually.[24] The Hospital Elder Life Program and Acute Care for Elders units are examples of programs that have been shown to decrease the incidence of delirium.[25, 26]

Despite vigorous efforts to prevent delirium, a subgroup of patients still will become delirious. These patients are at high risk for death. Our mortality prediction model revealed that patients who were discharged to postacute facilities were 4 times more likely to die during the subsequent year compared to patients who were discharged home. Patients discharged to postacute facilities are likely to have a higher burden of disease, greater functional and cognitive impairment, and more frailty than those who are able to return to the community. Very ill and/or frail patients receiving APs in the hospital and requiring APs on discharge to postacute care facilities have limited survival and may benefit from expedited palliative care interventions to clarify prognosis and goals, and relieve suffering. At a minimum, our study identifies a need for further study to identify this very high‐risk group of elders. It is notable that 50% of patients were found to have a post‐treatment ECG with a QTc of >500 ms, a finding that has not been previously described. This would put these patients at higher risk of mortality, and as such we suggest that current guidelines should continue to emphasize the importance of post‐treatment ECGs and set clear criteria for discontinuation in elderly patients.

Our study is limited by its retrospective, single‐center design and small sample size, therefore limiting the interpretation and generalizability of the results to other hospitals. Quetiapine was the most common AP medication used in our hospital; therefore, our findings cannot be generalized to hospitals that utilize other AP agents. Future studies should examine antipsychotic use across hospitals to determine variation in prescribing patterns and outcomes. Nevertheless, the care of these patients were transitioned to a large number of geriatricians and primary care and nursing home physicians after discharge, and the reflected practice patterns extended beyond our hospital. Additionally, we were unable to determine when and why APs were discontinued or started in the outpatient setting. We were only able to detect readmissions to the 3 hospitals within our health system and therefore may have missed some readmissions to other institutions, although the majority of patients in our region tend to return to the same hospital. For patients who were not readmitted, we were also unable to identify whether they remained on the APs initiated during their index hospitalizations. Any retrospective study is limited by the difficulty of distinguishing delirium from the behavioral and psychiatric symptoms of dementia, but we identified delirium using standard terms described in previous literature.[10] We were unable to determine the types of delirium (hyperactive vs hypoactive) given that the documentations on behavioral symptoms were largely missing from the charts. The number of patients with preexisting diagnosis of dementia was likely underestimated, as we were only able to verify the diagnosis from the medical history. Additionally, the retrospective design based on chart review limited the factors that we could detect and grade accurately for inclusion in our mortality prediction model. Of note, our model did not contain objective measures of cognition, agitation, function, and markers for frailty such as walking speed, weak grip strength, weight loss, and low physical activity.

CONCLUSION

Initiating an AP (eg, haloperidol, quetiapine, olanzapine, and risperidone) in the hospital is likely to result in long‐term use of these medications despite the fact that AP use has been associated with multiple risks including falls, fractures, stroke, cardiovascular disease, and increased mortality in those with underlying dementia.[27] When possible, behavioral interventions to prevent delirium and slow the trajectory of decline should be implemented to reduce AP use. If patients with delirium are started on antipsychotics, it is important to monitor for prolonged QTc given the associated risk of mortality. In a subgroup of patients at high risk for death in the upcoming year, occurrence of delirium or use of APs during a hospitalization should both be considered triggers for early advance care planning and possibly palliative care and end‐of‐life discussions, with an emphasis on quality of life.

Disclosures: The research was supported by the Department of Medicine, Baystate Medical Center/Tufts University School of Medicine. Dr. Lagu is supported by the National Heart, Lung, and Blood Institute of the National Institutes of Health under award number K01HL114745. Drs. Lagu and Loh had full access to all of the data in the study. They take responsibility for the integrity of the data and the accuracy of the analysis. Drs. Loh, Brennan, Lindenauer, and Lagu conceived of the study. Drs. Loh, Ramdass, and Ms. Garb acquired the data. Ms Garb analyzed and interpreted the data. Drs. Loh, Ramdass, and Thim drafted the manuscript. Drs. Brennan, Lindenauer, and Lagu and Ms. Garb critically reviewed the manuscript for important intellectual content. Dr. Lagu has received consulting fees from the Institute for Healthcare Improvement, under contract to the Centers for Medicare and Medicaid Services, for her work on a project to help health systems achieve disability competence. Dr. Brennan is supported by a Geriatric Work Force Enhancement Grant from the US Department of Health and Human Services award number 1 U1QHP287020100. The authors report no conflicts of interest.

Delirium, a clinical syndrome characterized by inattention and acute cognitive dysfunction, is very common in older hospitalized patients, with a reported incidence of 18% to 35% at time of admission and overall occurrence rates of 29% to 64%.[1] Previous studies have reported that a diagnosis of delirium is not benign and is associated with other adverse outcomes including prolonged hospitalization, institutionalization, increased cost, and mortality. These outcomes occurred independent of age, prior cognitive functioning, and comorbidities.[2] Guidelines recommend that management of inpatient delirium should be focused on addressing the underlying etiology and managed with nonpharmacological interventions whenever possible.[3, 4, 5] However, implementing these recommendations can prove to be very challenging in hospital settings. Providers frequently have to resort to medical therapies, including antipsychotics (APs). Although these medications are commonly used to treat delirium in elderly patients, there is limited evidence to support their efficacy, and there are currently no proven pharmacological alternatives to these medications.[6] Furthermore, previous studies have demonstrated an increased risk of stroke, infection, cognitive impairment, and mortality in elders with dementia who receive long‐term AP therapy.[7, 8, 9] Yet as many as 48% of hospitalized elders who were newly started on APs had these drugs continued at time of discharge.[10]

There have been few studies describing the long‐term outcomes of elderly patient who are started on APs in the hospital. Most information on outcomes comes from patients with dementia. Therefore, we studied the 1‐year outcomes of a cohort of patients with and without dementia who were started on APs in the hospital and then discharged on these medications. In this cohort, we aimed to describe the number of readmissions, reasons for readmissions, duration of AP therapy, use of other sedating medications such as anxiolytics, hypnotics, and antihistamines as well as the incidence of readmission and death 1 year after the index hospital discharge.

METHODS

We previously described a retrospective cohort of 300 elders (65 years old) admitted to a tertiary care hospital between October 1, 2012 and September 31, 2013 who were newly prescribed APs while hospitalized.[10] Of patients alive at the time of discharge (260), 56% (146 patients) were discharged on APs. Two investigators extracted these 148 patient charts independently to identify and quantify the number of readmissions to the index hospital. We then limited the sample to only the first readmission per patient following the index admission and extracted this readmission for each patient. We first determined if APs were present on the admission medication reconciliation. If APs were not present on admission, we examined whether they were resumed during the hospitalization using the electronic medication administration summary. If they were present on admission, we looked to see if they were discontinued during the readmission and if additional new APs were started during the hospitalizations. We documented the circumstances around APs use and identified patients who died during their hospitalizations. We identified delirium using the same terms that were described in our prior study on the same cohort of patients.[10] We determined if patients were delirious using a predetermined algorithm (Figure 1). Briefly, we first determined delirium was documented. We then examined whether there was a Confusion Assessment Method (CAM) instrument included in the record. If a CAM instrument was not documented, we then looked for documentation using specific terms (eg, disorientation, confusions). We identified patients with dementia by determining whether dementia was documented along with other admission medical comorbidities. If it was not, we determined whether dementia was newly diagnosed during the hospital stay using progress notes or consultation notes. We did not objectively define criteria for diagnosis of dementia. We used the National Death Index (NDI) to determine mortality for all patients 1 year after discharge from the index hospitalization. The NDI is a national database of death records maintained by the National Center for Health Statistics. It has shown consistently high sensitivity and specificity for detection of death.[11]

Figure 1

We used descriptive statistics (means, standard deviations, range, and percents as appropriate to the scale of measurement) to describe the patient sample. We then used multiple logistic regression to identify significant predictors of death within 1 year of discharge.[12] Univariate analysis was used to select candidates for the logistic model (t tests for continuous factors and ² for discrete factors). All factors with a significance level <0.2 on univariate analysis were included in the logistic regression, in addition to age and sex (regardless of significance). A maximum likelihood procedure was used to calculate the regression coefficients for the logistic model. The likelihood ratio criterion was used to determine the significance of individual factors in the regression model.[13] Factors with a significance level of 0.15 or less were retained in the final model, in addition to age and sex.

RESULTS

The 260 patients discharged alive from their index admissions had a 1‐year mortality rate of 29% (75/260). Of the 146/260 patients discharged on APs, 60 (41%) patients experienced at least 1 readmission (mean = 2 readmissions per patient; range, 18, with 111 total readmissions for 60 patients) within 1 year from discharge (Figure 2). Most common diagnoses at the time of readmissions were related neurological and psychiatric disorders (14%), cardiovascular and circulation disorders (13%), renal injury and electrolyte disorders (11%), and infections (6%). Among patients with at least 1 readmission, the mean age was 81.3 (range, 65.599.7), 60% were male, and 45% were admitted from a skilled nursing facility or rehabilitation facility (Table 1). Median time to readmission was 43.5 days (range, 1343 days), and 79% were readmitted to a medical service. The remaining 20% were admitted to a surgical service. Inpatient mortality during first readmissions was 8% (5/60). At the time of first readmission, 39/60 (65%) of patients were still on the same APs on which they had been discharged, and the APs were continued during the hospitalization in 79% of the patients (61% quetiapine, 19% olanzapine, and 13% risperidone). About half of patients whose APs were discontinued prior to readmission received a new AP during their hospital stays (9/20; 45%). One patient had been started on quetiapine in the outpatient setting. No patients were found to have new benzodiazepines, nonbenzodiazepine hypnotic, or antihistamines on their admission medication list.

Figure 2

Eighteen patients received 1 or more new APs during the readmission hospitalizations. These included haloperidol (89%) and quetiapine (39%). Delirium was the main reported indication for starting APs (78%), but in 17% of cases no indication was documented. An electrocardiogram (ECG) was performed in 94% prior to APs administration and for 22% after APs administration. Corrected QT interval (QTc) of >500 ms was present in 18% of patients in pretreatment ECG and 50% of patients in post‐AP ECG. Of patients who survived readmission, 58% (32/55) were discharged to postacute facilities. Of the 39 patients who were on the same APs from index admission, 27 (69%) patients were eventually discharged on the same APs or new APs started during the readmission.

In the multivariable model (Table 2), predictors of death at 1 year included discharge to postacute facilities after index admission (odds ratio [OR]: 2.28; 95% confidence interval [CI]: 1.10‐4.73, P = 0.03) and QTc prolongation >500 ms during index admission (OR: 3.41; 95% CI: 1.34‐8.67, P = 0.01). Age and gender were not associated with 1‐year mortality.

DISCUSSION

In a cohort of elderly patients who were discharged on APs, nearly one‐third (29%) died within 1 year of the hospitalization in which APs were initiated. Nearly half of the survivors from the index admission (41%) experienced at least 1 admission within 1 year from discharge. Of readmitted patients, two‐thirds were taking the same APs that had been started during the index hospitalization. Half of the patients not on APs on readmission were started on an AP during the hospitalization, most often because they became delirious on return to the acute care setting. Compared to patients discharged home after an index admission, patients who were discharged to postacute facilities were almost 4 times as likely to die during the year subsequent to the admission. These data suggest that once patients are started on APs, most are continued on them until the next admission or are restarted during that readmission. Moreover, hospitalized elders who require an AP are at high risk for mortality in the coming year.

Prior studies have reported that patients with delirium have elevated 1‐year mortality rates.[14, 15, 16, 17, 18, 19] A secondary analysis of the Delirium Prevention Trial, which included 437 hospitalized older patients, revealed a 1‐year mortality rate of 20% in those who were never delirious during hospitalization, compared to 26% to 38% in patients with delirium.[19] Additionally, 1‐year mortality in hospitalized older patients with delirium (36%) was shown to be higher than patients with dementia (29%) or depression (26%).[17] Unlike these studies, not all of the patients in our study had documented delirium, but all received an AP. Still, it is notable that the 1‐year mortality rate for delirium in general is similar to what we found in this study.

The literature has also reported that long‐term AP use is associated with excess mortality in elder patients, especially those with dementia.[20, 21, 22] In a retrospective cohort study, older patients with dementia who were taking antipsychotics had significantly higher 1‐year mortality rates (23%29%) than patients not taking antipsychotic medications (15%). In a large Canadian propensity score‐matched cohort study that included over 13,000 demented older adults, the mortality was higher in the community‐dwelling elders who received atypical APs compared to no APs, with a difference of 1.1% in 180‐day mortality rate after initiation of APs.[21] The absolute mortality rate was 2.6% higher in patients who received typical compared to atypical APs. Unlike these studies, not every patient in our cohort had a diagnosis of dementia, but again, mortality rates in these studies appear similar to our cohort.

In contrast, other observational studies have not found an increased risk associated with receipt of APs. For example, a prospective study that enrolled approximately 950 patients with probable dementia showed that AP use was not associated with time to death after adjustment for comorbidities, demographic and cognitive variables.[23] These conflicting results highlight the difficulties of attributing outcomes in high‐risk populations. Although the excess mortality observed in patients taking APs may be related to the risks of APs, it is quite possible that patients who require APs (most often for delirium or agitated dementia) are at higher risk of death. This confounding by indication may be nearly impossible to adjust for retrospectively, even using techniques such as propensity matching.

Our report adds to the literature; we know of no studies to date describing a cohort of patients, most with delirium, who were started on APs in the hospital. We also attempted to identify the reasons that patients were started on APs, which have been infrequently reported. As noted above, our 1‐year mortality rate of 29% among older patients prescribed APs in the hospital was quite similar to mortality rates both for patients with delirium who were not necessarily treated with APs and patients with dementia who were treated with APs. This finding further supports the argument that risk factors for mortality, including dementia, delirium, and AP use are very difficult to tease apart. It is possible that the reasons that APs are prescribed (agitated delirium or dementia) have as much to do with the excess mortality reported in observational studies of APs as the use of APs themselves.

The high rate of continued AP use we observed (two‐thirds of readmitted patients) may reflect limited pharmacological alternatives to these medications with little evidence to support treating the symptoms of delirium with other drug classes, along with suboptimal environmental and behavioral modifications in postacute facilities and hospitals. This is unfortunate given that delirium is often preventable. Systematic implementation of well‐documented strategies to decrease delirium in hospitals and postacute facilities would likely reduce the prescription of APs and has the potential to slow the decline in this vulnerable population. A meta‐analysis incorporating both randomized and nonrandomized trials of medical and surgical patients showed that multicomponent nonpharmacologic interventions decreased delirium by 50%.[24] Thus, simple interventions such as reorientation, early mobilization, optimizing vision and hearing, sleepwake cycle preservation, and hydration might avoid roughly 1 million cases of delirium in hospitalized older adults annually.[24] The Hospital Elder Life Program and Acute Care for Elders units are examples of programs that have been shown to decrease the incidence of delirium.[25, 26]

Despite vigorous efforts to prevent delirium, a subgroup of patients still will become delirious. These patients are at high risk for death. Our mortality prediction model revealed that patients who were discharged to postacute facilities were 4 times more likely to die during the subsequent year compared to patients who were discharged home. Patients discharged to postacute facilities are likely to have a higher burden of disease, greater functional and cognitive impairment, and more frailty than those who are able to return to the community. Very ill and/or frail patients receiving APs in the hospital and requiring APs on discharge to postacute care facilities have limited survival and may benefit from expedited palliative care interventions to clarify prognosis and goals, and relieve suffering. At a minimum, our study identifies a need for further study to identify this very high‐risk group of elders. It is notable that 50% of patients were found to have a post‐treatment ECG with a QTc of >500 ms, a finding that has not been previously described. This would put these patients at higher risk of mortality, and as such we suggest that current guidelines should continue to emphasize the importance of post‐treatment ECGs and set clear criteria for discontinuation in elderly patients.

Our study is limited by its retrospective, single‐center design and small sample size, therefore limiting the interpretation and generalizability of the results to other hospitals. Quetiapine was the most common AP medication used in our hospital; therefore, our findings cannot be generalized to hospitals that utilize other AP agents. Future studies should examine antipsychotic use across hospitals to determine variation in prescribing patterns and outcomes. Nevertheless, the care of these patients were transitioned to a large number of geriatricians and primary care and nursing home physicians after discharge, and the reflected practice patterns extended beyond our hospital. Additionally, we were unable to determine when and why APs were discontinued or started in the outpatient setting. We were only able to detect readmissions to the 3 hospitals within our health system and therefore may have missed some readmissions to other institutions, although the majority of patients in our region tend to return to the same hospital. For patients who were not readmitted, we were also unable to identify whether they remained on the APs initiated during their index hospitalizations. Any retrospective study is limited by the difficulty of distinguishing delirium from the behavioral and psychiatric symptoms of dementia, but we identified delirium using standard terms described in previous literature.[10] We were unable to determine the types of delirium (hyperactive vs hypoactive) given that the documentations on behavioral symptoms were largely missing from the charts. The number of patients with preexisting diagnosis of dementia was likely underestimated, as we were only able to verify the diagnosis from the medical history. Additionally, the retrospective design based on chart review limited the factors that we could detect and grade accurately for inclusion in our mortality prediction model. Of note, our model did not contain objective measures of cognition, agitation, function, and markers for frailty such as walking speed, weak grip strength, weight loss, and low physical activity.

CONCLUSION

Initiating an AP (eg, haloperidol, quetiapine, olanzapine, and risperidone) in the hospital is likely to result in long‐term use of these medications despite the fact that AP use has been associated with multiple risks including falls, fractures, stroke, cardiovascular disease, and increased mortality in those with underlying dementia.[27] When possible, behavioral interventions to prevent delirium and slow the trajectory of decline should be implemented to reduce AP use. If patients with delirium are started on antipsychotics, it is important to monitor for prolonged QTc given the associated risk of mortality. In a subgroup of patients at high risk for death in the upcoming year, occurrence of delirium or use of APs during a hospitalization should both be considered triggers for early advance care planning and possibly palliative care and end‐of‐life discussions, with an emphasis on quality of life.

Disclosures: The research was supported by the Department of Medicine, Baystate Medical Center/Tufts University School of Medicine. Dr. Lagu is supported by the National Heart, Lung, and Blood Institute of the National Institutes of Health under award number K01HL114745. Drs. Lagu and Loh had full access to all of the data in the study. They take responsibility for the integrity of the data and the accuracy of the analysis. Drs. Loh, Brennan, Lindenauer, and Lagu conceived of the study. Drs. Loh, Ramdass, and Ms. Garb acquired the data. Ms Garb analyzed and interpreted the data. Drs. Loh, Ramdass, and Thim drafted the manuscript. Drs. Brennan, Lindenauer, and Lagu and Ms. Garb critically reviewed the manuscript for important intellectual content. Dr. Lagu has received consulting fees from the Institute for Healthcare Improvement, under contract to the Centers for Medicare and Medicaid Services, for her work on a project to help health systems achieve disability competence. Dr. Brennan is supported by a Geriatric Work Force Enhancement Grant from the US Department of Health and Human Services award number 1 U1QHP287020100. The authors report no conflicts of interest.