Original Research

Patients with diabetes currently comprise over 8% of the US population (over 25 million people) and more than 20% of hospitalized patients.[1, 2] Hospitalizations of patients with diabetes account for 23% of total hospital costs in the United States,[2] and patients with diabetes have worse outcomes after hospitalization for a variety of common medical conditions,[3, 4, 5, 6] as well as in intensive care unit (ICU) settings.[7, 8] Individuals with diabetes have historically experienced higher inpatient mortality than individuals without diabetes.[9] However, we recently reported that patients with diabetes at our large academic medical center have experienced a disproportionate reduction in in‐hospital mortality relative to patients without diabetes over the past decade.[10] This surprising trend begs further inquiry.

Improvement in in‐hospital mortality among patients with diabetes may stem from improved inpatient glycemic management. The landmark 2001 study by van den Berghe et al. demonstrating that intensive insulin therapy reduced postsurgical mortality among ICU patients ushered in an era of intensive inpatient glucose control.[11] However, follow‐up multicenter studies have not been able to replicate these results.[12, 13, 14, 15] In non‐ICU and nonsurgical settings, intensive glucose control has not yet been shown to have any mortality benefit, although it may impact other morbidities, such as postoperative infections.[16] Consequently, less stringent glycemic targets are now recommended.[17] Nonetheless, hospitals are being held accountable for certain aspects of inpatient glucose control. For example, the Centers for Medicare & Medicaid Services (CMS) began asking hospitals to report inpatient glucose control in cardiac surgery patients in 2004.[18] This measure is now publicly reported, and as of 2013 is included in the CMS Value‐Based Purchasing Program, which financially penalizes hospitals that do not meet targets.

Outpatient diabetes standards have also evolved in the past decade. The Diabetes Control and Complications Trial in 1993 and the United Kingdom Prospective Diabetes Study in 1997 demonstrated that better glycemic control in type 1 and newly diagnosed type 2 diabetes patients, respectively, improved clinical outcomes, and prompted guidelines for pharmacologic treatment of diabetic patients.[19, 20] However, subsequent randomized clinical trials have failed to establish a clear beneficial effect of intensive glucose control on primary cardiovascular endpoints among higher‐risk patients with longstanding type 2 diabetes,[21, 22, 23] and clinical practice recommendations now accept a more individualized approach to glycemic control.[24] Nonetheless, clinicians are also being held accountable for outpatient glucose control.[25]

To better understand the disproportionate reduction in mortality among hospitalized patients with diabetes that we observed, we first examined whether it was limited to surgical patients or patients in the ICU, the populations that have been demonstrated to benefit from intensive inpatient glucose control. Furthermore, given recent improvements in inpatient and outpatient glycemic control,[26, 27] we examined whether inpatient or outpatient glucose control explained the mortality trends. Results from this study contribute empirical evidence on real‐world effects of efforts to improve inpatient and outpatient glycemic control.

METHODS

RESULTS

We included 322,938 patient admissions. Of this sample, 54,645 (16.9%) had spent at least 1 night in the ICU. Overall, 76,758 patients (23.8%) had diabetes, representing 26.3% of ICU patients, 23.2% of non‐ICU patients, 28.2% of medical patients, and 16.8% of surgical patients (see Table 1 for demographic characteristics).

Mortality Trends Within Strata

Among ICU patients, the overall mortality rate was 9.9%: 10.5% of patients with diabetes and 9.8% of patients without diabetes. Among non‐ICU patients, the overall mortality rate was 0.8%: 0.9% of patients with diabetes and 0.7% of patients without diabetes.

Among medical patients, the overall mortality rate was 2.9%: 3.1% of patients with diabetes and 2.8% of patients without diabetes. Among surgical patients, the overall mortality rate was 1.4%: 1.8% of patients with diabetes and 1.4% of patients without diabetes. Figure 1 shows quarterly in‐hospital mortality for patients with and without diabetes from 2000 to 2010 stratified by ICU status and by service assignment.

Figure 1

Table 2 describes the difference‐in‐differences regression analyses, stratified by ICU status and service assignment. Among ICU patients (Table 2, model 1), each successive year was associated with a 2.6% relative reduction in the adjusted odds of mortality (odds ratio [OR]: 0.974, 95% confidence interval [CI]: 0.963‐0.985) for patients without diabetes compared to a 7.8% relative reduction for those with diabetes (OR: 0.923, 95% CI: 0.906‐0.940). In other words, patients with diabetes compared to patients without diabetes had a significantly greater decline in odds of adjusted mortality of 5.3% per year (OR: 0.947, 95% CI: 0.927‐0.967). As a result, the adjusted odds of mortality among patients with versus without diabetes decreased from 1.352 in 2000 to 0.772 in 2010.

Table 2.Regression Analysis of Mortality Trends

Independent Variables	ICU Patients, N=54,646, OR (95% CI)	Non‐ICU Patients, N=268,293, OR (95% CI)	Medical Patients, N=196,325, OR (95% CI)	Surgical Patients, N=126,614, OR (95% CI)
	Model 1	Model 2	Model 3	Model 4
NOTE: All models control for sex, age at time of admission, race, payer, length of stay in days, principal discharge diagnosis, and Elixhauser comorbidity variables. Models 1 and 2 additionally control for service assignment, whereas models 3 and 4 control for ICU status. Abbreviations: CI, confidence interval; ICU, intensive care unit; OR, odds ratio.
Year	0.974 (0.963‐0.985)	0.925 (0.909‐0.940)	0.943 (0.933‐0.954)	0.995 (0.977‐1.103)
Diabetes	1.352 (1.562‐1.171)	0.958 (0.783‐1.173)	1.186 (1.037‐1.356)	1.213 (0.942‐1.563)
Diabetes*year	0.947 (0.927‐0.967)	0.977 (0.946‐1.008)	0.961 (0.942‐0.980)	0.955 (0.918‐0.994)
C statistic	0.812	0.907	0.880	0.919

Among non‐ICU patients (Table 2, model 2), each successive year was associated with a 7.5% relative reduction in the adjusted odds of mortality (OR: 0.925, 95% CI: 0.909‐0.940) for patients without diabetes compared to a 9.6% relative reduction for those with diabetes (OR: 0.904, 95% CI: 0.879‐0.929); this greater decline in odds of adjusted mortality of 2.3% per year (OR: 0.977, 95% CI: 0.946‐1.008; P=0.148) was not statistically significant.

We found greater decline in odds of mortality among patients with diabetes than among patients without diabetes over time in both medical patients (3.9% greater decline per year; OR: 0.961, 95% CI: 0.942‐0.980) and surgical patients (4.5% greater decline per year; OR: 0.955, 95% CI: 0.918‐0.994), without a difference between the 2. Detailed results are shown in Table 2, models 3 and 4.

Glycemic Control

Among ICU patients with diabetes (N=14,364), at least 2 inpatient point‐of‐care glucose readings were available for 13,136 (91.5%), with a mean of 4.67 readings per day, whereas hemoglobin A1c data were available for only 5321 patients (37.0%). Both inpatient glucose data and hemoglobin A1c were available for 4989 patients (34.7%). Figure 2 shows trends in inpatient and outpatient glycemic control measures among ICU patients with diabetes over the study period. Mean hemoglobin A1c decreased from 7.7 in 2000 to 7.3 in 2010. Mean hospitalization glucose began at 187.2, reached a nadir of 162.4 in the third quarter (Q3) of 2007, and rose subsequently to 174.4 with loosened glucose control targets. Standard deviation of mean glucose and percentage of patient‐days with a severe hyperglycemic episode followed a similar pattern, though with nadirs in Q4 2007 and Q2 2008, respectively, whereas percentage of patient‐days with a hypoglycemic episode rose from 1.46% in 2000, peaked at 3.00% in Q3 2005, and returned to 2.15% in 2010. All changes in glucose control are significant with P<0.001.

Figure 2

DISCUSSION

We conducted a difference‐in‐difference analysis of in‐hospital mortality rates among adult patients with diabetes compared to patients without diabetes over 10 years, stratifying by ICU status and service assignment. For patients with any ICU stay, we found that the reduction in odds of mortality for patients with diabetes has been 3 times larger than the reduction in odds of mortality for patients without diabetes. For those without an ICU stay, we found no significant difference between patients with and without diabetes in the rate at which in‐hospital mortality declined. We did not find stratification by assignment to a medical or surgical service to be an effect modifier. Finally, despite the fact that our institution achieved better aggregate inpatient glucose control, less severe hyperglycemia, and better long‐term glucose control over the course of the decade, we did not find that either inpatient or outpatient glucose control explained the trend in mortality for patients with diabetes in the ICU. Our study is unique in its inclusion of all hospitalized patients and its ability to simultaneously assess whether both inpatient and outpatient glucose control are explanatory factors in the observed mortality trends.

The fact that improved inpatient glucose control did not explain the trend in mortality for patients with diabetes in the ICU is consistent with the majority of the literature on intensive inpatient glucose control. In randomized trials, intensive glucose control appears to be of greater benefit for patients without diabetes than for patients with diabetes.[31] In fact, in 1 study, patients with diabetes were the only group that did not benefit from intensive glucose control.[32] In our study, it is possible that the rise in hypoglycemia nullified some of the benefits of glucose control. Nationally, hospital admissions for hypoglycemia among Medicare beneficiaries now outnumber admissions for hyperglycemia.[27]

We also do not find that the decline in hemoglobin A1c attenuated the reduction in mortality in the minority of patients for whom these data were available. This is concordant with evidence from 3 randomized clinical trials that have failed to establish a clear beneficial effect of intensive outpatient glucose control on primary cardiovascular endpoints among older, high‐risk patients with type 2 diabetes using glucose‐lowering agents.[21, 22, 23] It is notable, however, that the population for whom we had available hemoglobin A1c results was not representative of the overall population of ICU patients with diabetes. Consequently, there may be an association of outpatient glucose control with inpatient mortality in the overall population of ICU patients with diabetes that we were not able to detect.

The decline in mortality among ICU patients with diabetes in our study may stem from factors other than glycemic control. It is possible that patients were diagnosed earlier in their course of disease in later years of the study period, making the population of patients with diabetes younger or healthier. Of note, however, our risk adjustment models were very robust, with C statistics from 0.82 to 0.92, suggesting that we were able to account for much of the mortality risk attributable to patient clinical and demographic factors. More intensive glucose management may have nonglycemic benefits, such as closer patient observation, which may themselves affect mortality. Alternatively, improved cardiovascular management for patients with diabetes may have decreased the incidence of cardiovascular events. During the study period, evidence from large clinical trials demonstrated the importance of tight blood pressure and lipid management in improving outcomes for patients with diabetes,[33, 34, 35, 36] guidelines for lipid management for patients with diabetes changed,[37] and fewer patients developed cardiovascular complications.[38] Finally, it is possible that our findings can be explained by an improvement in treatment of complications for which patients with diabetes previously have had disproportionately worse outcomes, such as percutaneous coronary intervention.[39]

Our findings may have important implications for both clinicians and policymakers. Changes in inpatient glucose management have required substantial additional resources on the part of hospitals. Our evidence regarding the questionable impact of inpatient glucose control on in‐hospital mortality trends for patients with diabetes is disappointing and highlights the need for multifaceted evaluation of the impact of such quality initiatives. There may, for instance, be benefits from tighter blood glucose control in the hospital beyond mortality, such as reduced infections, costs, or length of stay. On the outpatient side, our more limited data are consistent with recent studies that have not been able to show a mortality benefit in older diabetic patients from more stringent glycemic control. A reassessment of prevailing diabetes‐related quality measures, as recently called for by some,[40, 41] seems reasonable.

Our study must be interpreted in light of its limitations. It is possible that the improvements in glucose management were too small to result in a mortality benefit. The overall reduction of 25 mg dL achieved at our institution is less than the 33 to 50 mg/dL difference between intensive and conventional groups in those randomized clinical trials that have found reductions in mortality.[11, 42] In addition, an increase in mean glucose during the last 1 to 2 years of the observation period (in response to prevailing guidelines) could potentially have attenuated any benefit on mortality. The study does not include other important clinical endpoints, such as infections, complications, length of stay, and hospital costs. Additionally, we did not examine postdischarge mortality, which might have shown a different pattern. The small proportion of patients with hemoglobin A1c results may have hampered our ability to detect an effect of outpatient glucose control. Consequently, our findings regarding outpatient glucose control are only suggestive. Finally, our findings represent the experience of a single, large academic medical center and may not be generalizable to all settings.

Overall, we found that patients with diabetes in the ICU have experienced a disproportionate reduction in in‐hospital mortality over time that does not appear to be explained by improvements in either inpatient or outpatient glucose control. Although improved glycemic control may have other benefits, it does not appear to impact in‐hospital mortality. Our real‐world empirical results contribute to the discourse among clinicians and policymakers with regards to refocusing the approach to managing glucose in‐hospital and readjudication of diabetes‐related quality measures.

Patients with diabetes currently comprise over 8% of the US population (over 25 million people) and more than 20% of hospitalized patients.[1, 2] Hospitalizations of patients with diabetes account for 23% of total hospital costs in the United States,[2] and patients with diabetes have worse outcomes after hospitalization for a variety of common medical conditions,[3, 4, 5, 6] as well as in intensive care unit (ICU) settings.[7, 8] Individuals with diabetes have historically experienced higher inpatient mortality than individuals without diabetes.[9] However, we recently reported that patients with diabetes at our large academic medical center have experienced a disproportionate reduction in in‐hospital mortality relative to patients without diabetes over the past decade.[10] This surprising trend begs further inquiry.

Improvement in in‐hospital mortality among patients with diabetes may stem from improved inpatient glycemic management. The landmark 2001 study by van den Berghe et al. demonstrating that intensive insulin therapy reduced postsurgical mortality among ICU patients ushered in an era of intensive inpatient glucose control.[11] However, follow‐up multicenter studies have not been able to replicate these results.[12, 13, 14, 15] In non‐ICU and nonsurgical settings, intensive glucose control has not yet been shown to have any mortality benefit, although it may impact other morbidities, such as postoperative infections.[16] Consequently, less stringent glycemic targets are now recommended.[17] Nonetheless, hospitals are being held accountable for certain aspects of inpatient glucose control. For example, the Centers for Medicare & Medicaid Services (CMS) began asking hospitals to report inpatient glucose control in cardiac surgery patients in 2004.[18] This measure is now publicly reported, and as of 2013 is included in the CMS Value‐Based Purchasing Program, which financially penalizes hospitals that do not meet targets.

Outpatient diabetes standards have also evolved in the past decade. The Diabetes Control and Complications Trial in 1993 and the United Kingdom Prospective Diabetes Study in 1997 demonstrated that better glycemic control in type 1 and newly diagnosed type 2 diabetes patients, respectively, improved clinical outcomes, and prompted guidelines for pharmacologic treatment of diabetic patients.[19, 20] However, subsequent randomized clinical trials have failed to establish a clear beneficial effect of intensive glucose control on primary cardiovascular endpoints among higher‐risk patients with longstanding type 2 diabetes,[21, 22, 23] and clinical practice recommendations now accept a more individualized approach to glycemic control.[24] Nonetheless, clinicians are also being held accountable for outpatient glucose control.[25]

To better understand the disproportionate reduction in mortality among hospitalized patients with diabetes that we observed, we first examined whether it was limited to surgical patients or patients in the ICU, the populations that have been demonstrated to benefit from intensive inpatient glucose control. Furthermore, given recent improvements in inpatient and outpatient glycemic control,[26, 27] we examined whether inpatient or outpatient glucose control explained the mortality trends. Results from this study contribute empirical evidence on real‐world effects of efforts to improve inpatient and outpatient glycemic control.

METHODS

RESULTS

We included 322,938 patient admissions. Of this sample, 54,645 (16.9%) had spent at least 1 night in the ICU. Overall, 76,758 patients (23.8%) had diabetes, representing 26.3% of ICU patients, 23.2% of non‐ICU patients, 28.2% of medical patients, and 16.8% of surgical patients (see Table 1 for demographic characteristics).

Mortality Trends Within Strata

Among ICU patients, the overall mortality rate was 9.9%: 10.5% of patients with diabetes and 9.8% of patients without diabetes. Among non‐ICU patients, the overall mortality rate was 0.8%: 0.9% of patients with diabetes and 0.7% of patients without diabetes.

Among medical patients, the overall mortality rate was 2.9%: 3.1% of patients with diabetes and 2.8% of patients without diabetes. Among surgical patients, the overall mortality rate was 1.4%: 1.8% of patients with diabetes and 1.4% of patients without diabetes. Figure 1 shows quarterly in‐hospital mortality for patients with and without diabetes from 2000 to 2010 stratified by ICU status and by service assignment.

Figure 1

Table 2 describes the difference‐in‐differences regression analyses, stratified by ICU status and service assignment. Among ICU patients (Table 2, model 1), each successive year was associated with a 2.6% relative reduction in the adjusted odds of mortality (odds ratio [OR]: 0.974, 95% confidence interval [CI]: 0.963‐0.985) for patients without diabetes compared to a 7.8% relative reduction for those with diabetes (OR: 0.923, 95% CI: 0.906‐0.940). In other words, patients with diabetes compared to patients without diabetes had a significantly greater decline in odds of adjusted mortality of 5.3% per year (OR: 0.947, 95% CI: 0.927‐0.967). As a result, the adjusted odds of mortality among patients with versus without diabetes decreased from 1.352 in 2000 to 0.772 in 2010.

Table 2.Regression Analysis of Mortality Trends

Independent Variables	ICU Patients, N=54,646, OR (95% CI)	Non‐ICU Patients, N=268,293, OR (95% CI)	Medical Patients, N=196,325, OR (95% CI)	Surgical Patients, N=126,614, OR (95% CI)
	Model 1	Model 2	Model 3	Model 4
NOTE: All models control for sex, age at time of admission, race, payer, length of stay in days, principal discharge diagnosis, and Elixhauser comorbidity variables. Models 1 and 2 additionally control for service assignment, whereas models 3 and 4 control for ICU status. Abbreviations: CI, confidence interval; ICU, intensive care unit; OR, odds ratio.
Year	0.974 (0.963‐0.985)	0.925 (0.909‐0.940)	0.943 (0.933‐0.954)	0.995 (0.977‐1.103)
Diabetes	1.352 (1.562‐1.171)	0.958 (0.783‐1.173)	1.186 (1.037‐1.356)	1.213 (0.942‐1.563)
Diabetes*year	0.947 (0.927‐0.967)	0.977 (0.946‐1.008)	0.961 (0.942‐0.980)	0.955 (0.918‐0.994)
C statistic	0.812	0.907	0.880	0.919

Among non‐ICU patients (Table 2, model 2), each successive year was associated with a 7.5% relative reduction in the adjusted odds of mortality (OR: 0.925, 95% CI: 0.909‐0.940) for patients without diabetes compared to a 9.6% relative reduction for those with diabetes (OR: 0.904, 95% CI: 0.879‐0.929); this greater decline in odds of adjusted mortality of 2.3% per year (OR: 0.977, 95% CI: 0.946‐1.008; P=0.148) was not statistically significant.

We found greater decline in odds of mortality among patients with diabetes than among patients without diabetes over time in both medical patients (3.9% greater decline per year; OR: 0.961, 95% CI: 0.942‐0.980) and surgical patients (4.5% greater decline per year; OR: 0.955, 95% CI: 0.918‐0.994), without a difference between the 2. Detailed results are shown in Table 2, models 3 and 4.

Glycemic Control

Among ICU patients with diabetes (N=14,364), at least 2 inpatient point‐of‐care glucose readings were available for 13,136 (91.5%), with a mean of 4.67 readings per day, whereas hemoglobin A1c data were available for only 5321 patients (37.0%). Both inpatient glucose data and hemoglobin A1c were available for 4989 patients (34.7%). Figure 2 shows trends in inpatient and outpatient glycemic control measures among ICU patients with diabetes over the study period. Mean hemoglobin A1c decreased from 7.7 in 2000 to 7.3 in 2010. Mean hospitalization glucose began at 187.2, reached a nadir of 162.4 in the third quarter (Q3) of 2007, and rose subsequently to 174.4 with loosened glucose control targets. Standard deviation of mean glucose and percentage of patient‐days with a severe hyperglycemic episode followed a similar pattern, though with nadirs in Q4 2007 and Q2 2008, respectively, whereas percentage of patient‐days with a hypoglycemic episode rose from 1.46% in 2000, peaked at 3.00% in Q3 2005, and returned to 2.15% in 2010. All changes in glucose control are significant with P<0.001.

Figure 2

DISCUSSION

We conducted a difference‐in‐difference analysis of in‐hospital mortality rates among adult patients with diabetes compared to patients without diabetes over 10 years, stratifying by ICU status and service assignment. For patients with any ICU stay, we found that the reduction in odds of mortality for patients with diabetes has been 3 times larger than the reduction in odds of mortality for patients without diabetes. For those without an ICU stay, we found no significant difference between patients with and without diabetes in the rate at which in‐hospital mortality declined. We did not find stratification by assignment to a medical or surgical service to be an effect modifier. Finally, despite the fact that our institution achieved better aggregate inpatient glucose control, less severe hyperglycemia, and better long‐term glucose control over the course of the decade, we did not find that either inpatient or outpatient glucose control explained the trend in mortality for patients with diabetes in the ICU. Our study is unique in its inclusion of all hospitalized patients and its ability to simultaneously assess whether both inpatient and outpatient glucose control are explanatory factors in the observed mortality trends.

The fact that improved inpatient glucose control did not explain the trend in mortality for patients with diabetes in the ICU is consistent with the majority of the literature on intensive inpatient glucose control. In randomized trials, intensive glucose control appears to be of greater benefit for patients without diabetes than for patients with diabetes.[31] In fact, in 1 study, patients with diabetes were the only group that did not benefit from intensive glucose control.[32] In our study, it is possible that the rise in hypoglycemia nullified some of the benefits of glucose control. Nationally, hospital admissions for hypoglycemia among Medicare beneficiaries now outnumber admissions for hyperglycemia.[27]

We also do not find that the decline in hemoglobin A1c attenuated the reduction in mortality in the minority of patients for whom these data were available. This is concordant with evidence from 3 randomized clinical trials that have failed to establish a clear beneficial effect of intensive outpatient glucose control on primary cardiovascular endpoints among older, high‐risk patients with type 2 diabetes using glucose‐lowering agents.[21, 22, 23] It is notable, however, that the population for whom we had available hemoglobin A1c results was not representative of the overall population of ICU patients with diabetes. Consequently, there may be an association of outpatient glucose control with inpatient mortality in the overall population of ICU patients with diabetes that we were not able to detect.

The decline in mortality among ICU patients with diabetes in our study may stem from factors other than glycemic control. It is possible that patients were diagnosed earlier in their course of disease in later years of the study period, making the population of patients with diabetes younger or healthier. Of note, however, our risk adjustment models were very robust, with C statistics from 0.82 to 0.92, suggesting that we were able to account for much of the mortality risk attributable to patient clinical and demographic factors. More intensive glucose management may have nonglycemic benefits, such as closer patient observation, which may themselves affect mortality. Alternatively, improved cardiovascular management for patients with diabetes may have decreased the incidence of cardiovascular events. During the study period, evidence from large clinical trials demonstrated the importance of tight blood pressure and lipid management in improving outcomes for patients with diabetes,[33, 34, 35, 36] guidelines for lipid management for patients with diabetes changed,[37] and fewer patients developed cardiovascular complications.[38] Finally, it is possible that our findings can be explained by an improvement in treatment of complications for which patients with diabetes previously have had disproportionately worse outcomes, such as percutaneous coronary intervention.[39]

Our findings may have important implications for both clinicians and policymakers. Changes in inpatient glucose management have required substantial additional resources on the part of hospitals. Our evidence regarding the questionable impact of inpatient glucose control on in‐hospital mortality trends for patients with diabetes is disappointing and highlights the need for multifaceted evaluation of the impact of such quality initiatives. There may, for instance, be benefits from tighter blood glucose control in the hospital beyond mortality, such as reduced infections, costs, or length of stay. On the outpatient side, our more limited data are consistent with recent studies that have not been able to show a mortality benefit in older diabetic patients from more stringent glycemic control. A reassessment of prevailing diabetes‐related quality measures, as recently called for by some,[40, 41] seems reasonable.

Our study must be interpreted in light of its limitations. It is possible that the improvements in glucose management were too small to result in a mortality benefit. The overall reduction of 25 mg dL achieved at our institution is less than the 33 to 50 mg/dL difference between intensive and conventional groups in those randomized clinical trials that have found reductions in mortality.[11, 42] In addition, an increase in mean glucose during the last 1 to 2 years of the observation period (in response to prevailing guidelines) could potentially have attenuated any benefit on mortality. The study does not include other important clinical endpoints, such as infections, complications, length of stay, and hospital costs. Additionally, we did not examine postdischarge mortality, which might have shown a different pattern. The small proportion of patients with hemoglobin A1c results may have hampered our ability to detect an effect of outpatient glucose control. Consequently, our findings regarding outpatient glucose control are only suggestive. Finally, our findings represent the experience of a single, large academic medical center and may not be generalizable to all settings.

Overall, we found that patients with diabetes in the ICU have experienced a disproportionate reduction in in‐hospital mortality over time that does not appear to be explained by improvements in either inpatient or outpatient glucose control. Although improved glycemic control may have other benefits, it does not appear to impact in‐hospital mortality. Our real‐world empirical results contribute to the discourse among clinicians and policymakers with regards to refocusing the approach to managing glucose in‐hospital and readjudication of diabetes‐related quality measures.

Patients with diabetes currently comprise over 8% of the US population (over 25 million people) and more than 20% of hospitalized patients.[1, 2] Hospitalizations of patients with diabetes account for 23% of total hospital costs in the United States,[2] and patients with diabetes have worse outcomes after hospitalization for a variety of common medical conditions,[3, 4, 5, 6] as well as in intensive care unit (ICU) settings.[7, 8] Individuals with diabetes have historically experienced higher inpatient mortality than individuals without diabetes.[9] However, we recently reported that patients with diabetes at our large academic medical center have experienced a disproportionate reduction in in‐hospital mortality relative to patients without diabetes over the past decade.[10] This surprising trend begs further inquiry.

Improvement in in‐hospital mortality among patients with diabetes may stem from improved inpatient glycemic management. The landmark 2001 study by van den Berghe et al. demonstrating that intensive insulin therapy reduced postsurgical mortality among ICU patients ushered in an era of intensive inpatient glucose control.[11] However, follow‐up multicenter studies have not been able to replicate these results.[12, 13, 14, 15] In non‐ICU and nonsurgical settings, intensive glucose control has not yet been shown to have any mortality benefit, although it may impact other morbidities, such as postoperative infections.[16] Consequently, less stringent glycemic targets are now recommended.[17] Nonetheless, hospitals are being held accountable for certain aspects of inpatient glucose control. For example, the Centers for Medicare & Medicaid Services (CMS) began asking hospitals to report inpatient glucose control in cardiac surgery patients in 2004.[18] This measure is now publicly reported, and as of 2013 is included in the CMS Value‐Based Purchasing Program, which financially penalizes hospitals that do not meet targets.

Outpatient diabetes standards have also evolved in the past decade. The Diabetes Control and Complications Trial in 1993 and the United Kingdom Prospective Diabetes Study in 1997 demonstrated that better glycemic control in type 1 and newly diagnosed type 2 diabetes patients, respectively, improved clinical outcomes, and prompted guidelines for pharmacologic treatment of diabetic patients.[19, 20] However, subsequent randomized clinical trials have failed to establish a clear beneficial effect of intensive glucose control on primary cardiovascular endpoints among higher‐risk patients with longstanding type 2 diabetes,[21, 22, 23] and clinical practice recommendations now accept a more individualized approach to glycemic control.[24] Nonetheless, clinicians are also being held accountable for outpatient glucose control.[25]

To better understand the disproportionate reduction in mortality among hospitalized patients with diabetes that we observed, we first examined whether it was limited to surgical patients or patients in the ICU, the populations that have been demonstrated to benefit from intensive inpatient glucose control. Furthermore, given recent improvements in inpatient and outpatient glycemic control,[26, 27] we examined whether inpatient or outpatient glucose control explained the mortality trends. Results from this study contribute empirical evidence on real‐world effects of efforts to improve inpatient and outpatient glycemic control.

METHODS

RESULTS

We included 322,938 patient admissions. Of this sample, 54,645 (16.9%) had spent at least 1 night in the ICU. Overall, 76,758 patients (23.8%) had diabetes, representing 26.3% of ICU patients, 23.2% of non‐ICU patients, 28.2% of medical patients, and 16.8% of surgical patients (see Table 1 for demographic characteristics).

Mortality Trends Within Strata

Among ICU patients, the overall mortality rate was 9.9%: 10.5% of patients with diabetes and 9.8% of patients without diabetes. Among non‐ICU patients, the overall mortality rate was 0.8%: 0.9% of patients with diabetes and 0.7% of patients without diabetes.

Among medical patients, the overall mortality rate was 2.9%: 3.1% of patients with diabetes and 2.8% of patients without diabetes. Among surgical patients, the overall mortality rate was 1.4%: 1.8% of patients with diabetes and 1.4% of patients without diabetes. Figure 1 shows quarterly in‐hospital mortality for patients with and without diabetes from 2000 to 2010 stratified by ICU status and by service assignment.

Figure 1

Table 2 describes the difference‐in‐differences regression analyses, stratified by ICU status and service assignment. Among ICU patients (Table 2, model 1), each successive year was associated with a 2.6% relative reduction in the adjusted odds of mortality (odds ratio [OR]: 0.974, 95% confidence interval [CI]: 0.963‐0.985) for patients without diabetes compared to a 7.8% relative reduction for those with diabetes (OR: 0.923, 95% CI: 0.906‐0.940). In other words, patients with diabetes compared to patients without diabetes had a significantly greater decline in odds of adjusted mortality of 5.3% per year (OR: 0.947, 95% CI: 0.927‐0.967). As a result, the adjusted odds of mortality among patients with versus without diabetes decreased from 1.352 in 2000 to 0.772 in 2010.

Table 2.Regression Analysis of Mortality Trends

Independent Variables	ICU Patients, N=54,646, OR (95% CI)	Non‐ICU Patients, N=268,293, OR (95% CI)	Medical Patients, N=196,325, OR (95% CI)	Surgical Patients, N=126,614, OR (95% CI)
	Model 1	Model 2	Model 3	Model 4
NOTE: All models control for sex, age at time of admission, race, payer, length of stay in days, principal discharge diagnosis, and Elixhauser comorbidity variables. Models 1 and 2 additionally control for service assignment, whereas models 3 and 4 control for ICU status. Abbreviations: CI, confidence interval; ICU, intensive care unit; OR, odds ratio.
Year	0.974 (0.963‐0.985)	0.925 (0.909‐0.940)	0.943 (0.933‐0.954)	0.995 (0.977‐1.103)
Diabetes	1.352 (1.562‐1.171)	0.958 (0.783‐1.173)	1.186 (1.037‐1.356)	1.213 (0.942‐1.563)
Diabetes*year	0.947 (0.927‐0.967)	0.977 (0.946‐1.008)	0.961 (0.942‐0.980)	0.955 (0.918‐0.994)
C statistic	0.812	0.907	0.880	0.919

Among non‐ICU patients (Table 2, model 2), each successive year was associated with a 7.5% relative reduction in the adjusted odds of mortality (OR: 0.925, 95% CI: 0.909‐0.940) for patients without diabetes compared to a 9.6% relative reduction for those with diabetes (OR: 0.904, 95% CI: 0.879‐0.929); this greater decline in odds of adjusted mortality of 2.3% per year (OR: 0.977, 95% CI: 0.946‐1.008; P=0.148) was not statistically significant.

We found greater decline in odds of mortality among patients with diabetes than among patients without diabetes over time in both medical patients (3.9% greater decline per year; OR: 0.961, 95% CI: 0.942‐0.980) and surgical patients (4.5% greater decline per year; OR: 0.955, 95% CI: 0.918‐0.994), without a difference between the 2. Detailed results are shown in Table 2, models 3 and 4.

Glycemic Control

Among ICU patients with diabetes (N=14,364), at least 2 inpatient point‐of‐care glucose readings were available for 13,136 (91.5%), with a mean of 4.67 readings per day, whereas hemoglobin A1c data were available for only 5321 patients (37.0%). Both inpatient glucose data and hemoglobin A1c were available for 4989 patients (34.7%). Figure 2 shows trends in inpatient and outpatient glycemic control measures among ICU patients with diabetes over the study period. Mean hemoglobin A1c decreased from 7.7 in 2000 to 7.3 in 2010. Mean hospitalization glucose began at 187.2, reached a nadir of 162.4 in the third quarter (Q3) of 2007, and rose subsequently to 174.4 with loosened glucose control targets. Standard deviation of mean glucose and percentage of patient‐days with a severe hyperglycemic episode followed a similar pattern, though with nadirs in Q4 2007 and Q2 2008, respectively, whereas percentage of patient‐days with a hypoglycemic episode rose from 1.46% in 2000, peaked at 3.00% in Q3 2005, and returned to 2.15% in 2010. All changes in glucose control are significant with P<0.001.

Figure 2

DISCUSSION

We conducted a difference‐in‐difference analysis of in‐hospital mortality rates among adult patients with diabetes compared to patients without diabetes over 10 years, stratifying by ICU status and service assignment. For patients with any ICU stay, we found that the reduction in odds of mortality for patients with diabetes has been 3 times larger than the reduction in odds of mortality for patients without diabetes. For those without an ICU stay, we found no significant difference between patients with and without diabetes in the rate at which in‐hospital mortality declined. We did not find stratification by assignment to a medical or surgical service to be an effect modifier. Finally, despite the fact that our institution achieved better aggregate inpatient glucose control, less severe hyperglycemia, and better long‐term glucose control over the course of the decade, we did not find that either inpatient or outpatient glucose control explained the trend in mortality for patients with diabetes in the ICU. Our study is unique in its inclusion of all hospitalized patients and its ability to simultaneously assess whether both inpatient and outpatient glucose control are explanatory factors in the observed mortality trends.

The fact that improved inpatient glucose control did not explain the trend in mortality for patients with diabetes in the ICU is consistent with the majority of the literature on intensive inpatient glucose control. In randomized trials, intensive glucose control appears to be of greater benefit for patients without diabetes than for patients with diabetes.[31] In fact, in 1 study, patients with diabetes were the only group that did not benefit from intensive glucose control.[32] In our study, it is possible that the rise in hypoglycemia nullified some of the benefits of glucose control. Nationally, hospital admissions for hypoglycemia among Medicare beneficiaries now outnumber admissions for hyperglycemia.[27]

We also do not find that the decline in hemoglobin A1c attenuated the reduction in mortality in the minority of patients for whom these data were available. This is concordant with evidence from 3 randomized clinical trials that have failed to establish a clear beneficial effect of intensive outpatient glucose control on primary cardiovascular endpoints among older, high‐risk patients with type 2 diabetes using glucose‐lowering agents.[21, 22, 23] It is notable, however, that the population for whom we had available hemoglobin A1c results was not representative of the overall population of ICU patients with diabetes. Consequently, there may be an association of outpatient glucose control with inpatient mortality in the overall population of ICU patients with diabetes that we were not able to detect.

The decline in mortality among ICU patients with diabetes in our study may stem from factors other than glycemic control. It is possible that patients were diagnosed earlier in their course of disease in later years of the study period, making the population of patients with diabetes younger or healthier. Of note, however, our risk adjustment models were very robust, with C statistics from 0.82 to 0.92, suggesting that we were able to account for much of the mortality risk attributable to patient clinical and demographic factors. More intensive glucose management may have nonglycemic benefits, such as closer patient observation, which may themselves affect mortality. Alternatively, improved cardiovascular management for patients with diabetes may have decreased the incidence of cardiovascular events. During the study period, evidence from large clinical trials demonstrated the importance of tight blood pressure and lipid management in improving outcomes for patients with diabetes,[33, 34, 35, 36] guidelines for lipid management for patients with diabetes changed,[37] and fewer patients developed cardiovascular complications.[38] Finally, it is possible that our findings can be explained by an improvement in treatment of complications for which patients with diabetes previously have had disproportionately worse outcomes, such as percutaneous coronary intervention.[39]

Our findings may have important implications for both clinicians and policymakers. Changes in inpatient glucose management have required substantial additional resources on the part of hospitals. Our evidence regarding the questionable impact of inpatient glucose control on in‐hospital mortality trends for patients with diabetes is disappointing and highlights the need for multifaceted evaluation of the impact of such quality initiatives. There may, for instance, be benefits from tighter blood glucose control in the hospital beyond mortality, such as reduced infections, costs, or length of stay. On the outpatient side, our more limited data are consistent with recent studies that have not been able to show a mortality benefit in older diabetic patients from more stringent glycemic control. A reassessment of prevailing diabetes‐related quality measures, as recently called for by some,[40, 41] seems reasonable.

Our study must be interpreted in light of its limitations. It is possible that the improvements in glucose management were too small to result in a mortality benefit. The overall reduction of 25 mg dL achieved at our institution is less than the 33 to 50 mg/dL difference between intensive and conventional groups in those randomized clinical trials that have found reductions in mortality.[11, 42] In addition, an increase in mean glucose during the last 1 to 2 years of the observation period (in response to prevailing guidelines) could potentially have attenuated any benefit on mortality. The study does not include other important clinical endpoints, such as infections, complications, length of stay, and hospital costs. Additionally, we did not examine postdischarge mortality, which might have shown a different pattern. The small proportion of patients with hemoglobin A1c results may have hampered our ability to detect an effect of outpatient glucose control. Consequently, our findings regarding outpatient glucose control are only suggestive. Finally, our findings represent the experience of a single, large academic medical center and may not be generalizable to all settings.

Overall, we found that patients with diabetes in the ICU have experienced a disproportionate reduction in in‐hospital mortality over time that does not appear to be explained by improvements in either inpatient or outpatient glucose control. Although improved glycemic control may have other benefits, it does not appear to impact in‐hospital mortality. Our real‐world empirical results contribute to the discourse among clinicians and policymakers with regards to refocusing the approach to managing glucose in‐hospital and readjudication of diabetes‐related quality measures.

Testing for breast cancer is traditionally offered in outpatient settings, and screening mammography rates have plateaued since 2000.[1] Current data suggest that the mammography utilization gap by race has narrowed; however, disparity remains among low‐income, uninsured, and underinsured populations.[2, 3] The lowest compliance with screening mammography recommendations have been reported among women with low income (63.2%), uninsured (50.4%), and those without a usual source of healthcare (43.6%).[4] Although socioeconomic status, access to the healthcare system, and awareness about screening benefits can all influence women's willingness to have screening, the most common reason that women report for not having mammograms were that no one recommended the test.[5, 6] These findings support previous reports that physicians' recommendations about the need for screening mammography is an influential factor in determining women's decisions related to compliance.[7] Hence, the role of healthcare providers in all clinical care settings is pivotal in reducing mammography utilization disparities.

A recent study evaluating the breast cancer screening adherence among the hospitalized women aged 50 to 75 years noted that many (60%) were low income (annual household income <$20,000), 39% were nonadherent, and 35% were at high risk of developing breast cancer.[8] Further, a majority of these hospitalized women were amenable to inpatient screening mammography if due and offered during the hospital stay.[8] As a follow‐up, the purpose of the current study was to explore how hospitalists feel about getting involved in breast cancer screening and ordering screening mammograms for hospitalized women. We hypothesized that a greater proportion of hospitalists would order mammography for hospitalized women who were both overdue for screening and at high risk for developing breast cancer if they fundamentally believe that they have a role in breast cancer screening. This study also explored anticipated barriers that may be of concern to hospitalists when ordering inpatient screening mammography.

METHODS

RESULTS

Out of 106 study subjects willing to participate, 8 did not respond, yielding a response rate of 92%. The mean age of the study participants was 37.6 years, and 55% were female. Almost two‐thirds of study participants (59%) were faculty physicians at an academic hospital, and the average clinical experience as a hospitalist was 4.6 years. Study participants were diverse with respect to ethnicity, and only 30% reported having a family member with breast cancer (Table 1). Because breast cancer is a disease that affects primarily women, stratified analysis by gender showed that most of these characteristic were similar across genders, except fewer women were full time (76% vs 93%, P=0.04) and on the faculty (44% vs 77%, P=0.003).

Only 38% believed that hospitalists should be involved with breast cancer screening. The most commonly cited concern related to ordering an inpatient screening mammography was follow‐up of the results of the mammography, followed by the test may not be covered by patient's insurance. As shown in Table 2, these concerns were not perceived differently among providers who believed that hospitalists should be involved in breast cancer screening as compared to those who do not. Demographic variables from Table 1 failed to discern any significant associations related to believing that hospitalists should be involved with breast cancer screening or with concerns about the barriers to screening presented in Table 2 (data not shown). As shown in Table 2, overall, 32% hospitalists were willing to order a screening mammography during a hospital stay for the scenario of the woman at high risk for developing breast cancer (5‐year risk prediction using Gail model 2.1%) and 33% for the low‐risk scenario (5‐year risk prediction using Gail model 0.6%).

DISCUSSION

Our study suggests that most hospitalists do not believe that they should be involved in breast cancer screening for their hospitalized patients. This perspective was not influenced by either the physician gender, family history for breast cancer, or by the patient's level of risk for developing breast cancer. When patients are in the hospital, both the setting and the acute illness are known to promote reflection and consideration of self‐care.[10] With major healthcare system changes on the horizon and the passing of the Affordable Care Act, we are becoming teams of providers who are collectively responsible for optimal care delivery. It may be possible to increase breast cancer screening rates by educating our patients and offering inpatient screening mammography while they are in the hospital, particularly to those who are at high risk of developing breast cancer.

Physician recommendations for preventive health and screening have consistently been found to be among the strongest predictors of screening utilization.[11] This is the first study to our knowledge that has attempted to understand hospitalists' views and concerns about ordering screening tests to detect occult malignancy. Although addressing preventive care during a hospitalization may seem complex and difficult, helping these women understand their personal risk profile (eg, family history of breast cancer, use of estrogen, race, age, and genetic risk factors) may be what is needed for beginning to influence perspective that might ultimately translate into a willingness to undergo screening.[12, 13, 14] Such delivery of patient‐centered care is built on a foundation of shared decision‐making, which takes into account the patient's preferences, values, and wishes.[15]

Ordering screening mammography for hospitalized patients will require a deeper understanding of hospitalists' attitudes, because the way that these physicians feel about the tests utility will dramatically influence the way that this opportunity is presented to patients, and ultimately the patients' preference to have or forego testing. Our study results are consistent with another publication that highlighted incongruence between physicians' views and patients' preferences for screening practices.[8, 11] Concerns cited, such as interference with patient's acute care, deserve attention, because it may be possible to carry out the screening in ways and at times that do not interfere with treatment or prolong length of stay. Exploring this with a feasibility study will be necessary. Such an approach has been advocated by Trimble et al. for inpatient cervical cancer screening as an efficient strategy to target high‐risk, nonadherent women.[16]

The inpatient setting allows for the elimination of major barriers to screening (like transportation and remembering to get to screening appointments),[8] thereby actively facilitating this needed service. Costs associated with inpatient screening mammography may deter both hospitalists and patients from screening; however, some insurers and Medicare pay for the full cost of screening tests, irrespective of the clinical setting.[17] Further, as hospitals or accountable care organizations become responsible for total cost per beneficiary, screening costs will be preferable when compared with the expenses associated with later detection of pathology and caring for advanced disease states.

One might question whether the mortality benefit of screening mammography is comparable among hospitalized women (who are theoretically sicker and with shorter life expectancy) and those cared for in outpatient practices. Unfortunately, we do not yet know the answer to this question, because data for inpatient screening mammography are nonexistent, and currently this is not considered as a standard of care. However, one can expect the benefits to be similar, if not greater, when performed in the outpatient setting, if preliminary efforts are directed at those who are both nonadherent and at high risk for breast cancer. According to 1 study, increasing mammography utilization by 5% in our country would prevent 560 deaths from breast cancer each year.[18]

Several limitations of this study should be considered. First, this cross‐sectional study was conducted at hospitals associated with a single institution and the results may not be generalizable. Second, although physicians' concerns were explored in this study, we did not solicit input about the potential impact of prevention and screening on the nursing staff. Third, there may be concerns about the hypothetical nature of anchoring and possible framing effects with the 2 clinical scenarios. Finally, it is possible that the hospitalists' response may have been subject to social desirability bias. That said, the response to the key question Do you think hospitalists should be involved in breast cancer screening? do not support a socially desirable bias.

Given the current policy emphasis on reducing disparities in cancer screening, it may be reasonable to expand the role of all healthcare providers and healthcare facilities in screening high‐risk populations. Screening tests that may seem difficult to coordinate in hospitals currently may become easier as our hospitals evolve to become more patient centered. Future studies are needed to evaluate the feasibility and potential barriers to inpatient screening mammography.

Testing for breast cancer is traditionally offered in outpatient settings, and screening mammography rates have plateaued since 2000.[1] Current data suggest that the mammography utilization gap by race has narrowed; however, disparity remains among low‐income, uninsured, and underinsured populations.[2, 3] The lowest compliance with screening mammography recommendations have been reported among women with low income (63.2%), uninsured (50.4%), and those without a usual source of healthcare (43.6%).[4] Although socioeconomic status, access to the healthcare system, and awareness about screening benefits can all influence women's willingness to have screening, the most common reason that women report for not having mammograms were that no one recommended the test.[5, 6] These findings support previous reports that physicians' recommendations about the need for screening mammography is an influential factor in determining women's decisions related to compliance.[7] Hence, the role of healthcare providers in all clinical care settings is pivotal in reducing mammography utilization disparities.

A recent study evaluating the breast cancer screening adherence among the hospitalized women aged 50 to 75 years noted that many (60%) were low income (annual household income <$20,000), 39% were nonadherent, and 35% were at high risk of developing breast cancer.[8] Further, a majority of these hospitalized women were amenable to inpatient screening mammography if due and offered during the hospital stay.[8] As a follow‐up, the purpose of the current study was to explore how hospitalists feel about getting involved in breast cancer screening and ordering screening mammograms for hospitalized women. We hypothesized that a greater proportion of hospitalists would order mammography for hospitalized women who were both overdue for screening and at high risk for developing breast cancer if they fundamentally believe that they have a role in breast cancer screening. This study also explored anticipated barriers that may be of concern to hospitalists when ordering inpatient screening mammography.

METHODS

RESULTS

Out of 106 study subjects willing to participate, 8 did not respond, yielding a response rate of 92%. The mean age of the study participants was 37.6 years, and 55% were female. Almost two‐thirds of study participants (59%) were faculty physicians at an academic hospital, and the average clinical experience as a hospitalist was 4.6 years. Study participants were diverse with respect to ethnicity, and only 30% reported having a family member with breast cancer (Table 1). Because breast cancer is a disease that affects primarily women, stratified analysis by gender showed that most of these characteristic were similar across genders, except fewer women were full time (76% vs 93%, P=0.04) and on the faculty (44% vs 77%, P=0.003).

Only 38% believed that hospitalists should be involved with breast cancer screening. The most commonly cited concern related to ordering an inpatient screening mammography was follow‐up of the results of the mammography, followed by the test may not be covered by patient's insurance. As shown in Table 2, these concerns were not perceived differently among providers who believed that hospitalists should be involved in breast cancer screening as compared to those who do not. Demographic variables from Table 1 failed to discern any significant associations related to believing that hospitalists should be involved with breast cancer screening or with concerns about the barriers to screening presented in Table 2 (data not shown). As shown in Table 2, overall, 32% hospitalists were willing to order a screening mammography during a hospital stay for the scenario of the woman at high risk for developing breast cancer (5‐year risk prediction using Gail model 2.1%) and 33% for the low‐risk scenario (5‐year risk prediction using Gail model 0.6%).

DISCUSSION

Our study suggests that most hospitalists do not believe that they should be involved in breast cancer screening for their hospitalized patients. This perspective was not influenced by either the physician gender, family history for breast cancer, or by the patient's level of risk for developing breast cancer. When patients are in the hospital, both the setting and the acute illness are known to promote reflection and consideration of self‐care.[10] With major healthcare system changes on the horizon and the passing of the Affordable Care Act, we are becoming teams of providers who are collectively responsible for optimal care delivery. It may be possible to increase breast cancer screening rates by educating our patients and offering inpatient screening mammography while they are in the hospital, particularly to those who are at high risk of developing breast cancer.

Physician recommendations for preventive health and screening have consistently been found to be among the strongest predictors of screening utilization.[11] This is the first study to our knowledge that has attempted to understand hospitalists' views and concerns about ordering screening tests to detect occult malignancy. Although addressing preventive care during a hospitalization may seem complex and difficult, helping these women understand their personal risk profile (eg, family history of breast cancer, use of estrogen, race, age, and genetic risk factors) may be what is needed for beginning to influence perspective that might ultimately translate into a willingness to undergo screening.[12, 13, 14] Such delivery of patient‐centered care is built on a foundation of shared decision‐making, which takes into account the patient's preferences, values, and wishes.[15]

Ordering screening mammography for hospitalized patients will require a deeper understanding of hospitalists' attitudes, because the way that these physicians feel about the tests utility will dramatically influence the way that this opportunity is presented to patients, and ultimately the patients' preference to have or forego testing. Our study results are consistent with another publication that highlighted incongruence between physicians' views and patients' preferences for screening practices.[8, 11] Concerns cited, such as interference with patient's acute care, deserve attention, because it may be possible to carry out the screening in ways and at times that do not interfere with treatment or prolong length of stay. Exploring this with a feasibility study will be necessary. Such an approach has been advocated by Trimble et al. for inpatient cervical cancer screening as an efficient strategy to target high‐risk, nonadherent women.[16]

The inpatient setting allows for the elimination of major barriers to screening (like transportation and remembering to get to screening appointments),[8] thereby actively facilitating this needed service. Costs associated with inpatient screening mammography may deter both hospitalists and patients from screening; however, some insurers and Medicare pay for the full cost of screening tests, irrespective of the clinical setting.[17] Further, as hospitals or accountable care organizations become responsible for total cost per beneficiary, screening costs will be preferable when compared with the expenses associated with later detection of pathology and caring for advanced disease states.

One might question whether the mortality benefit of screening mammography is comparable among hospitalized women (who are theoretically sicker and with shorter life expectancy) and those cared for in outpatient practices. Unfortunately, we do not yet know the answer to this question, because data for inpatient screening mammography are nonexistent, and currently this is not considered as a standard of care. However, one can expect the benefits to be similar, if not greater, when performed in the outpatient setting, if preliminary efforts are directed at those who are both nonadherent and at high risk for breast cancer. According to 1 study, increasing mammography utilization by 5% in our country would prevent 560 deaths from breast cancer each year.[18]

Several limitations of this study should be considered. First, this cross‐sectional study was conducted at hospitals associated with a single institution and the results may not be generalizable. Second, although physicians' concerns were explored in this study, we did not solicit input about the potential impact of prevention and screening on the nursing staff. Third, there may be concerns about the hypothetical nature of anchoring and possible framing effects with the 2 clinical scenarios. Finally, it is possible that the hospitalists' response may have been subject to social desirability bias. That said, the response to the key question Do you think hospitalists should be involved in breast cancer screening? do not support a socially desirable bias.

Given the current policy emphasis on reducing disparities in cancer screening, it may be reasonable to expand the role of all healthcare providers and healthcare facilities in screening high‐risk populations. Screening tests that may seem difficult to coordinate in hospitals currently may become easier as our hospitals evolve to become more patient centered. Future studies are needed to evaluate the feasibility and potential barriers to inpatient screening mammography.

Testing for breast cancer is traditionally offered in outpatient settings, and screening mammography rates have plateaued since 2000.[1] Current data suggest that the mammography utilization gap by race has narrowed; however, disparity remains among low‐income, uninsured, and underinsured populations.[2, 3] The lowest compliance with screening mammography recommendations have been reported among women with low income (63.2%), uninsured (50.4%), and those without a usual source of healthcare (43.6%).[4] Although socioeconomic status, access to the healthcare system, and awareness about screening benefits can all influence women's willingness to have screening, the most common reason that women report for not having mammograms were that no one recommended the test.[5, 6] These findings support previous reports that physicians' recommendations about the need for screening mammography is an influential factor in determining women's decisions related to compliance.[7] Hence, the role of healthcare providers in all clinical care settings is pivotal in reducing mammography utilization disparities.

A recent study evaluating the breast cancer screening adherence among the hospitalized women aged 50 to 75 years noted that many (60%) were low income (annual household income <$20,000), 39% were nonadherent, and 35% were at high risk of developing breast cancer.[8] Further, a majority of these hospitalized women were amenable to inpatient screening mammography if due and offered during the hospital stay.[8] As a follow‐up, the purpose of the current study was to explore how hospitalists feel about getting involved in breast cancer screening and ordering screening mammograms for hospitalized women. We hypothesized that a greater proportion of hospitalists would order mammography for hospitalized women who were both overdue for screening and at high risk for developing breast cancer if they fundamentally believe that they have a role in breast cancer screening. This study also explored anticipated barriers that may be of concern to hospitalists when ordering inpatient screening mammography.

METHODS

RESULTS

Out of 106 study subjects willing to participate, 8 did not respond, yielding a response rate of 92%. The mean age of the study participants was 37.6 years, and 55% were female. Almost two‐thirds of study participants (59%) were faculty physicians at an academic hospital, and the average clinical experience as a hospitalist was 4.6 years. Study participants were diverse with respect to ethnicity, and only 30% reported having a family member with breast cancer (Table 1). Because breast cancer is a disease that affects primarily women, stratified analysis by gender showed that most of these characteristic were similar across genders, except fewer women were full time (76% vs 93%, P=0.04) and on the faculty (44% vs 77%, P=0.003).

Only 38% believed that hospitalists should be involved with breast cancer screening. The most commonly cited concern related to ordering an inpatient screening mammography was follow‐up of the results of the mammography, followed by the test may not be covered by patient's insurance. As shown in Table 2, these concerns were not perceived differently among providers who believed that hospitalists should be involved in breast cancer screening as compared to those who do not. Demographic variables from Table 1 failed to discern any significant associations related to believing that hospitalists should be involved with breast cancer screening or with concerns about the barriers to screening presented in Table 2 (data not shown). As shown in Table 2, overall, 32% hospitalists were willing to order a screening mammography during a hospital stay for the scenario of the woman at high risk for developing breast cancer (5‐year risk prediction using Gail model 2.1%) and 33% for the low‐risk scenario (5‐year risk prediction using Gail model 0.6%).

DISCUSSION

Our study suggests that most hospitalists do not believe that they should be involved in breast cancer screening for their hospitalized patients. This perspective was not influenced by either the physician gender, family history for breast cancer, or by the patient's level of risk for developing breast cancer. When patients are in the hospital, both the setting and the acute illness are known to promote reflection and consideration of self‐care.[10] With major healthcare system changes on the horizon and the passing of the Affordable Care Act, we are becoming teams of providers who are collectively responsible for optimal care delivery. It may be possible to increase breast cancer screening rates by educating our patients and offering inpatient screening mammography while they are in the hospital, particularly to those who are at high risk of developing breast cancer.

Physician recommendations for preventive health and screening have consistently been found to be among the strongest predictors of screening utilization.[11] This is the first study to our knowledge that has attempted to understand hospitalists' views and concerns about ordering screening tests to detect occult malignancy. Although addressing preventive care during a hospitalization may seem complex and difficult, helping these women understand their personal risk profile (eg, family history of breast cancer, use of estrogen, race, age, and genetic risk factors) may be what is needed for beginning to influence perspective that might ultimately translate into a willingness to undergo screening.[12, 13, 14] Such delivery of patient‐centered care is built on a foundation of shared decision‐making, which takes into account the patient's preferences, values, and wishes.[15]

Ordering screening mammography for hospitalized patients will require a deeper understanding of hospitalists' attitudes, because the way that these physicians feel about the tests utility will dramatically influence the way that this opportunity is presented to patients, and ultimately the patients' preference to have or forego testing. Our study results are consistent with another publication that highlighted incongruence between physicians' views and patients' preferences for screening practices.[8, 11] Concerns cited, such as interference with patient's acute care, deserve attention, because it may be possible to carry out the screening in ways and at times that do not interfere with treatment or prolong length of stay. Exploring this with a feasibility study will be necessary. Such an approach has been advocated by Trimble et al. for inpatient cervical cancer screening as an efficient strategy to target high‐risk, nonadherent women.[16]

The inpatient setting allows for the elimination of major barriers to screening (like transportation and remembering to get to screening appointments),[8] thereby actively facilitating this needed service. Costs associated with inpatient screening mammography may deter both hospitalists and patients from screening; however, some insurers and Medicare pay for the full cost of screening tests, irrespective of the clinical setting.[17] Further, as hospitals or accountable care organizations become responsible for total cost per beneficiary, screening costs will be preferable when compared with the expenses associated with later detection of pathology and caring for advanced disease states.

One might question whether the mortality benefit of screening mammography is comparable among hospitalized women (who are theoretically sicker and with shorter life expectancy) and those cared for in outpatient practices. Unfortunately, we do not yet know the answer to this question, because data for inpatient screening mammography are nonexistent, and currently this is not considered as a standard of care. However, one can expect the benefits to be similar, if not greater, when performed in the outpatient setting, if preliminary efforts are directed at those who are both nonadherent and at high risk for breast cancer. According to 1 study, increasing mammography utilization by 5% in our country would prevent 560 deaths from breast cancer each year.[18]

Several limitations of this study should be considered. First, this cross‐sectional study was conducted at hospitals associated with a single institution and the results may not be generalizable. Second, although physicians' concerns were explored in this study, we did not solicit input about the potential impact of prevention and screening on the nursing staff. Third, there may be concerns about the hypothetical nature of anchoring and possible framing effects with the 2 clinical scenarios. Finally, it is possible that the hospitalists' response may have been subject to social desirability bias. That said, the response to the key question Do you think hospitalists should be involved in breast cancer screening? do not support a socially desirable bias.

Given the current policy emphasis on reducing disparities in cancer screening, it may be reasonable to expand the role of all healthcare providers and healthcare facilities in screening high‐risk populations. Screening tests that may seem difficult to coordinate in hospitals currently may become easier as our hospitals evolve to become more patient centered. Future studies are needed to evaluate the feasibility and potential barriers to inpatient screening mammography.

Although bronchiolitis is the leading cause of hospitalization for US infants,[1] there is a lack of basic prospective data about the expected inpatient clinical course and ongoing uncertainty about when a hospitalized child is ready for discharge to home.[2] This lack of data about children's readiness for discharge may result in variable hospital length‐of‐stay (LOS).[3, 4, 5]

One specific source of variability in discharge readiness and LOS variability may be the lack of consensus about safe threshold oxygen saturation values for discharge in children hospitalized with bronchiolitis.[6, 7] In 2006, the Scottish Intercollegiate Guidelines Network recommended a discharge room air oxygen (RAO2) saturation threshold of 95%.[8] The same year, the American Academy of Pediatrics (AAP) bronchiolitis clinical practice guideline stated that oxygen is not needed for children with RAO2 saturations 90% who are feeding well and have minimal respiratory distress.[9] There is a need for prospective studies to help clinicians make evidenced‐based discharge decisions for this common condition.

We performed a prospective, multicenter, multiyear study[10, 11, 12] to examine the typical inpatient clinical course of and to develop hospital discharge guidelines for children age <2 years hospitalized with bronchiolitis. We hypothesized that children would not worsen clinically and would be safe to discharge home once their respiratory status improved and they were able to remain hydrated.

METHODS

RESULTS

There were 1916 children hospitalized with bronchiolitis with data on all factors used to define clinical improvement and clinical worsening. The median number of days from the beginning of difficulty breathing until admission was 2 days (IQR, 15.5 days; range, 18 days) and from the beginning of difficulty breathing until clinical improvement was 4 days (IQR, 37.5 days; range, 133 days) (Figure 1). The variance for days to admission was significantly less than the variance for days to clinical improvement (P<0.001).

Figure 1

In this observational study, clinicians discharged 214 (11%) of the 1916 children before meeting the definition of clinical improvement. Thus, 1702 (89%; 95% CI: 87%‐90%) children reached the clinical improvement criteria, had a LOS >1 day, and had data on all factors (Figure 2).

Figure 2

Of the 1702 children who met the clinical improvement criteria, there were 76 children (4%; 95% CI: 3%5%) who worsened (Figure 2). The worsening occurred within a median of 1 day (IQR, 13 days) of clinical improvement. Forty‐six (3%) of the children required transfer to the ICU (1 required intubation, 1 required continuous positive airway pressure, and 4 had apnea), 23 (1%) required oxygen, and 17 (1%) required IV fluids. Eight percent of children met multiple criteria for worsening. A comparison between children who did and did not worsen is shown in Table 1. In general, children who worsened after improvement were younger and born earlier. These children also presented in more severe respiratory distress, had moderate or severe retractions, oxygen saturation <85% at hospitalization, inadequate oral intake, and apnea documented during the hospitalization. Neither viral etiology nor site of care influenced whether the children worsened after improving. However, stratified analysis of children based on initial location of admission (ie, ICU or ward) showed that among the children admitted to the ICU from the emergency department (ED), 89% met the improvement criteria and 19% clinically worsened. In contrast, among children admitted to the ward from the ED, 89% met the improvement criteria, and only 2% clinically worsened. Stratified multivariable models based on the initial location of admission from the ED (ie, ICU or ward) were not possible due to small sample sizes after stratification. None of these children had relapse events requiring rehospitalization within either 24 hours or 7 days of discharge.

On multivariable analysis (Table 2), independent risk factors for worsening after reaching the clinical improvement criteria were young age, preterm birth, and presenting to care with more severe bronchiolitis represented by severe retractions, inadequate oral intake, or apnea. To further evaluate the improvement criteria in the current analysis, multiple sensitivity analyses were conducted. The frequency of clinical worsening after reaching the improvement criteria was stable when we examined different RA02 criteria in sensitivity analyses: (1) excluding RA02 as a criterion for improvement: 90% met improvement criteria and 4% experienced clinical worsening, (2) changing the average RA02 threshold for clinical improvement to 94%: 62% met improvement criteria and 6% experienced clinical worsening, and (3) changing the average RA02 threshold for clinical improvement to 95%: 47% met improvement criteria and 5% experienced clinical worsening. Furthermore, stratifying by age <2 months and restricting to more stringent definitions of bronchiolitis (ie, age <1 year or age <1 year+no history of wheezing) also did not materially change the results (see Supporting Figure 1 in the online version of this article).

We compared the 214 children who were discharged prior to reaching clinical improvement with the 1702 children who reached the clinical improvement criteria. The 214 children were less likely to be age <2 months (22% vs 30%; P=0.02). These 2 groups (214 vs 1702) were similar with respect to severe retractions (2% vs 4%; P=0.13), median respiratory rate (48 vs 48; P=0.42), oxygen saturation <90% (15% vs 11%; P=0.07), inadequate oral intake (50% vs 43%; P=0.13), and rates of relapse events requiring rehospitalization within both 24 hours (0.6% vs 0.6%; P=0.88) and 7 days (1% vs 1%; P=0.90) of discharge.

DISCUSSION

In this large, multicenter, multiyear study of children hospitalized with bronchiolitis, we found that children present to a hospital in a relatively narrow time frame, but their time to recovery in the hospital is highly variable. Nonetheless, 96% of children continued to improve once they had: (1) improving or stable retractions rated as none/mild, (2) a decreasing or stable RR by age, (3) estimated average RAO2 saturation 90% and lowest RAO2 saturation of 88%, and (4) were hydrated. The 4% of children who worsened after clinically improving were more likely to be age <2 months, born <37 weeks, and present with more severe distress (ie, severe retractions, inadequate oral intake, or apnea). Based on the low risk of worsening after clinical improvement, especially among children admitted to the regular ward (2%), we believe these 4 clinical criteria could be used as discharge criteria for this common pediatric illness with a predominantly monophasic clinical course.

Variability in hospital LOS for children with bronchiolitis exists in the United States[3] and internationally.[4, 5] Cheung and colleagues analyzed administrative data from over 75,000 children admitted for bronchiolitis in England between April 2007 and March 2010 and found sixfold variation in LOS between sites. They concluded that this LOS variability was due in part to providers' clinical decision making.[5] Srivastava and colleagues[23] addressed variable clinician decision making in bronchiolitis and 10 other common pediatric conditions by embedding discharge criteria developed by expert consensus into admission order sets. They found that for children with bronchiolitis, the embedded discharge criteria reduced the median LOS from 1.91 to 1.87 days. In contrast to the single‐center data presented by White and colleagues,[24] the prospective, multicenter MARC‐30 data provide a clear understanding of the normal clinical course for children hospitalized with bronchiolitis, determine if children clinically worsen after clinical improvement, and provide data about discharge criteria for children hospitalized with bronchiolitis. Although there is a lack of rigorous published data, the lower tract symptoms of bronchiolitis (eg, cough, retractions) are said to peak on days 5 to 7 of illness and then gradually resolve.[25] In the present study, we found that the time from the onset of difficulty breathing until hospital admission is less variable than the time from the onset of difficulty breathing until either clinical improvement or discharge. Although 75% of children have clinically improved within 7.5 days of difficulty breathing based on the IQR results, the remaining 25% may have a more prolonged recovery in the hospital of up to 3 weeks. Interestingly, prolonged recovery times from bronchiolitis have also been noted in children presenting to the ED[26] and in an outpatient population.[27] It is unclear why 20% to 25% of children at different levels of severity of illness have prolonged recovery from bronchiolitis, but this group of children requires further investigation.

Given the variability of recovery times, clinicians may have difficulty knowing when a child is ready for hospital discharge. One of the main stumbling blocks for discharge readiness in children with bronchiolitis is the interpretation of the oxygen saturation value.[6, 8, 9, 20, 28] However, it should be considered that interpreting the oxygen saturation in a child who is clinically improving in the hospital setting is different than interpreting the oxygen saturation of a child in the ED or the clinic whose clinical course is less certain.[22] In the hospital setting, using the oxygen saturation value in in the AAP guideline,[9] 4% of children clinically worsened after they met the improvement criteria, a clinical pattern observed previously with supplemental oxygen.[28] This unpredictability may explain some of the variation in providers' clinical decision making.[5] The children who worsened, and therefore deserve more cautious discharge planning, were young (<2 months), premature (<37 weeks gestational age), and presented in more severe distress. Those children admitted to the ICU from the ED worsened more commonly than children admitted to the ward (19% vs 2%). Interestingly, the viral etiology of the child's bronchiolitis did not influence whether a child worsened after reaching the improvement criteria. Therefore, although children with RV bronchiolitis have a shorter hospital LOS than children with RSV bronchiolitis,[11] the pattern of recovery did not differ by viral etiology.

In addition to unsafe discharges, clinicians may be concerned about the possibility of readmissions. Although somewhat controversial, hospital readmission is being used as a quality of care metric.[29, 30, 31] One response to minimize readmissions would be for clinicians to observe children for longer than clinically indicated.[32] However, shorter LOS is not necessarily associated with increased readmission rates.[33] Given that the geometric mean of hospital charges per child with bronchiolitis increased from $6380 in 2000 to $8530 in 2009,[34] the potential for safely reducing hospital LOS by using the discharge criteria proposed in the current study instead of other criteria[8] may net substantial cost savings. Furthermore, reducing LOS would decrease the time children expose others to these respiratory viruses and possibly reduce medical errors.[35]

Our study has some potential limitations. Because the study participants were all hospitalized, these data do not inform admission or discharge decisions from either the ED or the clinic; but other data address those clinical scenarios.[22] Also, the 16 sites that participated in this study were large, urban teaching hospitals. Consequently, these results are not necessarily generalizable to smaller community hospitals. Although numerous data points were required to enter the analytic cohort, only 9% of the sample was excluded for missing data. There were 214 children who did not meet our improvement criteria by the time of discharge. Although the inability to include these children in the analysis may be seen as a limitation, this practice variability underscores the need for more data about discharging hospitalized children with bronchiolitis. Last, site teams reviewed medical records daily. More frequent recording of the clinical course would have yielded more granular data, but the current methodology replicates how data are generally presented during patient care rounds, when decisions about suitability for discharge are often considered.

CONCLUSION

We documented in this large multicenter study that most children hospitalized with bronchiolitis had a wide range of time to recovery, but the vast majority continued to improve once they reached the identified clinical criteria that predict a safe discharge to home. The children who worsened after clinical improvement were more likely to be younger, premature infants presenting in more severe distress. Although additional prospective validation of these hospital discharge criteria is warranted, these data may help clinicians make more evidence‐based discharge decisions for a common pediatric illness with high practice variation, both in the United States[3] and in other countries.[4, 5]

Although bronchiolitis is the leading cause of hospitalization for US infants,[1] there is a lack of basic prospective data about the expected inpatient clinical course and ongoing uncertainty about when a hospitalized child is ready for discharge to home.[2] This lack of data about children's readiness for discharge may result in variable hospital length‐of‐stay (LOS).[3, 4, 5]

One specific source of variability in discharge readiness and LOS variability may be the lack of consensus about safe threshold oxygen saturation values for discharge in children hospitalized with bronchiolitis.[6, 7] In 2006, the Scottish Intercollegiate Guidelines Network recommended a discharge room air oxygen (RAO2) saturation threshold of 95%.[8] The same year, the American Academy of Pediatrics (AAP) bronchiolitis clinical practice guideline stated that oxygen is not needed for children with RAO2 saturations 90% who are feeding well and have minimal respiratory distress.[9] There is a need for prospective studies to help clinicians make evidenced‐based discharge decisions for this common condition.

We performed a prospective, multicenter, multiyear study[10, 11, 12] to examine the typical inpatient clinical course of and to develop hospital discharge guidelines for children age <2 years hospitalized with bronchiolitis. We hypothesized that children would not worsen clinically and would be safe to discharge home once their respiratory status improved and they were able to remain hydrated.

METHODS

RESULTS

There were 1916 children hospitalized with bronchiolitis with data on all factors used to define clinical improvement and clinical worsening. The median number of days from the beginning of difficulty breathing until admission was 2 days (IQR, 15.5 days; range, 18 days) and from the beginning of difficulty breathing until clinical improvement was 4 days (IQR, 37.5 days; range, 133 days) (Figure 1). The variance for days to admission was significantly less than the variance for days to clinical improvement (P<0.001).

Figure 1

In this observational study, clinicians discharged 214 (11%) of the 1916 children before meeting the definition of clinical improvement. Thus, 1702 (89%; 95% CI: 87%‐90%) children reached the clinical improvement criteria, had a LOS >1 day, and had data on all factors (Figure 2).

Figure 2

Of the 1702 children who met the clinical improvement criteria, there were 76 children (4%; 95% CI: 3%5%) who worsened (Figure 2). The worsening occurred within a median of 1 day (IQR, 13 days) of clinical improvement. Forty‐six (3%) of the children required transfer to the ICU (1 required intubation, 1 required continuous positive airway pressure, and 4 had apnea), 23 (1%) required oxygen, and 17 (1%) required IV fluids. Eight percent of children met multiple criteria for worsening. A comparison between children who did and did not worsen is shown in Table 1. In general, children who worsened after improvement were younger and born earlier. These children also presented in more severe respiratory distress, had moderate or severe retractions, oxygen saturation <85% at hospitalization, inadequate oral intake, and apnea documented during the hospitalization. Neither viral etiology nor site of care influenced whether the children worsened after improving. However, stratified analysis of children based on initial location of admission (ie, ICU or ward) showed that among the children admitted to the ICU from the emergency department (ED), 89% met the improvement criteria and 19% clinically worsened. In contrast, among children admitted to the ward from the ED, 89% met the improvement criteria, and only 2% clinically worsened. Stratified multivariable models based on the initial location of admission from the ED (ie, ICU or ward) were not possible due to small sample sizes after stratification. None of these children had relapse events requiring rehospitalization within either 24 hours or 7 days of discharge.

On multivariable analysis (Table 2), independent risk factors for worsening after reaching the clinical improvement criteria were young age, preterm birth, and presenting to care with more severe bronchiolitis represented by severe retractions, inadequate oral intake, or apnea. To further evaluate the improvement criteria in the current analysis, multiple sensitivity analyses were conducted. The frequency of clinical worsening after reaching the improvement criteria was stable when we examined different RA02 criteria in sensitivity analyses: (1) excluding RA02 as a criterion for improvement: 90% met improvement criteria and 4% experienced clinical worsening, (2) changing the average RA02 threshold for clinical improvement to 94%: 62% met improvement criteria and 6% experienced clinical worsening, and (3) changing the average RA02 threshold for clinical improvement to 95%: 47% met improvement criteria and 5% experienced clinical worsening. Furthermore, stratifying by age <2 months and restricting to more stringent definitions of bronchiolitis (ie, age <1 year or age <1 year+no history of wheezing) also did not materially change the results (see Supporting Figure 1 in the online version of this article).

We compared the 214 children who were discharged prior to reaching clinical improvement with the 1702 children who reached the clinical improvement criteria. The 214 children were less likely to be age <2 months (22% vs 30%; P=0.02). These 2 groups (214 vs 1702) were similar with respect to severe retractions (2% vs 4%; P=0.13), median respiratory rate (48 vs 48; P=0.42), oxygen saturation <90% (15% vs 11%; P=0.07), inadequate oral intake (50% vs 43%; P=0.13), and rates of relapse events requiring rehospitalization within both 24 hours (0.6% vs 0.6%; P=0.88) and 7 days (1% vs 1%; P=0.90) of discharge.

DISCUSSION

In this large, multicenter, multiyear study of children hospitalized with bronchiolitis, we found that children present to a hospital in a relatively narrow time frame, but their time to recovery in the hospital is highly variable. Nonetheless, 96% of children continued to improve once they had: (1) improving or stable retractions rated as none/mild, (2) a decreasing or stable RR by age, (3) estimated average RAO2 saturation 90% and lowest RAO2 saturation of 88%, and (4) were hydrated. The 4% of children who worsened after clinically improving were more likely to be age <2 months, born <37 weeks, and present with more severe distress (ie, severe retractions, inadequate oral intake, or apnea). Based on the low risk of worsening after clinical improvement, especially among children admitted to the regular ward (2%), we believe these 4 clinical criteria could be used as discharge criteria for this common pediatric illness with a predominantly monophasic clinical course.

Variability in hospital LOS for children with bronchiolitis exists in the United States[3] and internationally.[4, 5] Cheung and colleagues analyzed administrative data from over 75,000 children admitted for bronchiolitis in England between April 2007 and March 2010 and found sixfold variation in LOS between sites. They concluded that this LOS variability was due in part to providers' clinical decision making.[5] Srivastava and colleagues[23] addressed variable clinician decision making in bronchiolitis and 10 other common pediatric conditions by embedding discharge criteria developed by expert consensus into admission order sets. They found that for children with bronchiolitis, the embedded discharge criteria reduced the median LOS from 1.91 to 1.87 days. In contrast to the single‐center data presented by White and colleagues,[24] the prospective, multicenter MARC‐30 data provide a clear understanding of the normal clinical course for children hospitalized with bronchiolitis, determine if children clinically worsen after clinical improvement, and provide data about discharge criteria for children hospitalized with bronchiolitis. Although there is a lack of rigorous published data, the lower tract symptoms of bronchiolitis (eg, cough, retractions) are said to peak on days 5 to 7 of illness and then gradually resolve.[25] In the present study, we found that the time from the onset of difficulty breathing until hospital admission is less variable than the time from the onset of difficulty breathing until either clinical improvement or discharge. Although 75% of children have clinically improved within 7.5 days of difficulty breathing based on the IQR results, the remaining 25% may have a more prolonged recovery in the hospital of up to 3 weeks. Interestingly, prolonged recovery times from bronchiolitis have also been noted in children presenting to the ED[26] and in an outpatient population.[27] It is unclear why 20% to 25% of children at different levels of severity of illness have prolonged recovery from bronchiolitis, but this group of children requires further investigation.

Given the variability of recovery times, clinicians may have difficulty knowing when a child is ready for hospital discharge. One of the main stumbling blocks for discharge readiness in children with bronchiolitis is the interpretation of the oxygen saturation value.[6, 8, 9, 20, 28] However, it should be considered that interpreting the oxygen saturation in a child who is clinically improving in the hospital setting is different than interpreting the oxygen saturation of a child in the ED or the clinic whose clinical course is less certain.[22] In the hospital setting, using the oxygen saturation value in in the AAP guideline,[9] 4% of children clinically worsened after they met the improvement criteria, a clinical pattern observed previously with supplemental oxygen.[28] This unpredictability may explain some of the variation in providers' clinical decision making.[5] The children who worsened, and therefore deserve more cautious discharge planning, were young (<2 months), premature (<37 weeks gestational age), and presented in more severe distress. Those children admitted to the ICU from the ED worsened more commonly than children admitted to the ward (19% vs 2%). Interestingly, the viral etiology of the child's bronchiolitis did not influence whether a child worsened after reaching the improvement criteria. Therefore, although children with RV bronchiolitis have a shorter hospital LOS than children with RSV bronchiolitis,[11] the pattern of recovery did not differ by viral etiology.

In addition to unsafe discharges, clinicians may be concerned about the possibility of readmissions. Although somewhat controversial, hospital readmission is being used as a quality of care metric.[29, 30, 31] One response to minimize readmissions would be for clinicians to observe children for longer than clinically indicated.[32] However, shorter LOS is not necessarily associated with increased readmission rates.[33] Given that the geometric mean of hospital charges per child with bronchiolitis increased from $6380 in 2000 to $8530 in 2009,[34] the potential for safely reducing hospital LOS by using the discharge criteria proposed in the current study instead of other criteria[8] may net substantial cost savings. Furthermore, reducing LOS would decrease the time children expose others to these respiratory viruses and possibly reduce medical errors.[35]

Our study has some potential limitations. Because the study participants were all hospitalized, these data do not inform admission or discharge decisions from either the ED or the clinic; but other data address those clinical scenarios.[22] Also, the 16 sites that participated in this study were large, urban teaching hospitals. Consequently, these results are not necessarily generalizable to smaller community hospitals. Although numerous data points were required to enter the analytic cohort, only 9% of the sample was excluded for missing data. There were 214 children who did not meet our improvement criteria by the time of discharge. Although the inability to include these children in the analysis may be seen as a limitation, this practice variability underscores the need for more data about discharging hospitalized children with bronchiolitis. Last, site teams reviewed medical records daily. More frequent recording of the clinical course would have yielded more granular data, but the current methodology replicates how data are generally presented during patient care rounds, when decisions about suitability for discharge are often considered.

CONCLUSION

We documented in this large multicenter study that most children hospitalized with bronchiolitis had a wide range of time to recovery, but the vast majority continued to improve once they reached the identified clinical criteria that predict a safe discharge to home. The children who worsened after clinical improvement were more likely to be younger, premature infants presenting in more severe distress. Although additional prospective validation of these hospital discharge criteria is warranted, these data may help clinicians make more evidence‐based discharge decisions for a common pediatric illness with high practice variation, both in the United States[3] and in other countries.[4, 5]

Although bronchiolitis is the leading cause of hospitalization for US infants,[1] there is a lack of basic prospective data about the expected inpatient clinical course and ongoing uncertainty about when a hospitalized child is ready for discharge to home.[2] This lack of data about children's readiness for discharge may result in variable hospital length‐of‐stay (LOS).[3, 4, 5]

One specific source of variability in discharge readiness and LOS variability may be the lack of consensus about safe threshold oxygen saturation values for discharge in children hospitalized with bronchiolitis.[6, 7] In 2006, the Scottish Intercollegiate Guidelines Network recommended a discharge room air oxygen (RAO2) saturation threshold of 95%.[8] The same year, the American Academy of Pediatrics (AAP) bronchiolitis clinical practice guideline stated that oxygen is not needed for children with RAO2 saturations 90% who are feeding well and have minimal respiratory distress.[9] There is a need for prospective studies to help clinicians make evidenced‐based discharge decisions for this common condition.

We performed a prospective, multicenter, multiyear study[10, 11, 12] to examine the typical inpatient clinical course of and to develop hospital discharge guidelines for children age <2 years hospitalized with bronchiolitis. We hypothesized that children would not worsen clinically and would be safe to discharge home once their respiratory status improved and they were able to remain hydrated.

METHODS

RESULTS

There were 1916 children hospitalized with bronchiolitis with data on all factors used to define clinical improvement and clinical worsening. The median number of days from the beginning of difficulty breathing until admission was 2 days (IQR, 15.5 days; range, 18 days) and from the beginning of difficulty breathing until clinical improvement was 4 days (IQR, 37.5 days; range, 133 days) (Figure 1). The variance for days to admission was significantly less than the variance for days to clinical improvement (P<0.001).

Figure 1

In this observational study, clinicians discharged 214 (11%) of the 1916 children before meeting the definition of clinical improvement. Thus, 1702 (89%; 95% CI: 87%‐90%) children reached the clinical improvement criteria, had a LOS >1 day, and had data on all factors (Figure 2).

Figure 2

Of the 1702 children who met the clinical improvement criteria, there were 76 children (4%; 95% CI: 3%5%) who worsened (Figure 2). The worsening occurred within a median of 1 day (IQR, 13 days) of clinical improvement. Forty‐six (3%) of the children required transfer to the ICU (1 required intubation, 1 required continuous positive airway pressure, and 4 had apnea), 23 (1%) required oxygen, and 17 (1%) required IV fluids. Eight percent of children met multiple criteria for worsening. A comparison between children who did and did not worsen is shown in Table 1. In general, children who worsened after improvement were younger and born earlier. These children also presented in more severe respiratory distress, had moderate or severe retractions, oxygen saturation <85% at hospitalization, inadequate oral intake, and apnea documented during the hospitalization. Neither viral etiology nor site of care influenced whether the children worsened after improving. However, stratified analysis of children based on initial location of admission (ie, ICU or ward) showed that among the children admitted to the ICU from the emergency department (ED), 89% met the improvement criteria and 19% clinically worsened. In contrast, among children admitted to the ward from the ED, 89% met the improvement criteria, and only 2% clinically worsened. Stratified multivariable models based on the initial location of admission from the ED (ie, ICU or ward) were not possible due to small sample sizes after stratification. None of these children had relapse events requiring rehospitalization within either 24 hours or 7 days of discharge.

On multivariable analysis (Table 2), independent risk factors for worsening after reaching the clinical improvement criteria were young age, preterm birth, and presenting to care with more severe bronchiolitis represented by severe retractions, inadequate oral intake, or apnea. To further evaluate the improvement criteria in the current analysis, multiple sensitivity analyses were conducted. The frequency of clinical worsening after reaching the improvement criteria was stable when we examined different RA02 criteria in sensitivity analyses: (1) excluding RA02 as a criterion for improvement: 90% met improvement criteria and 4% experienced clinical worsening, (2) changing the average RA02 threshold for clinical improvement to 94%: 62% met improvement criteria and 6% experienced clinical worsening, and (3) changing the average RA02 threshold for clinical improvement to 95%: 47% met improvement criteria and 5% experienced clinical worsening. Furthermore, stratifying by age <2 months and restricting to more stringent definitions of bronchiolitis (ie, age <1 year or age <1 year+no history of wheezing) also did not materially change the results (see Supporting Figure 1 in the online version of this article).

We compared the 214 children who were discharged prior to reaching clinical improvement with the 1702 children who reached the clinical improvement criteria. The 214 children were less likely to be age <2 months (22% vs 30%; P=0.02). These 2 groups (214 vs 1702) were similar with respect to severe retractions (2% vs 4%; P=0.13), median respiratory rate (48 vs 48; P=0.42), oxygen saturation <90% (15% vs 11%; P=0.07), inadequate oral intake (50% vs 43%; P=0.13), and rates of relapse events requiring rehospitalization within both 24 hours (0.6% vs 0.6%; P=0.88) and 7 days (1% vs 1%; P=0.90) of discharge.

DISCUSSION

In this large, multicenter, multiyear study of children hospitalized with bronchiolitis, we found that children present to a hospital in a relatively narrow time frame, but their time to recovery in the hospital is highly variable. Nonetheless, 96% of children continued to improve once they had: (1) improving or stable retractions rated as none/mild, (2) a decreasing or stable RR by age, (3) estimated average RAO2 saturation 90% and lowest RAO2 saturation of 88%, and (4) were hydrated. The 4% of children who worsened after clinically improving were more likely to be age <2 months, born <37 weeks, and present with more severe distress (ie, severe retractions, inadequate oral intake, or apnea). Based on the low risk of worsening after clinical improvement, especially among children admitted to the regular ward (2%), we believe these 4 clinical criteria could be used as discharge criteria for this common pediatric illness with a predominantly monophasic clinical course.

Variability in hospital LOS for children with bronchiolitis exists in the United States[3] and internationally.[4, 5] Cheung and colleagues analyzed administrative data from over 75,000 children admitted for bronchiolitis in England between April 2007 and March 2010 and found sixfold variation in LOS between sites. They concluded that this LOS variability was due in part to providers' clinical decision making.[5] Srivastava and colleagues[23] addressed variable clinician decision making in bronchiolitis and 10 other common pediatric conditions by embedding discharge criteria developed by expert consensus into admission order sets. They found that for children with bronchiolitis, the embedded discharge criteria reduced the median LOS from 1.91 to 1.87 days. In contrast to the single‐center data presented by White and colleagues,[24] the prospective, multicenter MARC‐30 data provide a clear understanding of the normal clinical course for children hospitalized with bronchiolitis, determine if children clinically worsen after clinical improvement, and provide data about discharge criteria for children hospitalized with bronchiolitis. Although there is a lack of rigorous published data, the lower tract symptoms of bronchiolitis (eg, cough, retractions) are said to peak on days 5 to 7 of illness and then gradually resolve.[25] In the present study, we found that the time from the onset of difficulty breathing until hospital admission is less variable than the time from the onset of difficulty breathing until either clinical improvement or discharge. Although 75% of children have clinically improved within 7.5 days of difficulty breathing based on the IQR results, the remaining 25% may have a more prolonged recovery in the hospital of up to 3 weeks. Interestingly, prolonged recovery times from bronchiolitis have also been noted in children presenting to the ED[26] and in an outpatient population.[27] It is unclear why 20% to 25% of children at different levels of severity of illness have prolonged recovery from bronchiolitis, but this group of children requires further investigation.

Given the variability of recovery times, clinicians may have difficulty knowing when a child is ready for hospital discharge. One of the main stumbling blocks for discharge readiness in children with bronchiolitis is the interpretation of the oxygen saturation value.[6, 8, 9, 20, 28] However, it should be considered that interpreting the oxygen saturation in a child who is clinically improving in the hospital setting is different than interpreting the oxygen saturation of a child in the ED or the clinic whose clinical course is less certain.[22] In the hospital setting, using the oxygen saturation value in in the AAP guideline,[9] 4% of children clinically worsened after they met the improvement criteria, a clinical pattern observed previously with supplemental oxygen.[28] This unpredictability may explain some of the variation in providers' clinical decision making.[5] The children who worsened, and therefore deserve more cautious discharge planning, were young (<2 months), premature (<37 weeks gestational age), and presented in more severe distress. Those children admitted to the ICU from the ED worsened more commonly than children admitted to the ward (19% vs 2%). Interestingly, the viral etiology of the child's bronchiolitis did not influence whether a child worsened after reaching the improvement criteria. Therefore, although children with RV bronchiolitis have a shorter hospital LOS than children with RSV bronchiolitis,[11] the pattern of recovery did not differ by viral etiology.

In addition to unsafe discharges, clinicians may be concerned about the possibility of readmissions. Although somewhat controversial, hospital readmission is being used as a quality of care metric.[29, 30, 31] One response to minimize readmissions would be for clinicians to observe children for longer than clinically indicated.[32] However, shorter LOS is not necessarily associated with increased readmission rates.[33] Given that the geometric mean of hospital charges per child with bronchiolitis increased from $6380 in 2000 to $8530 in 2009,[34] the potential for safely reducing hospital LOS by using the discharge criteria proposed in the current study instead of other criteria[8] may net substantial cost savings. Furthermore, reducing LOS would decrease the time children expose others to these respiratory viruses and possibly reduce medical errors.[35]

Our study has some potential limitations. Because the study participants were all hospitalized, these data do not inform admission or discharge decisions from either the ED or the clinic; but other data address those clinical scenarios.[22] Also, the 16 sites that participated in this study were large, urban teaching hospitals. Consequently, these results are not necessarily generalizable to smaller community hospitals. Although numerous data points were required to enter the analytic cohort, only 9% of the sample was excluded for missing data. There were 214 children who did not meet our improvement criteria by the time of discharge. Although the inability to include these children in the analysis may be seen as a limitation, this practice variability underscores the need for more data about discharging hospitalized children with bronchiolitis. Last, site teams reviewed medical records daily. More frequent recording of the clinical course would have yielded more granular data, but the current methodology replicates how data are generally presented during patient care rounds, when decisions about suitability for discharge are often considered.

CONCLUSION

We documented in this large multicenter study that most children hospitalized with bronchiolitis had a wide range of time to recovery, but the vast majority continued to improve once they reached the identified clinical criteria that predict a safe discharge to home. The children who worsened after clinical improvement were more likely to be younger, premature infants presenting in more severe distress. Although additional prospective validation of these hospital discharge criteria is warranted, these data may help clinicians make more evidence‐based discharge decisions for a common pediatric illness with high practice variation, both in the United States[3] and in other countries.[4, 5]