At the Intersection of Health, Health Care and Policy Cite this article as: Karen E. Joynt, Paula Chatterjee, E. John Orav and Ashish K. Jha Hospital Closures Had No Measurable Impact On Local Hospitalization Rates Or Mortality Rates, 2003−11 Health Affairs, 34, no.5 (2015):765-772 doi: 10.1377/hlthaff.2014.1352
The online version of this article, along with updated information and services, is available at: http://content.healthaffairs.org/content/34/5/765.full.html
For Reprints, Links & Permissions: http://healthaffairs.org/1340_reprints.php E-mail Alerts : http://content.healthaffairs.org/subscriptions/etoc.dtl To Subscribe: http://content.healthaffairs.org/subscriptions/online.shtml
Health Affairs is published monthly by Project HOPE at 7500 Old Georgetown Road, Suite 600, Bethesda, MD 20814-6133. Copyright © 2015 by Project HOPE - The People-to-People Health Foundation. As provided by United States copyright law (Title 17, U.S. Code), no part of Health Affairs may be reproduced, displayed, or transmitted in any form or by any means, electronic or mechanical, including photocopying or by information storage or retrieval systems, without prior written permission from the Publisher. All rights reserved.
Not for commercial use or unauthorized distribution Downloaded from content.healthaffairs.org by Health Affairs on May 9, 2015 at NYU LANGONE MED CTR SCHOOL OF MEDICINE
Access To Care By Karen E. Joynt, Paula Chatterjee, E. John Orav, and Ashish K. Jha 10.1377/hlthaff.2014.1352 HEALTH AFFAIRS 34, NO. 5 (2015): 765–772 ©2015 Project HOPE— The People-to-People Health Foundation, Inc.
doi:
Hospital Closures Had No Measurable Impact On Local Hospitalization Rates Or Mortality Rates, 2003–11
Karen E. Joynt (kjoynt@ partners.org) is an assistant professor in the Division of Cardiovascular Medicine, Harvard Medical School and Brigham and Women’s Hospital, and an instructor at the Harvard T.H. Chan School of Public Health in the Department of Health Policy and Management, both in Boston, Massachusetts. She is currently serving as a senior adviser to the deputy assistant secretary for health policy in the Office of the Assistant Secretary for Planning and Evaluation, Department of Health and Human Services, in Washington, D.C.
The Affordable Care Act (ACA) set in motion payment changes that could put pressure on hospital finances and lead some hospitals to close. Understanding the impact of closures on patient care and outcomes is critically important. We identified 195 hospital closures in the United States between 2003 and 2011. We found no significant difference between the change in annual mortality rates for patients living in hospital service areas (HSAs) that experienced one or more closures and the change in rates in matched HSAs without a closure (5.5 percent to 5.2 percent versus 5.4 percent to 5.4 percent, respectively). Nor was there a significant difference in the change in all-cause mortality rates following hospitalization (9.1 percent to 8.2 percent in HSAs with a closure versus 9.0 percent to 8.4 percent in those without a closure). HSAs with a closure had a drop in readmission rates compared to controls (19.4 percent to 18.2 percent versus 18.8 percent to 18.3 percent). Overall, we found no evidence that hospital closures were associated with worse outcomes for patients living in those communities. These findings may offer reassurance to policy makers and clinical leaders concerned about the potential acceleration of hospital closures as a result of health care reform. ABSTRACT
H
ospital closures may accelerate in coming years as provisions of the Affordable Care Act (ACA) put increasing pressure on hospital finances—for example, through reductions in the rate of growth in payments for hospital services and reductions in disproportionate share and graduate medical education funding.1 ACA insurance expansion provisions may offset some of this financial pressure.2 However, many policy makers and political leaders worry that hospital closures could have important negative effects for patients seeking access to acute care services,3,4 especially for conditions for which timely care has an impact on the outcome. Whether hospital closures have meaningful
Paula Chatterjee is a resident in medicine at Brigham and Women’s Hospital.
adverse consequences for access and patient outcomes is unclear. If hospitals that close have a poor record of quality and safety, and if there are better-performing institutions in close proximity, patients may be better off seeking care elsewhere. Alternatively, if hospital closures occur in areas with few other health care alternatives, patients may be worse off. There are very few data on the impact of hospital closures nationally: Most previous research focused on primarily rural areas or specific regions of the country. National data on what happens to patients after hospitals close would be very useful in helping federal policy makers and city and state government officials determine how and when to step in to prevent hospital closures. MAY 2 015
34:5
Downloaded from content.healthaffairs.org by Health Affairs on May 9, 2015 at NYU LANGONE MED CTR SCHOOL OF MEDICINE
E. John Orav is an associate professor of medicine (biostatistics) at Harvard Medical School and Brigham and Women’s Hospital and an associate professor of biostatistics at the Harvard T.H. Chan School of Public Health. Ashish K. Jha is the K.T. Li Professor of Health Policy at the Harvard T.H. Chan School of Public Health.
Health Affa irs
765
Access To Care Given the relatively limited empirical evidence on hospital closures nationwide and their impact on patient outcomes, we sought to answer three questions: First, what are the characteristics of hospitals that closed over the past decade and the communities they served? Second, what is the relationship between hospital closures and patient outcomes (mortality and readmission rates) in a community? And third, does the relationship between hospital closures and patient outcomes differ by the acuity of the medical condition considered or by the rurality of the area in which the closure occurred?
Study Data And Methods Hospitals Our primary predictor was hospital closure. To identify closed hospitals, we first used the “landscape change” reports from the American Hospital Association from the period 2005–10.5 These reports document year-to-year changes in the national hospital census and verify the cause for the changes (such as a hospital merger or acquisition, the opening of a new hospital, or a closure). We confirmed the closures using Medicare cost reports for the same time period. Given the high degree of fidelity between our two sources, we expanded the study period to include closures during 2003–05 and during 2010–11 using solely the Medicare cost reports. Thus, the total study period was from 2003 to 2011. For each suspected closure identified by one or both of our sources, we confirmed the hospital’s status using multiple additional sources, including state health departments and local newspaper articles describing specific closures. We also examined the number of yearly hospitalizations for each hospital, using acute care claims in Medicare inpatient files. We considered that closures and the year of closure were confirmed if the number of hospitalizations declined to zero at the appropriate time. After identifying all closed hospitals, we used the complete census of US hospitals for the study period to define a comparison group of currently open hospitals. Hospital Service Areas To understand the impact of a closure on the community of patients who could feasibly have sought care at the closed hospital, we performed all analyses at the hospital service area (HSA) level. As defined by the Dartmouth Atlas of Health Care,6 an HSA is a local health care market that represents travel patterns for patients receiving primary hospital care. There are 3,436 HSAs in the United States, many of which contain a single hospital. We assigned each hospital closure to its designated HSA. For comparison, control HSAs were 76 6
Health Affa irs
M AY 20 1 5
3 4 :5
identified using three matching variables: geographic region (one of the ten regions defined by the Centers for Disease Control and Prevention), rurality (defined by rural-urban commuting area code), and baseline risk-adjusted all-cause annual mortality rates. We identified three control HSAs for each closure HSA to reduce the likelihood of unmeasured confounding, since there might have been differences in outcomes based on rurality or region that would otherwise be inadequately accounted for. To address this and be certain that our results were robust to our choice of controls, we also repeated our analyses using all nonclosure HSAs as the control group. In addition, we repeated our analysis using the hospital referral region (HRR) as the unit of analysis and comparing HRRs with closures to all other HRRs. This approach had the advantage of examining the larger hospital market. However, it also had the disadvantage of diluting our ability to detect the impact of a closure on access to care, since only a small proportion of patients in an HRR would be expected to be affected by any given closure. Patients We used Medicare denominator and inpatient files to identify people enrolled in fee-for-service Medicare from January 1, 2002, through December 31, 2012. We assigned beneficiaries to HSAs using their home ZIP codes.We obtained from the Medicare files the following patient characteristics: age, sex, self-reported race/ethnicity, Medicaid eligibility, and medical comorbidities as classified by Anne Elixhauser and coauthors.7 Other Variables Using the Medicare cost reports and American Hospital Association surveys, we extracted the following hospital characteristics: size, geographic region, ownership, rural-urban commuting area code, teaching status (the ratio of interns or residents to beds), critical access status, safety-net status (in the top 25 percent of the disproportionate-share hospital index), proportion of Medicare patients, and proportion of Medicaid patients. We also calculated markers of hospital financial status from the cost reports, including total margins (net income divided by total revenue) and operating margins (the product of net patient revenue minus total operating expenses, divided by net patient revenue) and costs of various types of uncompensated or undercompensated care. To understand more about the context in which each closure occurred, we also examined HSA-level variables from the year prior to a closure. Using the Dartmouth Atlas,6 we extracted the following HSA-level characteristics: age, sex, race, and price-adjusted total Medicare Parts A and B reimbursements per enrollee.We used the
Downloaded from content.healthaffairs.org by Health Affairs on May 9, 2015 at NYU LANGONE MED CTR SCHOOL OF MEDICINE
Existing structures of markets and competition may work reasonably well in selecting hospitals for closure.
same source to obtain the following supply-side variables: the numbers of all physicians, all specialists, and primary care physicians per 100,000 residents and of acute care hospital beds per 1,000 residents. Outcomes Our primary outcome was all-cause annual mortality at the HSA (population) level, calculated across all beneficiaries residing in the HSAs of interest. Secondary outcomes at the population level included annual hospital admission rates per 1,000 beneficiaries and annual average per beneficiary inpatient spending. We also examined the following outcomes that were limited to patients who were hospitalized: all-cause riskadjusted thirty-day mortality rates, all-cause riskadjusted thirty-day readmission rates, lengthof-stay, and risk-adjusted inpatient costs per hospitalization.We used the Elixhauser method7 for risk adjustment where appropriate. We assessed mortality at both the HSA and the hospitalization levels because the rates at the two levels allowed us to examine different issues. Population-based mortality rates offered a more global look at the impact of closures and accounted for deaths that might occur outside of the hospital as a result of increased travel time. In contrast, hospitalization-based mortality rates allowed us to examine the hospital care received by patients before and after their local hospital closed. Primary Analyses First, we created a map of each of the closures in our sample.We then compared the characteristics of closed versus open US hospitals using chi-square tests for categorical variables and analysis of variation tests for continuous variables. We used nonparametric Wilcoxon tests to compare the non-normally distributed financial measures of total margin and operating margin. Second, we compared patient demographic characteristics (age, sex, race/ethnicity, and comorbidities) and basic payment and supply characteristics (Medicare reimbursements per
enrollee and supply-side variables) among our three groups of HSAs: HSAs with closures, matched control HSAs without closures, and all HSAs without closures. Appropriate statistical tests were performed to obtain p values. Third, we created difference-in-differences regression models to test the relationship between hospital closure and the change over time in each of our outcomes of interest (population-level mortality, admissions, and costs; and, for hospitalized patients, thirty-day mortality and readmissions, length-of-stay, and costs). Additionally, as proof of concept, we tested for differences in the change in the proportion of hospitalizations occurring outside patients’ home HSAs after a closure. Our models included time period (preclosure versus postclosure), closure status for each HSA (closure versus nonclosure), and the interaction between the two as predictors. The models also included patient age, sex, and medical comorbidities for all mortality, length-of-stay, and cost analyses. We considered the preclosure year to be the calendar year prior to the closure and the postclosure year to be the calendar year following the closure. We excluded the calendar year of the closure. Correlation within HSAs and matched groups was accounted for by using a random effect for group. This matching strategy allowed us to effectively control for secular trends in any of our outcomes over the decade-long study period, since each HSA with a closure was being directly compared to contemporary controls. Sensitivity Analyses We conducted a number of additional analyses. First, to assess whether the impact of closure on mortality varied by condition acuity, we limited our analysis to patients with acute myocardial infarction, stroke, or trauma.8 Second, to determine whether the impact of closure on outcomes differed by rurality or supply of medical services, we subdivided HSAs into rural and urban designations according to their rural-urban commuting area classifications and then repeated our analyses for population-level all-cause mortality rates for each subgroup. We similarly repeated our analyses after stratifying the HSAs by the proportion of beds within an HSA that were removed as a result of a hospital closure, with a particular focus on the HSAs in which all beds were lost because of closure. Finally, to capture adverse effects that may have taken longer to accrue, we compared annual mortality two years before each closure to mortality two years after the closure. A two-tailed p value of less than 0.05 was considered significant. All analyses were performed M AY 2 0 1 5
3 4: 5
Downloaded from content.healthaffairs.org by Health Affairs on May 9, 2015 at NYU LANGONE MED CTR SCHOOL OF MEDICINE
Health A ffairs
767
Access To Care using SAS software, version 9.2. This project was considered exempt by the Office of Human Research Administration of the Harvard T.H. Chan School of Public Health because of the deidentified nature of the data. Limitations Our study had several limitations. First, the use of administrative data limited our ability to control for all clinical confounders. Nonetheless, we doubt whether unmeasured confounders differed over time between closure and nonclosure HSAs in ways that would have meaningfully biased our results. Second, because we used Medicare data, our
Exhibit 1 Baseline Characteristics Of Hospitals That Closed Versus All Other Open Hospitals, 2010 Hospitals Characteristic
Closed
Open
Average number of beds
63.6
93.8
Size Small Medium Large
47.3% 52.2 0.5
57.3% 35.8 6.9
Region Northeast Midwest South West
23.7 17.2 41.9 17.2
16.1 33.9 30.5 19.4
Ownership For-profit Private nonprofit Public
42.2 49.2 8.6
25.1 52.5 22.4
Rural-urban commuting area Urban Suburban Large rural town Small town or isolated rural
70.6 6.7 8.2 14.4
45.0 4.9 16.8 33.3
25.1 12.5 32.8 47.9 16.3
19.3 29.3 17.0 49.7 17.2
Teaching hospital Critical access hospital Safety-net hospital Medicare patients Medicaid patients Margins (percent) Totala Operatingb
−13 −20
Cost (percent of total revenue) Total state and local indigent care programc Total CHIPc Total gross Medicaidc Uncompensated carec
1.6% 0.2 10.4 11.9
4 −1 1.3% 0.2 9.6 7.1
SOURCE Authors’ analysis of data from the American Hospital Association and Medicare cost reports. NOTES Of the hospitals, 195 (4.3 percent) closed, and 4,335 (95.7 percent) remained open. Costs are reported by the hospitals and shown as a percentage of total hospital revenue. All differences are significant (p
The online version of this article, along with updated information and services, is available at: http://content.healthaffairs.org/content/34/5/765.full.html
For Reprints, Links & Permissions: http://healthaffairs.org/1340_reprints.php E-mail Alerts : http://content.healthaffairs.org/subscriptions/etoc.dtl To Subscribe: http://content.healthaffairs.org/subscriptions/online.shtml
Health Affairs is published monthly by Project HOPE at 7500 Old Georgetown Road, Suite 600, Bethesda, MD 20814-6133. Copyright © 2015 by Project HOPE - The People-to-People Health Foundation. As provided by United States copyright law (Title 17, U.S. Code), no part of Health Affairs may be reproduced, displayed, or transmitted in any form or by any means, electronic or mechanical, including photocopying or by information storage or retrieval systems, without prior written permission from the Publisher. All rights reserved.
Not for commercial use or unauthorized distribution Downloaded from content.healthaffairs.org by Health Affairs on May 9, 2015 at NYU LANGONE MED CTR SCHOOL OF MEDICINE
Access To Care By Karen E. Joynt, Paula Chatterjee, E. John Orav, and Ashish K. Jha 10.1377/hlthaff.2014.1352 HEALTH AFFAIRS 34, NO. 5 (2015): 765–772 ©2015 Project HOPE— The People-to-People Health Foundation, Inc.
doi:
Hospital Closures Had No Measurable Impact On Local Hospitalization Rates Or Mortality Rates, 2003–11
Karen E. Joynt (kjoynt@ partners.org) is an assistant professor in the Division of Cardiovascular Medicine, Harvard Medical School and Brigham and Women’s Hospital, and an instructor at the Harvard T.H. Chan School of Public Health in the Department of Health Policy and Management, both in Boston, Massachusetts. She is currently serving as a senior adviser to the deputy assistant secretary for health policy in the Office of the Assistant Secretary for Planning and Evaluation, Department of Health and Human Services, in Washington, D.C.
The Affordable Care Act (ACA) set in motion payment changes that could put pressure on hospital finances and lead some hospitals to close. Understanding the impact of closures on patient care and outcomes is critically important. We identified 195 hospital closures in the United States between 2003 and 2011. We found no significant difference between the change in annual mortality rates for patients living in hospital service areas (HSAs) that experienced one or more closures and the change in rates in matched HSAs without a closure (5.5 percent to 5.2 percent versus 5.4 percent to 5.4 percent, respectively). Nor was there a significant difference in the change in all-cause mortality rates following hospitalization (9.1 percent to 8.2 percent in HSAs with a closure versus 9.0 percent to 8.4 percent in those without a closure). HSAs with a closure had a drop in readmission rates compared to controls (19.4 percent to 18.2 percent versus 18.8 percent to 18.3 percent). Overall, we found no evidence that hospital closures were associated with worse outcomes for patients living in those communities. These findings may offer reassurance to policy makers and clinical leaders concerned about the potential acceleration of hospital closures as a result of health care reform. ABSTRACT
H
ospital closures may accelerate in coming years as provisions of the Affordable Care Act (ACA) put increasing pressure on hospital finances—for example, through reductions in the rate of growth in payments for hospital services and reductions in disproportionate share and graduate medical education funding.1 ACA insurance expansion provisions may offset some of this financial pressure.2 However, many policy makers and political leaders worry that hospital closures could have important negative effects for patients seeking access to acute care services,3,4 especially for conditions for which timely care has an impact on the outcome. Whether hospital closures have meaningful
Paula Chatterjee is a resident in medicine at Brigham and Women’s Hospital.
adverse consequences for access and patient outcomes is unclear. If hospitals that close have a poor record of quality and safety, and if there are better-performing institutions in close proximity, patients may be better off seeking care elsewhere. Alternatively, if hospital closures occur in areas with few other health care alternatives, patients may be worse off. There are very few data on the impact of hospital closures nationally: Most previous research focused on primarily rural areas or specific regions of the country. National data on what happens to patients after hospitals close would be very useful in helping federal policy makers and city and state government officials determine how and when to step in to prevent hospital closures. MAY 2 015
34:5
Downloaded from content.healthaffairs.org by Health Affairs on May 9, 2015 at NYU LANGONE MED CTR SCHOOL OF MEDICINE
E. John Orav is an associate professor of medicine (biostatistics) at Harvard Medical School and Brigham and Women’s Hospital and an associate professor of biostatistics at the Harvard T.H. Chan School of Public Health. Ashish K. Jha is the K.T. Li Professor of Health Policy at the Harvard T.H. Chan School of Public Health.
Health Affa irs
765
Access To Care Given the relatively limited empirical evidence on hospital closures nationwide and their impact on patient outcomes, we sought to answer three questions: First, what are the characteristics of hospitals that closed over the past decade and the communities they served? Second, what is the relationship between hospital closures and patient outcomes (mortality and readmission rates) in a community? And third, does the relationship between hospital closures and patient outcomes differ by the acuity of the medical condition considered or by the rurality of the area in which the closure occurred?
Study Data And Methods Hospitals Our primary predictor was hospital closure. To identify closed hospitals, we first used the “landscape change” reports from the American Hospital Association from the period 2005–10.5 These reports document year-to-year changes in the national hospital census and verify the cause for the changes (such as a hospital merger or acquisition, the opening of a new hospital, or a closure). We confirmed the closures using Medicare cost reports for the same time period. Given the high degree of fidelity between our two sources, we expanded the study period to include closures during 2003–05 and during 2010–11 using solely the Medicare cost reports. Thus, the total study period was from 2003 to 2011. For each suspected closure identified by one or both of our sources, we confirmed the hospital’s status using multiple additional sources, including state health departments and local newspaper articles describing specific closures. We also examined the number of yearly hospitalizations for each hospital, using acute care claims in Medicare inpatient files. We considered that closures and the year of closure were confirmed if the number of hospitalizations declined to zero at the appropriate time. After identifying all closed hospitals, we used the complete census of US hospitals for the study period to define a comparison group of currently open hospitals. Hospital Service Areas To understand the impact of a closure on the community of patients who could feasibly have sought care at the closed hospital, we performed all analyses at the hospital service area (HSA) level. As defined by the Dartmouth Atlas of Health Care,6 an HSA is a local health care market that represents travel patterns for patients receiving primary hospital care. There are 3,436 HSAs in the United States, many of which contain a single hospital. We assigned each hospital closure to its designated HSA. For comparison, control HSAs were 76 6
Health Affa irs
M AY 20 1 5
3 4 :5
identified using three matching variables: geographic region (one of the ten regions defined by the Centers for Disease Control and Prevention), rurality (defined by rural-urban commuting area code), and baseline risk-adjusted all-cause annual mortality rates. We identified three control HSAs for each closure HSA to reduce the likelihood of unmeasured confounding, since there might have been differences in outcomes based on rurality or region that would otherwise be inadequately accounted for. To address this and be certain that our results were robust to our choice of controls, we also repeated our analyses using all nonclosure HSAs as the control group. In addition, we repeated our analysis using the hospital referral region (HRR) as the unit of analysis and comparing HRRs with closures to all other HRRs. This approach had the advantage of examining the larger hospital market. However, it also had the disadvantage of diluting our ability to detect the impact of a closure on access to care, since only a small proportion of patients in an HRR would be expected to be affected by any given closure. Patients We used Medicare denominator and inpatient files to identify people enrolled in fee-for-service Medicare from January 1, 2002, through December 31, 2012. We assigned beneficiaries to HSAs using their home ZIP codes.We obtained from the Medicare files the following patient characteristics: age, sex, self-reported race/ethnicity, Medicaid eligibility, and medical comorbidities as classified by Anne Elixhauser and coauthors.7 Other Variables Using the Medicare cost reports and American Hospital Association surveys, we extracted the following hospital characteristics: size, geographic region, ownership, rural-urban commuting area code, teaching status (the ratio of interns or residents to beds), critical access status, safety-net status (in the top 25 percent of the disproportionate-share hospital index), proportion of Medicare patients, and proportion of Medicaid patients. We also calculated markers of hospital financial status from the cost reports, including total margins (net income divided by total revenue) and operating margins (the product of net patient revenue minus total operating expenses, divided by net patient revenue) and costs of various types of uncompensated or undercompensated care. To understand more about the context in which each closure occurred, we also examined HSA-level variables from the year prior to a closure. Using the Dartmouth Atlas,6 we extracted the following HSA-level characteristics: age, sex, race, and price-adjusted total Medicare Parts A and B reimbursements per enrollee.We used the
Downloaded from content.healthaffairs.org by Health Affairs on May 9, 2015 at NYU LANGONE MED CTR SCHOOL OF MEDICINE
Existing structures of markets and competition may work reasonably well in selecting hospitals for closure.
same source to obtain the following supply-side variables: the numbers of all physicians, all specialists, and primary care physicians per 100,000 residents and of acute care hospital beds per 1,000 residents. Outcomes Our primary outcome was all-cause annual mortality at the HSA (population) level, calculated across all beneficiaries residing in the HSAs of interest. Secondary outcomes at the population level included annual hospital admission rates per 1,000 beneficiaries and annual average per beneficiary inpatient spending. We also examined the following outcomes that were limited to patients who were hospitalized: all-cause riskadjusted thirty-day mortality rates, all-cause riskadjusted thirty-day readmission rates, lengthof-stay, and risk-adjusted inpatient costs per hospitalization.We used the Elixhauser method7 for risk adjustment where appropriate. We assessed mortality at both the HSA and the hospitalization levels because the rates at the two levels allowed us to examine different issues. Population-based mortality rates offered a more global look at the impact of closures and accounted for deaths that might occur outside of the hospital as a result of increased travel time. In contrast, hospitalization-based mortality rates allowed us to examine the hospital care received by patients before and after their local hospital closed. Primary Analyses First, we created a map of each of the closures in our sample.We then compared the characteristics of closed versus open US hospitals using chi-square tests for categorical variables and analysis of variation tests for continuous variables. We used nonparametric Wilcoxon tests to compare the non-normally distributed financial measures of total margin and operating margin. Second, we compared patient demographic characteristics (age, sex, race/ethnicity, and comorbidities) and basic payment and supply characteristics (Medicare reimbursements per
enrollee and supply-side variables) among our three groups of HSAs: HSAs with closures, matched control HSAs without closures, and all HSAs without closures. Appropriate statistical tests were performed to obtain p values. Third, we created difference-in-differences regression models to test the relationship between hospital closure and the change over time in each of our outcomes of interest (population-level mortality, admissions, and costs; and, for hospitalized patients, thirty-day mortality and readmissions, length-of-stay, and costs). Additionally, as proof of concept, we tested for differences in the change in the proportion of hospitalizations occurring outside patients’ home HSAs after a closure. Our models included time period (preclosure versus postclosure), closure status for each HSA (closure versus nonclosure), and the interaction between the two as predictors. The models also included patient age, sex, and medical comorbidities for all mortality, length-of-stay, and cost analyses. We considered the preclosure year to be the calendar year prior to the closure and the postclosure year to be the calendar year following the closure. We excluded the calendar year of the closure. Correlation within HSAs and matched groups was accounted for by using a random effect for group. This matching strategy allowed us to effectively control for secular trends in any of our outcomes over the decade-long study period, since each HSA with a closure was being directly compared to contemporary controls. Sensitivity Analyses We conducted a number of additional analyses. First, to assess whether the impact of closure on mortality varied by condition acuity, we limited our analysis to patients with acute myocardial infarction, stroke, or trauma.8 Second, to determine whether the impact of closure on outcomes differed by rurality or supply of medical services, we subdivided HSAs into rural and urban designations according to their rural-urban commuting area classifications and then repeated our analyses for population-level all-cause mortality rates for each subgroup. We similarly repeated our analyses after stratifying the HSAs by the proportion of beds within an HSA that were removed as a result of a hospital closure, with a particular focus on the HSAs in which all beds were lost because of closure. Finally, to capture adverse effects that may have taken longer to accrue, we compared annual mortality two years before each closure to mortality two years after the closure. A two-tailed p value of less than 0.05 was considered significant. All analyses were performed M AY 2 0 1 5
3 4: 5
Downloaded from content.healthaffairs.org by Health Affairs on May 9, 2015 at NYU LANGONE MED CTR SCHOOL OF MEDICINE
Health A ffairs
767
Access To Care using SAS software, version 9.2. This project was considered exempt by the Office of Human Research Administration of the Harvard T.H. Chan School of Public Health because of the deidentified nature of the data. Limitations Our study had several limitations. First, the use of administrative data limited our ability to control for all clinical confounders. Nonetheless, we doubt whether unmeasured confounders differed over time between closure and nonclosure HSAs in ways that would have meaningfully biased our results. Second, because we used Medicare data, our
Exhibit 1 Baseline Characteristics Of Hospitals That Closed Versus All Other Open Hospitals, 2010 Hospitals Characteristic
Closed
Open
Average number of beds
63.6
93.8
Size Small Medium Large
47.3% 52.2 0.5
57.3% 35.8 6.9
Region Northeast Midwest South West
23.7 17.2 41.9 17.2
16.1 33.9 30.5 19.4
Ownership For-profit Private nonprofit Public
42.2 49.2 8.6
25.1 52.5 22.4
Rural-urban commuting area Urban Suburban Large rural town Small town or isolated rural
70.6 6.7 8.2 14.4
45.0 4.9 16.8 33.3
25.1 12.5 32.8 47.9 16.3
19.3 29.3 17.0 49.7 17.2
Teaching hospital Critical access hospital Safety-net hospital Medicare patients Medicaid patients Margins (percent) Totala Operatingb
−13 −20
Cost (percent of total revenue) Total state and local indigent care programc Total CHIPc Total gross Medicaidc Uncompensated carec
1.6% 0.2 10.4 11.9
4 −1 1.3% 0.2 9.6 7.1
SOURCE Authors’ analysis of data from the American Hospital Association and Medicare cost reports. NOTES Of the hospitals, 195 (4.3 percent) closed, and 4,335 (95.7 percent) remained open. Costs are reported by the hospitals and shown as a percentage of total hospital revenue. All differences are significant (p
