At the Intersection of Health, Health Care and Policy Cite this article as: Arnold M. Epstein, Ashish K. Jha, E. John Orav, Daniel L. Liebman, Anne-Marie J. Audet, Mark A. Zezza and Stuart Guterman Analysis Of Early Accountable Care Organizations Defines Patient, Structural, Cost, And Quality-Of-Care Characteristics Health Affairs, 33, no.1 (2014):95-102 doi: 10.1377/hlthaff.2013.1063
The online version of this article, along with updated information and services, is available at: http://content.healthaffairs.org/content/33/1/95.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 © 2014 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 March 10, 2015 at Kungl. Tekniska högskolan
Accountable Care Organizations By Arnold M. Epstein, Ashish K. Jha, E. John Orav, Daniel L. Liebman, Anne-Marie J. Audet, Mark A. Zezza, and Stuart Guterman
T H E CA R E S PAN
Analysis Of Early Accountable Care Organizations Defines Patient, Structural, Cost, And Quality-Of-Care Characteristics Accountable care organizations (ACOs) have attracted interest from many policy makers and clinical leaders because of their potential to improve the quality of care and reduce costs. Federal ACO programs for Medicare beneficiaries are now up and running, but little information is available about the baseline characteristics of early entrants. In this descriptive study we present data on the structural and market characteristics of these early ACOs and compare ACOs’ patient populations, costs, and quality with those of their non-ACO counterparts at baseline. We found that ACO patients were more likely than non-ACO patients to be older than age eighty and had higher incomes. ACO patients were less likely than non-ACO patients to be black, covered by Medicaid, or disabled. The cost of care for ACO patients was slightly lower than that for non-ACO patients. Slightly fewer than half of the ACOs had a participating hospital. Hospitals that were in ACOs were more likely than non-ACO hospitals to be large, teaching, and not-forprofit, although there was little difference in their performance on quality metrics. Our findings can be useful in interpreting the early results from the federal ACO programs and in establishing a baseline to assess the programs’ development. ABSTRACT
I
mproving the quality of health care while curbing its cost remains an elusive goal for policy makers. The Affordable Care Act established a series of new programs intended to improve the structure of the US health care delivery system and address this goal. Among the most prominent of these new programs are two federal initiatives creating accountable care organizations (ACOs), in which various constellations of providers agree to assume collective responsibility for the care delivered to a defined Medicare population. In return, these providers will share in any savings and
(potentially) any overages in the cost of their care compared to historical benchmarks. These providers will also be responsible for meeting a set of quality standards. The appeal of ACOs is partly rooted in the shortcomings of current payment models. Historically, providers have been reimbursed either through fee-for-service payments, an approach that raises concerns about overuse leading to higher spending and unnecessary services,1,2 or through full capitation, which raises concerns about underuse and withholding appropriate care.3,4 The shared savings model and requirements for quality of care adopted for ACOs are J A N U A RY 201 4
3 3: 1
Downloaded from content.healthaffairs.org by Health Affairs on March 10, 2015 at Kungl. Tekniska högskolan
10.1377/hlthaff.2013.1063 HEALTH AFFAIRS 33, NO. 1 (2014): 95–102 ©2014 Project HOPE— The People-to-People Health Foundation, Inc.
doi:
Arnold M. Epstein (aepstein@ hsph.harvard.edu) is the John H. Foster Professor of Health Policy and Management, Harvard School of Public Health, in Boston, Massachusetts. Ashish K. Jha is a professor of health policy and management, Harvard School of Public Health. E. John Orav is an associate professor of medicine (biostatistics) at Harvard Medical School. Daniel L. Liebman is a research assistant in the Department of Health Policy and Management, Harvard School of Public Health. Anne-Marie J. Audet is vice president for delivery system reform and breakthrough healthcare opportunities at the Commonwealth Fund, in New York City. Mark A. Zezza is assistant vice president for delivery system reform and cost control at the Commonwealth Fund. Stuart Guterman is vice president for Medicare and cost control at the Commonwealth Fund.
Health Affai rs
95
Accountable Care Organizations an attempt to address both of these concerns.5 There is already strong enthusiasm for ACOs by both policy makers and some clinical leaders. However, the programs are still in their early days. Only 116 health care organizations had signed up for Medicare’s Shared Savings Program as of July 2012, and only 32 had joined Medicare’s Pioneer ACO Initiative as of December 2011.6 In January 2013 the Centers for Medicare and Medicaid Services announced that an additional 106 organizations had joined the Shared Savings Program.7 This rate of uptake is relatively brisk for new programs, but only a small proportion of health care providers in the United States are participating in ACOs.Whether the programs will grow in a robust way will probably depend on the experiences of these early entrants. Moreover, these early experiences will likely guide policy makers as they modify ACO regulations to address any concerns that arise. Despite the importance of timely information on the characteristics and performance of early ACO entrants, very little is known about them. In this study we sought to identify all ACOs that had joined the federal programs by January 2013 and to answer three questions. First, what are the characteristics of patient populations cared for by ACOs, hospitals participating in ACOs, and markets in which ACOs operate, and how do these characteristics compare to those of nonACO populations, hospitals, and markets? Second, what are the baseline quality and cost of care of the ACOs, and how do they compare to those of non-ACO organizations? And third, how do the patient populations and cost of care vary in different types of ACOs—specifically, in ACOs with a hospital participant compared to ACOs without one?
Study Data And Methods Study Sample We obtained publicly available lists from the Centers for Medicare and Medicaid Services (CMS) of the 254 ACOs that had joined the Pioneer ACO Initiative and the Shared Savings Program through January 2013.We used the ZIP codes of the ACOs’ headquarters on their websites and the 2012 and 2013 editions of the HealthQuest Accountable Care Directory to verify the location of the ACOs’ headquarters. ACOs are defined as a collection of one or more “ACO participants.” A “participant” can be a hospital, a community health center, a group or solo physician practice, or any other entity that bills Medicare for services under a unique federal tax ID number or a CMS Certification Number. Starting in the fall of 2012 we attempted to obtain a list of each ACO’s participants. We used 96
Health Affairs
J A N U A RY 2 0 1 4
33:1
phone and e-mail to try to contact responsible officials at each of the 148 ACOs that had signed up by that time for one of the two federal programs. When we were unable to obtain this information through direct outreach, we attempted to identify ACO participants using the ACOs’ websites. We obtained sixty-two ACO participant lists directly from ACO officials and sixty-one lists online, for a total sample of 5,442 participants in 123 ACOs (83 percent of the 148 organizations that we originally targeted—the cohort that had entered the programs by July 2012). It was necessary for us to obtain ACO participants’ Medicare-enrolled tax ID numbers to identify their service claims in our data. We obtained a total of 4,793 Medicare-enrolled ACO tax ID numbers (88 percent of the 5,442 participants we had identified). We used the numbers obtained directly from ACOs when such information was provided, and we identified the remainder through online databases or by linking ACO providers’ National Provider Identifier numbers with tax ID numbers on their Medicare claims data. For 12 percent of the participants, neither of these approaches was successful (for details on the tax ID number acquisition process, see the online Appendix).8 It was also necessary for us to identify CMS Certification Numbers for federally qualified health centers in our sample, because these participants are enrolled in ACOs via those numbers instead of tax ID numbers. We were able to identify CMS Certification Numbers for forty-five of the forty-nine federally qualified health centers in our sample (92 percent), using CMS’s online National Plan and Provider Enumeration System. Attributing Patients To ACOs We analyzed 2011 Medicare claims for a random 5 percent sample of all Medicare beneficiaries enrolled in the traditional fee-for-service program who were continuously enrolled in Parts A and B and who received at least one primary care service during the year. We followed the published CMS Shared Savings Program assignment algorithm to assign each beneficiary to the provider group that accounted for the most allowed charges for evaluation and management services among all groups providing those services to the beneficiary.9 Patient Characteristics And Cost Of Care We used data from Medicare enrollment files and the US census to characterize patients’ age; sex; race or ethnicity; Medicaid status; and income, as defined by the median income level in their ZIP code of residence. We obtained information on comorbidities from diagnosis codes on Parts A and B Medicare claims, tabulating whether or
Downloaded from content.healthaffairs.org by Health Affairs on March 10, 2015 at Kungl. Tekniska högskolan
not patients had any of the comorbidities listed in the CMS Hierarchical Condition Category mortality models. We also tabulated whether patients had any of the thirty Elixhauser comorbidities.10 We tabulated standardized costs for inpatient and outpatient care using the following Medicare claims files: the inpatient file; the carrier file, which reports on professional services; and the outpatient file. In tabulating costs for inpatient services, we began with the amount paid by Medicare for each hospitalization. We adjusted this amount for variations in input prices (using the Medicare Wage Index) and for graduate medical education and disproportionateshare payments. We used published Medicare fee schedules to assign standardized Medicare costs to each outpatient and carrier file service, regardless of the actual amount that Medicare paid for each service (for details on the methods for standardizing costs, see the online Appendix).8 The use of standardized costs allowed us to identify patients who used a comparable amount of medical care across areas of the country in which actual spending may have varied substantially. Hospital Characteristics We used data from the American Hospital Association’s 2011 Annual Survey of Hospitals to characterize hospitals participating and not participating in ACOs according to bed size, regional location, profit status, and teaching status. To assess each hospital’s performance on process quality metrics, we used data from Medicare’s Hospital Compare database on Hospital Quality Alliance metrics that reflected process quality performance during 2011 on four conditions: acute myocardial infarction, congestive heart failure, pneumonia, and surgical care. These data are based on a standard methodology prescribed by the Joint Commission and other organizations to calculate condition-specific summary scores. Each hospital’s condition-specific summary score was calculated as the number of times that “appropriate” care was given across all the measures for that condition, divided by the number of times patients were eligible for that care. The specific indicators involved are listed in Appendix Exhibit A1.8 To assess outcome measures reflecting quality of care, we examined rates of risk-adjusted mortality for acute myocardial infarction, congestive heart failure, and pneumonia. Rates were adjusted by age, sex, and Elixhauser comorbidities— a list developed by the Agency for Healthcare Research and Quality and used widely with administrative data.10 Finally, we examined rates of readmission for
the same three conditions, again adjusting for age, sex, and Elixhauser comorbidities. For each hospital, we calculated expected conditionspecific odds of thirty-day readmission and multiplied the hospital’s observed-to-expected ratio by the overall readmission rate from our national sample for each specific condition. Market Characteristics We compared market characteristics of Hospital Referral Regions (HRRs) containing one or more headquarters of ACOs that had joined one of the two federal programs with the characteristics of HRRs containing no such headquarters.We examined population size, competition (using the HerfindahlHirschman Index, which measures hospital competition based on Medicare discharges), beds per 1,000 population, and primary care physicians and specialist physicians per 100,000 population. The data sources for these characteristics were the US census and the Dartmouth Atlas of Health Care.11 We also tabulated the Prevention Quality Indicators (as measured by hospitalizations for ambulatory care–sensitive conditions) compiled by the Agency for Healthcare Research and Quality.12 Data Analysis We first plotted the geographic location of headquarters of ACOs that had joined the Pioneer ACO Initiative or Shared Savings Program as of January 2013. We then compared the demographic, socioeconomic, and clinical characteristics of patients cared for in ACOs and their cost of care in 2011 with those data for other beneficiaries in the Medicare fee-forservice system. To gauge whether differences were related to differences in the markets where ACOs were located, we randomly matched each ACO patient to a non-ACO patient residing in the same HRR. Then we repeated our analyses comparing the demographic, socioeconomic, and clinical characteristics of patients cared for in ACOs and their cost of care in 2011 with those data for non-ACO beneficiaries in the Medicare fee-for-service system in the same region. We examined the structural characteristics and quality performance of hospitals participating in ACOs, compared to those of hospitals not affiliated with an ACO. We also examined the market characteristics of HRRs containing one or more headquarters of ACOs participating in one of the federal programs, compared to the characteristics of regions without any ACO headquarters. Finally, we compared the characteristics of patients in ACOs that included an acute care hospital with those of patients in ACOs without such a hospital. Limitations As noted above, we were able to identify the participants in only 83 percent of JANUARY 2014
33:1
Downloaded from content.healthaffairs.org by Health Affairs on March 10, 2015 at Kungl. Tekniska högskolan
Health A ffairs
97
Accountable Care Organizations the ACOs joining the federal programs. However, that percentage compares favorably with the samples in most studies based on survey data. The composition of ACOs is fluid, with some participants joining or leaving over time. Thus, some of the ACO participant lists we were able to obtain likely had some inaccuracies by the end of our data collection and analyses. Our data were from the most recent lists available from the ACOs when we conducted our study. We searched for tax ID numbers using 2009 data, the most recent data available to us at the time. Therefore, some of the determinations might not fully and accurately reflect presentday billing. We treated Pioneer and Shared Savings Program ACOs consistently. However, they differ in how CMS assigns patients to the ACOs and how CMS considers providers within each enrolled participant. For instance, Pioneer ACOs identify a subset of providers within a large participant and are assigned patients prospectively. In contrast, the Shared Savings Program includes all providers within a participant and assigns patients retrospectively based on their use of services. Our data included all providers at a given participant organization as if it were a Shared Sav-
ings Program ACO.We were not able to obtain or confirm tax ID numbers for all ACO participants, although we did have numbers for nearly 90 percent of the participants in the study. The claims data that we used to risk-adjust outcomes had limited information on comorbidities. When we lacked the list of an ACO’s participants, we were not able to identify the entire ACO’s patient population. Thus, we could identify the number of ACOs in different HRRs but not the number and proportion of people enrolled in them. Finally, our analyses are merely descriptive and cannot be used to suggest causality.
Study Results ACOs were more heavily concentrated in the South than elsewhere in the United States, and the regional distribution did not change substantially between 2012 and 2013 (Exhibit 1). Thirtyfour percent of the ACOs joining one of the two federal programs by 2012 were in the South, versus 41 percent of those that joined in 2013. In the Midwest, 22 percent and 20 percent joined in 2012 and 2013, respectively, compared to 27 percent and 22 percent in the Northeast and 17 percent and 18 percent in the West.
Exhibit 1 Accountable Care Organization (ACO) Locations In The United States, 2013
SOURCE Authors’ analysis. NOTES The locations were determined by the ZIP code of the ACO’s headquarters. “Joined” means joined either the Pioneer ACO Initiative or the Shared Savings Program. Alaska and Hawaii contained no ACOs and are not depicted on this map.
98
H ea lt h A f fai r s
JANUARY 2014
33:1
Downloaded from content.healthaffairs.org by Health Affairs on March 10, 2015 at Kungl. Tekniska högskolan
As noted above, we were able to identify provider participants at 123 of the 148 ACOs entering one of the programs in 2012. Their geographic distribution was similar to that of the overall cohort of ACOs entering in 2012: 35 percent of the participants were in the South, 24 percent in the Midwest, 28 percent in the Northeast, and 14 percent in the West. Patients Compared to non-ACO patients, ACO patients were less likely to be younger than age sixty-five and more likely to be older than age eighty (Exhibit 2). ACO patients were also less likely to be black or eligible for Medicaid, and they had higher incomes. Differences in comorbidities between patients cared for in Medicare ACOs and other Medicare beneficiaries were small (Exhibit 2). However, differences in the prevalence of several clinical conditions were statistically significant. Compared to non-ACO patients, ACO patients had 5.8 percent lower total cost of care ($7,694 versus $8,164). Both hospital costs and nonhospital costs were lower for ACO patients. We found that differences between ACO patients and non-ACO patients were similar when we compared the characteristics of patients in ACOs to those of non-ACO patients residing in the HRRs that included one or more ACOs (Appendix Exhibit A2).8 Hospitals Fifty-six of the 123 ACOs whose provider participants we identified (46 percent) had one or more hospitals among their participants. Compared to other hospitals, those participating in an ACO were significantly more likely to be large, located in the Northeast and Midwest, not for profit, and in an urban area (Exhibit 3). Hospitals in ACOs were also significantly more likely to be teaching hospitals and to have a medical or cardiac intensive care unit. The quality of care at the hospitals participating in ACOs was similar to that at nonparticipating hospitals (Exhibit 3). Markets HRRs with one or more ACOs headquartered in them had larger average populations; more competition, based on the Herfindahl-Hirschman Index; and higher total Medicare spending per beneficiary, compared to regions that did not contain an ACO headquarters (Exhibit 4). HRRs containing ACO headquarters also had fewer beds per 1,000 population but more primary care physicians and specialist physicians per 100,000 population. Outpatient Medicare spending per beneficiary was slightly higher overall in HRRs with ACO headquarters than in regions without headquarters. The same was true of hospital spending, but there was no difference in the rate of hospitalization for ambulatory care–sensitive conditions, based on Prevention Quality Indicators.12
Exhibit 2 Demographic, Socioeconomic, And Clinical Characteristics Of Accountable Care Organization (ACO) Patients And Non-ACO Patients, 2011 ACO (n=85,712)
Non-ACO (n=1,338,677)
p value
73.5a 13.3% 39.0 16.6 31.0
72.5b 16.4% 38.8 16.2 28.7
The online version of this article, along with updated information and services, is available at: http://content.healthaffairs.org/content/33/1/95.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 © 2014 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 March 10, 2015 at Kungl. Tekniska högskolan
Accountable Care Organizations By Arnold M. Epstein, Ashish K. Jha, E. John Orav, Daniel L. Liebman, Anne-Marie J. Audet, Mark A. Zezza, and Stuart Guterman
T H E CA R E S PAN
Analysis Of Early Accountable Care Organizations Defines Patient, Structural, Cost, And Quality-Of-Care Characteristics Accountable care organizations (ACOs) have attracted interest from many policy makers and clinical leaders because of their potential to improve the quality of care and reduce costs. Federal ACO programs for Medicare beneficiaries are now up and running, but little information is available about the baseline characteristics of early entrants. In this descriptive study we present data on the structural and market characteristics of these early ACOs and compare ACOs’ patient populations, costs, and quality with those of their non-ACO counterparts at baseline. We found that ACO patients were more likely than non-ACO patients to be older than age eighty and had higher incomes. ACO patients were less likely than non-ACO patients to be black, covered by Medicaid, or disabled. The cost of care for ACO patients was slightly lower than that for non-ACO patients. Slightly fewer than half of the ACOs had a participating hospital. Hospitals that were in ACOs were more likely than non-ACO hospitals to be large, teaching, and not-forprofit, although there was little difference in their performance on quality metrics. Our findings can be useful in interpreting the early results from the federal ACO programs and in establishing a baseline to assess the programs’ development. ABSTRACT
I
mproving the quality of health care while curbing its cost remains an elusive goal for policy makers. The Affordable Care Act established a series of new programs intended to improve the structure of the US health care delivery system and address this goal. Among the most prominent of these new programs are two federal initiatives creating accountable care organizations (ACOs), in which various constellations of providers agree to assume collective responsibility for the care delivered to a defined Medicare population. In return, these providers will share in any savings and
(potentially) any overages in the cost of their care compared to historical benchmarks. These providers will also be responsible for meeting a set of quality standards. The appeal of ACOs is partly rooted in the shortcomings of current payment models. Historically, providers have been reimbursed either through fee-for-service payments, an approach that raises concerns about overuse leading to higher spending and unnecessary services,1,2 or through full capitation, which raises concerns about underuse and withholding appropriate care.3,4 The shared savings model and requirements for quality of care adopted for ACOs are J A N U A RY 201 4
3 3: 1
Downloaded from content.healthaffairs.org by Health Affairs on March 10, 2015 at Kungl. Tekniska högskolan
10.1377/hlthaff.2013.1063 HEALTH AFFAIRS 33, NO. 1 (2014): 95–102 ©2014 Project HOPE— The People-to-People Health Foundation, Inc.
doi:
Arnold M. Epstein (aepstein@ hsph.harvard.edu) is the John H. Foster Professor of Health Policy and Management, Harvard School of Public Health, in Boston, Massachusetts. Ashish K. Jha is a professor of health policy and management, Harvard School of Public Health. E. John Orav is an associate professor of medicine (biostatistics) at Harvard Medical School. Daniel L. Liebman is a research assistant in the Department of Health Policy and Management, Harvard School of Public Health. Anne-Marie J. Audet is vice president for delivery system reform and breakthrough healthcare opportunities at the Commonwealth Fund, in New York City. Mark A. Zezza is assistant vice president for delivery system reform and cost control at the Commonwealth Fund. Stuart Guterman is vice president for Medicare and cost control at the Commonwealth Fund.
Health Affai rs
95
Accountable Care Organizations an attempt to address both of these concerns.5 There is already strong enthusiasm for ACOs by both policy makers and some clinical leaders. However, the programs are still in their early days. Only 116 health care organizations had signed up for Medicare’s Shared Savings Program as of July 2012, and only 32 had joined Medicare’s Pioneer ACO Initiative as of December 2011.6 In January 2013 the Centers for Medicare and Medicaid Services announced that an additional 106 organizations had joined the Shared Savings Program.7 This rate of uptake is relatively brisk for new programs, but only a small proportion of health care providers in the United States are participating in ACOs.Whether the programs will grow in a robust way will probably depend on the experiences of these early entrants. Moreover, these early experiences will likely guide policy makers as they modify ACO regulations to address any concerns that arise. Despite the importance of timely information on the characteristics and performance of early ACO entrants, very little is known about them. In this study we sought to identify all ACOs that had joined the federal programs by January 2013 and to answer three questions. First, what are the characteristics of patient populations cared for by ACOs, hospitals participating in ACOs, and markets in which ACOs operate, and how do these characteristics compare to those of nonACO populations, hospitals, and markets? Second, what are the baseline quality and cost of care of the ACOs, and how do they compare to those of non-ACO organizations? And third, how do the patient populations and cost of care vary in different types of ACOs—specifically, in ACOs with a hospital participant compared to ACOs without one?
Study Data And Methods Study Sample We obtained publicly available lists from the Centers for Medicare and Medicaid Services (CMS) of the 254 ACOs that had joined the Pioneer ACO Initiative and the Shared Savings Program through January 2013.We used the ZIP codes of the ACOs’ headquarters on their websites and the 2012 and 2013 editions of the HealthQuest Accountable Care Directory to verify the location of the ACOs’ headquarters. ACOs are defined as a collection of one or more “ACO participants.” A “participant” can be a hospital, a community health center, a group or solo physician practice, or any other entity that bills Medicare for services under a unique federal tax ID number or a CMS Certification Number. Starting in the fall of 2012 we attempted to obtain a list of each ACO’s participants. We used 96
Health Affairs
J A N U A RY 2 0 1 4
33:1
phone and e-mail to try to contact responsible officials at each of the 148 ACOs that had signed up by that time for one of the two federal programs. When we were unable to obtain this information through direct outreach, we attempted to identify ACO participants using the ACOs’ websites. We obtained sixty-two ACO participant lists directly from ACO officials and sixty-one lists online, for a total sample of 5,442 participants in 123 ACOs (83 percent of the 148 organizations that we originally targeted—the cohort that had entered the programs by July 2012). It was necessary for us to obtain ACO participants’ Medicare-enrolled tax ID numbers to identify their service claims in our data. We obtained a total of 4,793 Medicare-enrolled ACO tax ID numbers (88 percent of the 5,442 participants we had identified). We used the numbers obtained directly from ACOs when such information was provided, and we identified the remainder through online databases or by linking ACO providers’ National Provider Identifier numbers with tax ID numbers on their Medicare claims data. For 12 percent of the participants, neither of these approaches was successful (for details on the tax ID number acquisition process, see the online Appendix).8 It was also necessary for us to identify CMS Certification Numbers for federally qualified health centers in our sample, because these participants are enrolled in ACOs via those numbers instead of tax ID numbers. We were able to identify CMS Certification Numbers for forty-five of the forty-nine federally qualified health centers in our sample (92 percent), using CMS’s online National Plan and Provider Enumeration System. Attributing Patients To ACOs We analyzed 2011 Medicare claims for a random 5 percent sample of all Medicare beneficiaries enrolled in the traditional fee-for-service program who were continuously enrolled in Parts A and B and who received at least one primary care service during the year. We followed the published CMS Shared Savings Program assignment algorithm to assign each beneficiary to the provider group that accounted for the most allowed charges for evaluation and management services among all groups providing those services to the beneficiary.9 Patient Characteristics And Cost Of Care We used data from Medicare enrollment files and the US census to characterize patients’ age; sex; race or ethnicity; Medicaid status; and income, as defined by the median income level in their ZIP code of residence. We obtained information on comorbidities from diagnosis codes on Parts A and B Medicare claims, tabulating whether or
Downloaded from content.healthaffairs.org by Health Affairs on March 10, 2015 at Kungl. Tekniska högskolan
not patients had any of the comorbidities listed in the CMS Hierarchical Condition Category mortality models. We also tabulated whether patients had any of the thirty Elixhauser comorbidities.10 We tabulated standardized costs for inpatient and outpatient care using the following Medicare claims files: the inpatient file; the carrier file, which reports on professional services; and the outpatient file. In tabulating costs for inpatient services, we began with the amount paid by Medicare for each hospitalization. We adjusted this amount for variations in input prices (using the Medicare Wage Index) and for graduate medical education and disproportionateshare payments. We used published Medicare fee schedules to assign standardized Medicare costs to each outpatient and carrier file service, regardless of the actual amount that Medicare paid for each service (for details on the methods for standardizing costs, see the online Appendix).8 The use of standardized costs allowed us to identify patients who used a comparable amount of medical care across areas of the country in which actual spending may have varied substantially. Hospital Characteristics We used data from the American Hospital Association’s 2011 Annual Survey of Hospitals to characterize hospitals participating and not participating in ACOs according to bed size, regional location, profit status, and teaching status. To assess each hospital’s performance on process quality metrics, we used data from Medicare’s Hospital Compare database on Hospital Quality Alliance metrics that reflected process quality performance during 2011 on four conditions: acute myocardial infarction, congestive heart failure, pneumonia, and surgical care. These data are based on a standard methodology prescribed by the Joint Commission and other organizations to calculate condition-specific summary scores. Each hospital’s condition-specific summary score was calculated as the number of times that “appropriate” care was given across all the measures for that condition, divided by the number of times patients were eligible for that care. The specific indicators involved are listed in Appendix Exhibit A1.8 To assess outcome measures reflecting quality of care, we examined rates of risk-adjusted mortality for acute myocardial infarction, congestive heart failure, and pneumonia. Rates were adjusted by age, sex, and Elixhauser comorbidities— a list developed by the Agency for Healthcare Research and Quality and used widely with administrative data.10 Finally, we examined rates of readmission for
the same three conditions, again adjusting for age, sex, and Elixhauser comorbidities. For each hospital, we calculated expected conditionspecific odds of thirty-day readmission and multiplied the hospital’s observed-to-expected ratio by the overall readmission rate from our national sample for each specific condition. Market Characteristics We compared market characteristics of Hospital Referral Regions (HRRs) containing one or more headquarters of ACOs that had joined one of the two federal programs with the characteristics of HRRs containing no such headquarters.We examined population size, competition (using the HerfindahlHirschman Index, which measures hospital competition based on Medicare discharges), beds per 1,000 population, and primary care physicians and specialist physicians per 100,000 population. The data sources for these characteristics were the US census and the Dartmouth Atlas of Health Care.11 We also tabulated the Prevention Quality Indicators (as measured by hospitalizations for ambulatory care–sensitive conditions) compiled by the Agency for Healthcare Research and Quality.12 Data Analysis We first plotted the geographic location of headquarters of ACOs that had joined the Pioneer ACO Initiative or Shared Savings Program as of January 2013. We then compared the demographic, socioeconomic, and clinical characteristics of patients cared for in ACOs and their cost of care in 2011 with those data for other beneficiaries in the Medicare fee-forservice system. To gauge whether differences were related to differences in the markets where ACOs were located, we randomly matched each ACO patient to a non-ACO patient residing in the same HRR. Then we repeated our analyses comparing the demographic, socioeconomic, and clinical characteristics of patients cared for in ACOs and their cost of care in 2011 with those data for non-ACO beneficiaries in the Medicare fee-for-service system in the same region. We examined the structural characteristics and quality performance of hospitals participating in ACOs, compared to those of hospitals not affiliated with an ACO. We also examined the market characteristics of HRRs containing one or more headquarters of ACOs participating in one of the federal programs, compared to the characteristics of regions without any ACO headquarters. Finally, we compared the characteristics of patients in ACOs that included an acute care hospital with those of patients in ACOs without such a hospital. Limitations As noted above, we were able to identify the participants in only 83 percent of JANUARY 2014
33:1
Downloaded from content.healthaffairs.org by Health Affairs on March 10, 2015 at Kungl. Tekniska högskolan
Health A ffairs
97
Accountable Care Organizations the ACOs joining the federal programs. However, that percentage compares favorably with the samples in most studies based on survey data. The composition of ACOs is fluid, with some participants joining or leaving over time. Thus, some of the ACO participant lists we were able to obtain likely had some inaccuracies by the end of our data collection and analyses. Our data were from the most recent lists available from the ACOs when we conducted our study. We searched for tax ID numbers using 2009 data, the most recent data available to us at the time. Therefore, some of the determinations might not fully and accurately reflect presentday billing. We treated Pioneer and Shared Savings Program ACOs consistently. However, they differ in how CMS assigns patients to the ACOs and how CMS considers providers within each enrolled participant. For instance, Pioneer ACOs identify a subset of providers within a large participant and are assigned patients prospectively. In contrast, the Shared Savings Program includes all providers within a participant and assigns patients retrospectively based on their use of services. Our data included all providers at a given participant organization as if it were a Shared Sav-
ings Program ACO.We were not able to obtain or confirm tax ID numbers for all ACO participants, although we did have numbers for nearly 90 percent of the participants in the study. The claims data that we used to risk-adjust outcomes had limited information on comorbidities. When we lacked the list of an ACO’s participants, we were not able to identify the entire ACO’s patient population. Thus, we could identify the number of ACOs in different HRRs but not the number and proportion of people enrolled in them. Finally, our analyses are merely descriptive and cannot be used to suggest causality.
Study Results ACOs were more heavily concentrated in the South than elsewhere in the United States, and the regional distribution did not change substantially between 2012 and 2013 (Exhibit 1). Thirtyfour percent of the ACOs joining one of the two federal programs by 2012 were in the South, versus 41 percent of those that joined in 2013. In the Midwest, 22 percent and 20 percent joined in 2012 and 2013, respectively, compared to 27 percent and 22 percent in the Northeast and 17 percent and 18 percent in the West.
Exhibit 1 Accountable Care Organization (ACO) Locations In The United States, 2013
SOURCE Authors’ analysis. NOTES The locations were determined by the ZIP code of the ACO’s headquarters. “Joined” means joined either the Pioneer ACO Initiative or the Shared Savings Program. Alaska and Hawaii contained no ACOs and are not depicted on this map.
98
H ea lt h A f fai r s
JANUARY 2014
33:1
Downloaded from content.healthaffairs.org by Health Affairs on March 10, 2015 at Kungl. Tekniska högskolan
As noted above, we were able to identify provider participants at 123 of the 148 ACOs entering one of the programs in 2012. Their geographic distribution was similar to that of the overall cohort of ACOs entering in 2012: 35 percent of the participants were in the South, 24 percent in the Midwest, 28 percent in the Northeast, and 14 percent in the West. Patients Compared to non-ACO patients, ACO patients were less likely to be younger than age sixty-five and more likely to be older than age eighty (Exhibit 2). ACO patients were also less likely to be black or eligible for Medicaid, and they had higher incomes. Differences in comorbidities between patients cared for in Medicare ACOs and other Medicare beneficiaries were small (Exhibit 2). However, differences in the prevalence of several clinical conditions were statistically significant. Compared to non-ACO patients, ACO patients had 5.8 percent lower total cost of care ($7,694 versus $8,164). Both hospital costs and nonhospital costs were lower for ACO patients. We found that differences between ACO patients and non-ACO patients were similar when we compared the characteristics of patients in ACOs to those of non-ACO patients residing in the HRRs that included one or more ACOs (Appendix Exhibit A2).8 Hospitals Fifty-six of the 123 ACOs whose provider participants we identified (46 percent) had one or more hospitals among their participants. Compared to other hospitals, those participating in an ACO were significantly more likely to be large, located in the Northeast and Midwest, not for profit, and in an urban area (Exhibit 3). Hospitals in ACOs were also significantly more likely to be teaching hospitals and to have a medical or cardiac intensive care unit. The quality of care at the hospitals participating in ACOs was similar to that at nonparticipating hospitals (Exhibit 3). Markets HRRs with one or more ACOs headquartered in them had larger average populations; more competition, based on the Herfindahl-Hirschman Index; and higher total Medicare spending per beneficiary, compared to regions that did not contain an ACO headquarters (Exhibit 4). HRRs containing ACO headquarters also had fewer beds per 1,000 population but more primary care physicians and specialist physicians per 100,000 population. Outpatient Medicare spending per beneficiary was slightly higher overall in HRRs with ACO headquarters than in regions without headquarters. The same was true of hospital spending, but there was no difference in the rate of hospitalization for ambulatory care–sensitive conditions, based on Prevention Quality Indicators.12
Exhibit 2 Demographic, Socioeconomic, And Clinical Characteristics Of Accountable Care Organization (ACO) Patients And Non-ACO Patients, 2011 ACO (n=85,712)
Non-ACO (n=1,338,677)
p value
73.5a 13.3% 39.0 16.6 31.0
72.5b 16.4% 38.8 16.2 28.7
