Medicare’s Bundled Payment Initiative: Most Hospitals Are Focused On A Few High-Volume Conditions
- Thomas C. Tsai is a research associate in the Department of Health Policy and Management at the Harvard T.H. Chan School of Public Health and a general surgery resident in the Department of Surgery at Brigham and Women’s Hospital. He 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 (ASPE), Department of Health and Human Services, in Washington, D.C.
- Karen E. Joynt is currently serving as a senior adviser to the deputy assistant secretary for health policy, ASPE, Department of Health and Human Services. At the time that this article was prepared, she was an instructor of medicine in the Division of Cardiovascular Medicine at Brigham and Women’s Hospital and an instructor in the Department of Health Policy and Management, Harvard T.H. Chan School of Public Health.
- Robert C. Wild is a statistical programmer and analyst in the Department of Health Policy and Management, Harvard T.H. Chan School of Public Health.
- E. John Orav is an associate professor of biostatistics at the Harvard T.H. Chan School of Public Health.
- Ashish K. Jha ( [email protected] ) is the K.T. Li Professor of International Health at the Harvard T.H. Chan School of Public Health, in Boston, Massachusetts.
The Bundled Payments for Care Improvement initiative is a federally funded innovation model mandated by the Affordable Care Act. It is designed to help transition Medicare away from fee-for-service payments and toward bundling a single payment for an episode of acute care in a hospital and related postacute care in an appropriate setting. While results from the initiative will not be available for several years, current data can help provide critical early insights. However, little is known about the participating organizations and how they are focusing their efforts. We identified participating hospitals and used national Medicare claims data to assess their characteristics and previous spending patterns. These hospitals are mostly large, nonprofit, teaching hospitals in the Northeast, and they have selectively enrolled in the bundled payment initiative covering patient conditions with high clinical volumes. We found no significant differences in episode-based spending between participating and nonparticipating hospitals. Postacute care explains the largest variation in overall episode-based spending, signaling an opportunity to align incentives across providers. However, the focus on a few selected clinical conditions and the high degree of integration that already exists between enrolled hospitals and postacute care providers may limit the generalizability of bundled payment across the Medicare system.
Given the substantial and ongoing need to improve the value of US health care, policy makers are increasingly focused on alternative payment models for health care services. There is broad consensus that the current system of paying for care through fee-for-service reimbursement leads to overprovision of services, drives up costs, and fails to guarantee optimal patient outcomes. 1,2 Bundling payments around episodes of care, including acute hospitalization and postacute care, has been identified as a promising approach to making health care more efficient and of higher value. 1,3 The notion behind bundled payment is simple: Providers are paid a fixed amount for the expected cost of a set of related services. By receiving a target price for a defined episode of care and its associated services, providers have strong incentives to eliminate unnecessary spending while maintaining high quality of care.
The Affordable Care Act (ACA) emphasizes shifting fee-for-service Medicare payment models toward value-based models. 2 To that end, the ACA established the Center for Medicare and Medicaid Innovation within the Centers for Medicare and Medicaid Services (CMS). In January 2013 the Innovation Center launched the Bundled Payments for Care Improvement (BPCI) initiative, a new payment model that allows participating hospitals to enroll in bundled payment agreements with CMS for up to forty-eight predefined clinical conditions aggregated from Medicare severity diagnosis-related groups (MS-DRGs). These conditions include amputations, urinary tract infections, stroke, chronic obstructive pulmonary disease, coronary artery bypass graft surgery, joint replacements, percutaneous coronary intervention, cardiac and other major surgeries, diabetes, and respiratory conditions. It is hoped that bundled payment will lead to higher-quality, more coordinated care and lower costs to Medicare. The initiative’s performance period lasts for three years, with the possibility of a two-year extension. 4
While the cost and quality impacts of the BPCI initiative will not be available for at least several years, currently available data can help provide critical early insights into the program. Despite the importance of the bundled payment initiative, little is known about which organizations have joined, how these organizations differ from those that did not join, and where the participating organizations are focusing their efforts. CMS has posted the names of participating institutions on its website but provides little information on their structural features and on whether they differ on important characteristics such as clinical volume and spending patterns. Understanding these basic features can be immensely helpful to organizations considering joining the program as well as to policy makers looking to design effective proposals that appeal to a broad range of health care providers.
To address this knowledge gap, we analyzed recently released CMS data to answer three key questions. First, what types of hospitals have entered into risk-bearing agreements with CMS through the BPCI initiative, and how do they compare to nonparticipating hospitals? Second, for which clinical conditions did participating hospitals enroll, and are there particular features, such as high volume, that might explain why the conditions were chosen? And finally, for the clinical conditions chosen by participating hospitals, are there important differences in baseline spending between hospitals that are enrolled and not enrolled in the initiative, and, if there are differences, what might explain them?
Study Data And Methods
The Bundled Payments For Care Improvement Initiative
Under the BPCI initiative, hospitals can enroll in any of four different bundled payment models, each of which includes a different combination of acute and postacute services. Model 1 involves hospital payments for inpatient stays for most Medicare fee-for-service discharges. Model 2 includes all Medicare Parts A and B payments for the forty-eight selected clinical conditions (index hospitalization). Model 3 includes payments only for postacute claims for a participating skilled nursing facility, inpatient rehabilitation facility, long-term care hospital, home health agency, or physician practice group. And finally, model 4 has a more restricted definition of a bundle that includes payments for the initial hospitalization, health care provider fees during the hospitalization, and payments for readmissions but does not include other postacute care costs. (See online Appendix Exhibit A1 for details about the four models.) 5
Within each model, enrollment can occur in two phases: a non-risk-bearing preparation phase 1 and a risk-bearing phase 2. During phase 1, institutions or provider groups are given claims data and information on historical spending from CMS in order to inform the design of the bundled payments. Participants additionally have the option of enrolling in any of forty-eight different clinical conditions based on a thirty-, sixty-, or ninety-day episode.
For models 1–3, payments are retrospective, with actual payments reconciled against a target price. By contrast, in model 4, a lump-sum payment is made prospectively to the participating hospital, which then pays physicians and other health care providers from the bundled payment. Target payments for model 1 are set at a discount from the DRG payment, with no discount in the first six months, a discount of 0.5 percent in months seven to twelve, and a discount of 1.0 percent in years two and three. For model 2, target payments are set at a 3 percent discount for thirty-day and sixty-day bundles and 2 percent for ninety-day bundles based on participant-specific patterns of historical spending. Model 3 discounts are 3.00 percent, and model 4 discounts are either 3.00 or 3.25 percent, depending on previous participation in the Medicare Acute Care Episode demonstration. If actual payments in the risk-bearing phase are less than the target price, the participants would receive additional payments as an incentive bonus. If payments exceed the target price, the participants are responsible for refunding the excess to CMS as a penalty.
We focused our main analyses on model 2 because it represents the most comprehensive bundled payment by including all services following an index hospitalization. We identified acute care hospital participants in phase 2 of model 2 by using the April 2014 release of participants from CMS, which includes names of participating institutions as well as their phase of enrollment, enrolled clinical conditions, and episode length (thirty, sixty, or ninety days). 6 Participants can be any provider billing CMS, including acute care hospitals, physician groups, home health agencies, and skilled nursing facilities (see Appendix Exhibits A2–A5). 5 However, because all episodes begin with an index inpatient admission, we focused our analyses on acute care hospitals only.
Our primary interest was in hospitals that opted to participate in phase 2, the risk-bearing phase. We chose mutually exclusive definitions for the phases: Hospitals that were part of phase 1 (non-risk-bearing) only but chose not to move on to phase 2 were classified as phase 1 hospitals. Hospitals that were part of phase 2 (risk-bearing) were classified as phase 2 whether or not they had been part of phase 1. Thus, we had three main groups in our analytic sample: phase 2 participants (our main analytic sample), phase 1 participants, and nonparticipants. Nonparticipants included all remaining nonfederal, non-critical-access, and nonspecialty acute care hospitals in the United States.
We chose to focus our main analytic sample on model 2 phase 2 hospitals for several reasons. First, as phase 2 participants, these hospitals had entered into risk-bearing agreements with CMS, instead of only expressing interest in the preparation phase 1. Second, although bundled payment models are not yet well established in practice, model 2 allows for the broadest definition of an episode of care. By including all related Parts A and B payments, an analysis of model 2 would, therefore, highlight which component of spending results in the greatest variation and the greatest opportunity for savings through an episode-based bundled payment. For simplicity, we refer to our analytic sample of hospitals participating under model 2 of phase 2 simply as “phase 2 hospitals.”
We were interested in key structural features of hospitals that chose to participate in the BPCI initiative versus those that did not, including size, teaching status, ownership, region, membership in a network, affiliation with postacute care facilities or providers, percentage Medicare discharges, and percentage Medicaid discharges. These variables were each obtained from the 2011 American Hospital Association Annual Survey. The urban or rural location of hospitals was classified using rural-urban commuting area codes. These structural characteristics were included in the analyses because prior studies have suggested that they are associated with differences in hospital quality and payments.
Medicare Standardized Payments And Total Thirty-Day Episode Spending
Medicare payments for the conditions of interest were calculated for all eligible Medicare beneficiaries in 2011 using 20 percent Medicare inpatient, outpatient, carrier, skilled nursing facility, and home health agency claims files. We excluded beneficiaries younger than age sixty-five, those not continuously enrolled during the study period, and those enrolled in Medicare Advantage. Based on the exclusion criteria of the initiative, beneficiaries with end-stage renal disease were also excluded, given the existing CMS bundled payment for outpatient maintenance dialysis.
We followed a methodology similar to CMS’s for the definition of episodes of care. Each episode began with an “anchor” index hospitalization for a qualifying clinical condition predefined by an MS-DRG or group of MS-DRG codes. Episodes were then created by including all Medicare Parts A and B payments from the start of the index hospitalization until thirty days after discharge from the index hospitalization, as specified under model 2 agreements. Using lists of International Classification of Diseases , Ninth Revision (ICD-9), diagnosis codes that were unrelated to the forty-eight predefined clinical conditions, we excluded ineligible Parts A and B claims based on CMS’s methodology for the initiative. 7 For example, instead of using all-cause readmissions, we excluded readmissions deemed by CMS to be unrelated to the index hospitalization (and, therefore, by CMS’s approach, not eligible to be included in the bundle). As a result, our episode-based bundled payments were condition-specific and adhered very closely to the approach used by CMS.
We used standardized payments in order to allow for national-level comparisons across different types of hospitals, employing a previously described methodology 8 (see Appendix Methods A1). 5 Thirty-day total spending for an episode was calculated by aggregating payments across the five components of care: index hospitalization, postacute care services, readmissions, health care providers, and outpatient care from the corresponding Medicare claim files (inpatient, skilled nursing, home health, carrier, and outpatient). Postacute care standardized payments were derived from the inpatient (for inpatient rehabilitation stays), skilled nursing, and home health claims files. These episode-based spending patterns were created for all forty-eight eligible clinical conditions, although for simplicity we present our analysis from the five most commonly enrolled clinical conditions: major joint replacement of the lower extremity; congestive heart failure; chronic obstructive pulmonary disease, bronchitis, or asthma; simple pneumonia or respiratory infections; and hip and femur procedures excepting major joint replacement.
We first compared the key hospital characteristics of interest between our analytic sample and two comparison groups (phase 1 participants and acute care hospitals not participating in the initiative) using chi-square tests and t -tests for categorical and continuous variables, respectively. We then aggregated the total number of clinical conditions in which each hospital enrolled in phase 1 and phase 2, to illustrate the shifting clinical condition enrollment patterns between the non-risk-bearing phase 1 part of the program and the risk-bearing phase 2 participants.
Next, we used enrollment information from each participating hospital to create enrollment summaries for clinical conditions for our analytic sample of phase 2 participants. We then determined whether phase 2 participants entering risk-bearing agreements with Medicare selected conditions for which they had high clinical volumes. For the top five clinical conditions, we summarized the mean number of index hospitalizations for phase 2 participants and compared their volumes to those of other US hospitals using the 100 percent Medicare inpatient file for 2011. Hospitals were then grouped into quartiles, and we calculated the percentage of phase 2 hospitals that were in the highest quartile, by volume, of all US hospitals. To assess whether the enrolled clinical conditions represented a meaningful percentage of discharges for the phase 2 hospitals, we again used the 100 percent national Medicare inpatient file to tabulate the total number of Medicare discharges for each participating hospital and then calculated the proportion of total Medicare discharges accounted for by the top ten enrolled clinical conditions.
We next assessed patterns of spending for the enrolled clinical conditions among phase 2 hospitals and compared them with non–phase 2 hospitals. Total thirty-day episode spending, as well as the individual components of spending, was averaged across hospitals. We used analyses of variance to compare total and component spending between participants and nonparticipants, as well as between phase 1 and phase 2 participants. Finally, we assessed the proportion of variation in total payments for phase 1 and phase 2 hospitals attributable to each component of care, focusing on payments for index hospitalizations, readmissions, postacute care, health care providers (such as physicians), and outpatient care, as previously described. We again show results for the top five clinical conditions for simplicity of presentation. To assess the variance of thirty-day total costs that were attributed to each component of care, we used a general linear regression model with thirty-day total cost as the dependent variable and each component of care as independent variables. The percentage variation explained by each component was then obtained by removing each payment component sequentially and calculating the corresponding change in R-squared.
A two-tailed p value less than 0.05 was considered statistically significant. All analyses were performed using SAS software, version 9.3. This study was approved by the Harvard School of Public Health Office of Human Research and Administration.
Our study had several limitations. Because hospitals are only beginning to enter risk-bearing agreements with CMS, we were unable to evaluate the effectiveness of the BPCI initiative in terms of its impact on reducing costs or improving quality. Instead, we sought only to describe the baseline characteristics of the hospitals that enrolled. We focused primarily on acute care hospitals, and, accordingly, our findings may not be generalizable to bundled payments that did not include an anchor or index inpatient admission. Although the CMS initiative encompasses four different bundled payment models, for analytic purposes our study focused primarily on model 2, which was also the model in phase 2 that had the most hospital participants. Only forty-three acute care hospitals enrolled in any of the other three models during phase 2.
Finally, because the initiative allows for participating institutions to submit specific definitions of bundles based on the number of partnering facilities or physician practice groups, our analyses may not reflect the actual bundled payment specifications agreed on between individual institutions and CMS. However, by following the CMS methodology of excluding unrelated Medicare Parts A and B diagnoses from the bundles, we believe that our results closely match the calculation of bundles proposed by CMS and are largely representative of the overall patterns of spending.
Characteristics Of Enrolled Hospitals
We began with 3,360 non-critical-access, acute care hospitals in the United States that provide care to Medicare beneficiaries. Of these, 225 were BPCI initiative phase 1 participants, 107 were phase 2 participants, and 3,028 were categorized as non-BPCI participants.
There was a significant attrition of hospitals as well as nonhospital participants between phases 1 and 2 (see Appendix Exhibits A2–A5). 5 Hospitals enrolling in either phase 1 or phase 2 of the initiative were somewhat different than non-BPCI-enrolling hospitals ( Exhibit 1 ). Compared to their non-BPCI counterparts, phase 2 hospitals were larger (35.5 percent versus 11.6 percent), more likely to be in the Northeast (41.1 percent versus 13.4 percent), more likely to be nonprofit (82.2 percent versus 58.1 percent), more likely to be major teaching hospitals (27.1 percent versus 6.7 percent), more likely to be urban (93.5 percent versus 61.1 percent), and more likely to be members of a hospital system (74.8 percent versus 61.0 percent, all p values ). BPCI initiative phase 2 hospitals were more likely to exhibit greater affiliation with postacute care providers than were non-BPCI hospitals (78.5 percent versus 66.8 percent, ).
| BPCI hospitals ( N = 332) |
|Hospital characteristic||Phase 1 enrollees ( n = 225)||Phase 2, model 2 enrollees ( n = 107)||Non-BPCI hospitals ( n = 3028)|
|Small ( beds)||20.9%||3.7%||36.7%|
|Medium (100–399 beds)||59.1||60.8||51.7|
|Large ( beds)||20.0||35.5||11.6|
|Not a teaching hospital||55.6||37.4||72.4|
|Large rural town||11.1||5.6||20.2|
|Small rural town/isolated rural||1.8||0.9||14.1|
|Has intensive care unit||80.9||81.3||73.1|
|Member of a system||79.6||74.8||61.0|
|Affiliated with postacute care ***||78.2||78.5||66.8|
|Percent Medicare patients||46.4||45.7||47.2|
|Percent Medicaid patients||18.3||19.2||18.4|
Hospitals that enrolled in phase 2 differed significantly from their phase 1 counterparts in a few key characteristics ( Exhibit 1 ): Compared to phase 1 participants, phase 2 hospitals were more likely to be large (35.5 percent versus 20.0 percent, ), located in the Northeast (41.1 percent versus 23.6 percent, ), and designated as major teaching hospitals (27.1 percent versus 15.1 percent, ). There was a similar degree of affiliation with postacute care providers between phase 2 and phase 1 hospitals (78.5 percent versus 78.2 percent, ).
Enrolled Clinical Conditions For Phase 1 Versus Phase 2 Hospitals
In phase 1, the exploratory preparation period of the BPCI initiative, the overwhelming majority of hospitals (85 percent) enrolled for all forty-eight eligible clinical conditions (see Appendix Exhibit A6). 5 However, when hospitals enrolled in the risk-bearing portion (phase 2), there was a significant reduction in the number of enrolled conditions: 50 percent of phase 2 hospitals enrolled in only one clinical condition, and 72 percent of hospitals enrolled in three or fewer conditions (see Appendix Exhibit A7). 5 In contrast to phase 1 participants, no hospital entered risk-bearing agreements for all forty-eight conditions.
Among phase 2 participants, the five most common clinical conditions were major joint replacement of the lower extremity (72.9 percent); congestive heart failure (35.5 percent); chronic obstructive pulmonary disease, bronchitis, or asthma (25.2 percent); simple pneumonia or respiratory infections (20.6 percent); and hip and femur procedures excepting major joint replacement (18.7 percent) ( Exhibit 2 ).
|Model 2, phase 2 participants ( N = 107)|
|Clinical condition||No.||Percent||Average volume for participating hospitals (no. of discharges)||Average volume for all hospitals (no. of discharges)||Percent in highest quartile of volume||Percent of admissions for specified condition out of total Medicare discharges|
|Major joint replacement of the lower extremity ***||78||72.9||205.8||102.7||48.7||4.8|
|Congestive heart failure ***||38||35.5||140.0||72.7||55.3||3.1|
|Chronic obstructive pulmonary disease, bronchitis, asthma ***||27||25.2||109.6||71.7||48.2||2.8|
|Simple pneumonia and respiratory infections ***||22||20.6||153.0||86.3||54.6||3.4|
|Hip and femur procedures except major joint||20||18.7||39.5||31.3||40.0||1.0|
|Revision of the hip or knee||18||16.8||21.3||10.9||40.0||0.5|
|Acute myocardial infarction ***||17||15.9||45.6||24.9||58.8||1.0|
|Coronary artery bypass graft||17||15.9||38.9||31.6||41.2||0.6|
|Lower extremity and humerus procedure except hip, foot, and femur||14||13.1||11.9||9.8||35.7||0.3|
|Double joint replacement of the lower extremity **||13||12.1||17.5||5.6||62.5||0.3|
Hospital Volume And Baseline Costs Of Care
When we examined annual hospital volume for the enrolled conditions, we found that hospitals tended to be high volume for the conditions in which they enrolled. Focusing on the two most common enrolled conditions—major joint replacement of the lower extremity and congestive heart failure—hospitals enrolling in major joint replacement averaged 205.8 discharges, compared to the national average of 102.7 discharges ( Exhibit 2 ), while hospitals enrolling in congestive heart failure averaged 140.0 discharges, compared to a national average of 72.7 discharges. This pattern was consistent for all of the top ten enrolled conditions. Phase 2 hospitals were also more likely to be in the highest quartile of discharge volume nationally for the enrolled conditions, with 48.7 percent of hospitals enrolling in major joint replacement and 55.3 percent of hospitals enrolling in congestive heart failure in the highest quartile of volume nationally ( Exhibit 2 ). Again, patterns were consistent for all top ten conditions.
Because these hospitals were large hospitals, we assessed the percentage of total annual discharges accounted for by the enrolled conditions to determine whether these hospitals were high volume for these conditions specifically or just high volume overall. For hospitals enrolling in major joint replacement and congestive heart failure, the former on average constituted 4.8 percent of their total annual Medicare discharges, while the latter constituted 3.1 percent of their total annual Medicare discharges ( Exhibit 2 ).
Patterns Of Spending
When we examined baseline costs of care for the two most common enrolled conditions, we found no significant difference in payments among phase 2, phase 1, and non-BPCI hospitals. The average thirty-day total payment for the seventy-eight hospitals enrolled in major joint replacement of the lower extremity under phase 2 was $26,924 compared to $26,661 for the 197 hospitals in phase 1 and $26,832 for the 2,496 non-BPCI US hospitals that performed this procedure ( p value for difference across the three groups equals 0.936) ( Exhibit 3 ). The average thirty-day total payment for the thirty-eight hospitals enrolled in congestive heart failure under phase 2 was $16,309 compared to $16,106 for the 193 hospitals in phase 1 and $15,859 for the 2,815 non-BPCI US hospitals that provided care for congestive heart failure ( p value for difference across three groups equals 0.710) ( Exhibit 3 ).
Contribution Of Postacute Care Spending To Variations In Total Thirty-Day Spending
For phase 2 BPCI hospitals, we found that the index hospitalization accounted for the largest component of spending, followed by postacute care ( Exhibit 3 ). For example, for major joint replacement, the index hospitalization accounted for $12,300, postacute care for $7,896, health care provider payments for $4,564, readmissions for $2,059, and outpatient care for $273.
When we assessed the percentage variation of total thirty-day episode spending among phase 2 hospitals by each component of the bundled payment, we found that postacute spending accounted for the largest variation in thirty-day episode-based payments among phase 2 hospitals ( Exhibit 4 ). However, the percentage variation in total thirty-day episode-based payments explained by postacute care spending varied in a condition-specific pattern, ranging from 26.3 percent for chronic obstructive pulmonary disease to 75.3 percent for hip and femur procedures excepting major joint replacement.
Spending on readmissions ranked behind postacute care for explaining the variation in total thirty-day spending. These patterns were again condition specific: For orthopedic admissions, in which readmission rates tend to be low, readmissions accounted for 1.6–7.3 percent of the variation in total thirty-day spending. For congestive heart failure, chronic obstructive pulmonary disease, and pneumonia, in which readmission rates tend to be high, readmissions accounted for 24.1–25.3 percent of the variation. These patterns were similar for phase 1 hospitals ( Exhibit 4 ) and non-BPCI hospitals (see Appendix Exhibit A8). 5
We found that hospitals enrolled in the Bundled Payments for Care Improvement initiative were most often large, nonprofit, teaching hospitals in the Northeast. While many hospitals initially enrolled for all forty-eight eligible clinical conditions to explore patterns of spending during phase 1, only a small proportion of them entered into the phase 2 risk-bearing program, and of those that did, most chose to focus on three or fewer conditions. These were generally conditions for which they had high volumes but not necessarily ones for which they were at higher cost at baseline. For all hospitals, postacute care spending was a large component of total thirty-day spending and explained the largest proportion of variation in episode-based spending across hospitals. Taken together, these findings suggest that while there are substantial opportunities for savings across conditions, few hospitals are currently participating in the BPCI initiative, and those hospitals’ efforts are narrowly targeted.
Given the potential to deliver highly integrated care through episode-based bundled payment, the fact that only a small proportion of hospitals enrolled in risk-bearing agreements with CMS may be disappointing. Those that did were mainly large academic teaching hospitals in the Northeast. Importantly, these hospitals were also more likely than others to have existing affiliations with postacute care providers such as skilled nursing facilities, rehabilitation centers, and home health agencies. Although it may be unsurprising that members of this select group of hospitals are more equipped than others to engage in innovative payment and delivery approaches, this also has important implications for the generalizability of the initiative.
Participating hospitals chose only a small number of conditions, generally ones for which they already had high clinical volumes. Clearly, the hospitals did not choose their targets randomly but instead focused on those conditions with which they had substantial experience. Based on 2011 inpatient Medicare claims, major joint replacement accounted for approximately one in twenty total annual discharges for hospitals enrolling in this clinical condition—a very large number given that a typical large, high-complexity hospital often manages more than 1,200 unique primary diagnoses and performs close to 600 unique procedures. 9 Our findings suggest that many hospital leaders likely believe that a certain clinical volume threshold is necessary for engagement in bundled payment efforts. Medicare, therefore, may need to tailor condition-specific, episode-based bundled payments to individual hospitals based on their clinical volumes and case-mix.
We expected to find that phase 2 hospitals might be significantly higher cost at baseline than their nonparticipating counterparts. However, we did not find any such effect. It is possible that the small sample size of hospitals in phase 2 (only seventy-eight hospitals for joint replacement and thirty-eight for congestive heart failure) affected our power to detect meaningful differences. Given that the point estimates for total spending were nearly identical across all three of our groups, we suspect that there likely is no meaningful difference in baseline spending for these conditions among the three groups of hospitals. This finding might not be surprising because we standardized Medicare payments for our analyses, which effectively minimalized price variations, leaving differences in utilization patterns as the key driver of spending. However, given the large variations in actual reimbursement and the application of target discount rates for bundled payments, the BPCI initiative may still offer opportunities for cost savings to Medicare among the phase 2 participants.
Finally, our findings suggest that if the BPCI hospitals wish to find opportunities for savings, they would do well to look at postacute care services such as those provided by skilled nursing facilities, rehabilitation facilities, or home health agencies. These services accounted for the highest amount of the variation in total episode-based spending and, therefore, represent important targets in any effort to reduce health care spending for these conditions. 10 This is in keeping with prior evidence showing that postacute care spending accounts for not only the largest component of geographic variation in spending in the Medicare population 11 but also the largest growth in spending among Medicare beneficiaries. 12 Because patterns of variation were condition specific, condition-specific delivery innovations are likely to be needed. For example, savings in a bundled payment program might be maximized by reducing readmissions for chronic obstructive pulmonary disease, whereas savings could be maximized by reducing the use of skilled nursing facilities for major joint replacement. Additional lessons could be drawn from earlier efforts to create a bundled payment for inpatient care through the inpatient prospective payment system that has helped curtail increases in inpatient spending. 13,14
An important remaining question is whether the BPCI hospitals will continue to focus narrowly on target clinical conditions or expand beyond the enrolled clinical conditions to invest more broadly in care coordination and delivery transformation. If bundled payment does not gain a large toehold beyond the few commonly enrolled clinical conditions, then the BPCI initiative may result in limited savings to Medicare. However, if hospitals and delivery systems use the initiative as a springboard to more comprehensive bundled payment across multiple conditions, then the demonstration project will have an opportunity to help transform health care delivery and how we pay for it. Previous efforts to bundle payments have had mixed results with the positive impact of bundled payment diluted by concurrent fee-for-service arrangements. 15 Unfortunately, prior CMS demonstration projects do not provide much guidance on this issue. Some prior demonstrations, such as with accountable care organizations, were ultimately expanded to include a wide range of organizations, while other demonstrations such as the Acute Care Episode Demonstration have not been widely adopted. 16,17 The validity of the Innovation Center’s demonstration projects has also recently been questioned, with some researchers suggesting that randomized controlled trials are a more valid approach to evaluating delivery innovation in Medicare. 18 Nevertheless, the final evaluation of the BPCI initiative will be immensely helpful for policy makers as they seek to understand whether bundled payment can be a permanent mechanism through which to reduce spending while simultaneously improving and streamlining care delivery.
Several studies have assessed variations in episode-based spending and the resulting implications for bundled payment, 1,2,19,20 but this is the first study to our knowledge to systematically describe both the structural features as well as the thirty-day spending patterns of hospitals enrolling in the initiative. Similar to these previous studies, we found large variations in episode-based payments for common clinical conditions in the Medicare population, and our data confirm that postacute care accounts for the largest component of variation in episode-based total spending. 19,21 We extend the existing understanding of how hospitals may respond to the changing financial incentives posed by bundled payment through our analysis of changing enrollment patterns (and motivations) between the phase 1 non-risk-bearing and the phase 2 risk-bearing hospitals.
We found that hospitals enrolling in Medicare’s Bundled Payment for Care Improvement initiative were likely to be large, nonprofit, teaching hospitals in the Northeast that had a high degree of existing affiliation with postacute care providers. These hospitals have been strategic in their enrollment in the initiative, selecting conditions for which they have high clinical volumes. Episode-based bundled payment may serve to align financial incentives across the spectrum of care and help curtail Medicare expenditures. However, for episode-based payment to work effectively, Medicare may need to tailor condition-specific bundled payments to individual hospitals based on their clinical volumes and patterns of care, to facilitate meaningful partnerships between hospitals and postacute care providers.
The views expressed herein are those of the authors only and do not express the official views of the Department of Health and Human Services. The authors thank the Center for Medicare and Medicaid Innovation for providing detailed information on the Bundled Payments for Care Improvement Initiative. At the time the manuscript was prepared, Thomas Tsai was supported by the National Cancer Institute (Grant No. R25CA92203).
- 1 . Payment reform options: episode payment is a good place to start . Health Aff (Millwood) . 2009 ; 28 ( 2 ): w262 – 71 . DOI:
10.1377/hlthaff.28.2.w262Go to the article, Google Scholar
- 2 . Medicare’s bundled payment pilot for acute and postacute care: analysis and recommendations on where to begin . Health Aff (Millwood) . 2011 ; 30 ( 9 ): 1708 – 17 . Go to the article, Google Scholar
- 3 Confronting costs: stabilizing US health spending while moving toward a high performance health care system . New York (NY) : Commonwealth Fund ; 2013 Jan . Report No. 1653. Google Scholar
- 4 Centers for Medicare and Medicaid Services . Bundled Payments for Care Improvement initiative: general information [Internet]. Baltimore (MD) : CMS ; [cited
2015 Jan 8]. Available from: http://innovation.cms.gov/initiatives/bundled-payments/ Google Scholar
- 5 To access the Appendix, click on the Appendix link in the box to the right of the article online.
- 6 Centers for Medicare and Medicaid Services . BPCI initiative episodes: details on the participating health care facilities [Internet]. Baltimore (MD) : CMS ; [cited
2015 Jan 8]. Available from: http://innovation.cms.gov/initiatives/Bundled-Payments/Participating-Health-Care-Facilities/index.html Google Scholar
- 7 Centers for Medicare and Medicaid Services . BPCI models 2–4 Part A and Part B exclusions [Internet]. Baltimore (MD) : CMS ; 2014 [cited
2015 Jan 8]. Available from: http://innovation.cms.gov/Files/x/BPCI2-4_PartA-B_Exclusion.pdf Google Scholar
- 8 . Contribution of preventable acute care spending to total spending for high-cost Medicare patients . JAMA . 2013 ; 309 ( 24 ): 2572 – 8 . Crossref, Medline, Google Scholar
- 9 . Beyond volume: does hospital complexity matter? An analysis of inpatient surgical mortality in the United States . Med Care . 2014 ; 52 ( 3 ): 235 – 42 . Medline, Google Scholar
- 10 . Post-acute care—the next frontier for controlling Medicare spending . N Engl J Med . 2014 ; 370 ( 8 ): 692 – 4 . Crossref, Medline, Google Scholar
- 11 . Geographic variation in Medicare services . N Engl J Med . 2013 ; 368 ( 16 ): 1465 – 8 . Crossref, Medline, Google Scholar
- 12 . Large increases in spending on postacute care in Medicare point to the potential for cost savings in these settings . Health Aff (Millwood) . 2013 ; 32 ( 5 ): 864 – 72 . Go to the article, Google Scholar
- 13 . Why did Medicare spending growth slow down? Health Aff (Millwood) . 2008 ; 27 ( 3 ): 793 – 802 . Go to the article, Google Scholar
- 14 . Slower growth in Medicare spending—is this the new normal? N Engl J Med . 2012 ; 366 ( 12 ): 1073 – 5 . Crossref, Medline, Google Scholar
- 15 . The PROMETHEUS bundled payment experiment: slow start shows problems in implementing new payment models . Health Aff (Millwood) . 2011 ; 30 ( 11 ): 2116 – 24 . Go to the article, Google Scholar
- 16 . The Pioneer accountable care organization model: improving quality and lowering costs . JAMA . 2014 ; 312 ( 16 ): 1635 – 6 . Crossref, Medline, Google Scholar
- 17 . Controlling costs by expanding the Medicare acute care episode demonstration . JAMA Intern Med . 2014 ; 174 ( 9 ): 1438 – 9 . Crossref, Medline, Google Scholar
- 18 . Method of study is criticized in group’s health policy tests . New York Times . 2014 Feb 3 ; Sect. A:1. Google Scholar
- 19 . Large variations in Medicare payments for surgery highlight savings potential from bundled payment programs . Health Aff (Millwood) . 2011 ; 30 ( 11 ): 2107 – 15 . Go to the article, Google Scholar
- 20 . Lessons learned preparing for Medicare bundled payments . N Engl J Med . 2012 ; 367 ( 20 ): 1873 – 5 . Crossref, Medline, Google Scholar
- 21 . Medicare spending and outcomes after postacute care for stroke and hip fracture . Med Care . 2010 ; 48 ( 9 ): 776 – 84 . Crossref, Medline, Google Scholar