Patient Population Loss At A Large Pioneer Accountable Care Organization And Implications For Refining The Program

Abstract

There is an ongoing move toward payment models that hold providers increasingly accountable for the care of their patients. The success of these new models depends in part on the stability of patient populations. We investigated the amount of population turnover in a large Medicare Pioneer Accountable Care Organization (ACO) in the period 2012‒14. We found that substantial numbers of beneficiaries became part of or left the ACO population during that period. For example, nearly one-third of beneficiaries who entered in 2012 left before 2014. Some of this turnover reflected that of ACO physicians—that is, beneficiaries whose physician left the ACO were more likely to leave than those whose physician remained. Some of the turnover also reflected changes in care delivery. For example, beneficiaries who were active in a care management program were less likely to leave the ACO than similar beneficiaries who had not yet started such a program. We recommend policy changes to increase the stability of ACO beneficiary populations, such as permitting lower cost sharing for care received within an ACO and requiring all beneficiaries to identify their primary care physician before being linked to an ACO.

Medicare’s Pioneer Accountable Care Organization (ACO) program began a new phase in 2015, following completion of the initial three-year contract period (2012‒14).[1] Payment and delivery innovations such as ACOs offer promise for stimulating improvements in health care, and arguably the Pioneer ACOs represent an advanced permutation of ACOs within the Medicare program.[2–6] Based in part on the perceived promise of ACOs, the Centers for Medicare and Medicaid Services (CMS) plans to expand the use of alternative payment models to 30 percent of all Medicare fee-for-service (FFS) payments by the end of 2016 and to 50 percent by 2018.[7,8]

Nonetheless, ACOs have not yet lived up to the aspirations of their proponents. Early evaluations of ACOs suggest only modest savings compared to traditional FFS payment models.[9–12] More important, thirteen of the original thirty-one Pioneer ACOs have withdrawn from their program contracts. Estimates suggest that the small magnitude of their savings were similar to savings achieved by the ACOs still in the program and that their withdrawal was based, in part, on program rules that are under revision.[13–16]

At least nine of the thirteen ACOs that have withdrawn have entered the Medicare Shared Savings Program, the larger of the agency’s two ACO programs. In this program, ACOs face one-sided risk—that is, they are eligible to receive bonuses for meeting program targets, but they face no financial penalties for failing to meet the targets. In contrast, Pioneer ACOs face both upside and downside risks: They are eligible for bonuses when their spending is below benchmarks but incur penalties when their spending exceeds the benchmarks. Recently, CMS has proposed a three-year delay in the transition from one-sided to two-sided risk for Medicare Shared Savings Program ACOs.[17,18] It also appears that ACOs have found it challenging to change their care delivery systems, especially when faced with considerable uncertainty about how best to meet beneficiary needs.[19]

The shift from traditional FFS Medicare to ACOs requires a well-defined beneficiary population for each ACO. Within FFS Medicare, beneficiaries may seek care from any physician who accepts Medicare, and they pay the same amount out of pocket for this care no matter which physician they pick; there is no need for defined populations. In contrast, the Pioneer ACO program introduces provider incentives based on the experience of ACO populations, but it continues to allow patients to choose any physician, thus creating a potentially one-sided instead of a two-sided relationship—meaning that both patients and providers are aware of and committed to each other. For example, Pioneer ACOs receive either financial bonuses or penalties depending on whether the actual Medicare spending for their group of beneficiaries falls below or above target spending amounts or benchmarks.

Fundamentally, population definition requires linking beneficiaries with ACOs in ways that preserve beneficiaries’ relationships with their existing outpatient physicians while permitting future financial risk sharing and delivery innovation. During the initial phase of the Pioneer ACO program (2012‒14), CMS used a passive approach to define prospective populations. The approach involved first attribution (linking beneficiaries with specific primary care physicians) and then alignment (linking beneficiaries with an ACO based on their primary care physicians’ participation in the ACO). This approach was based on algorithms that examined the patterns of primary care visits within historical claims data.

Importantly, CMS did not ask beneficiaries to identify their primary care physicians and imposed no requirements that beneficiaries receive their care from their aligned ACO. In other words, beneficiaries could obtain care from any physician or hospital that accepted Medicare, even though a single Pioneer ACO was “responsible” for all of their care. Finally, there were no systematic efforts to confirm that the ACO linkage process accurately assigned beneficiaries to their primary care physicians.

This combination of factors could result in a potentially lopsided relationship between patients and their health care providers. The approach is also in stark contrast with those created by well-defined private insurance networks or Medicare Advantage plans, in which patients have financial incentives to seek care within a network or plan. In short, high levels of turnover in the aligned beneficiary population could blunt or pervert ACO incentives to invest in new ways of delivering care because many beneficiaries could leave an ACO before it received the payoff from its investment, and ACOs are unlikely to know which beneficiaries will stay and which will leave.

Not surprisingly, there have been anecdotal reports of substantial changes in ACO membership from one year to the next.[20–24] Some simulation-based work suggests that the current approaches to defining an ACO’s population contribute to the amount of turnover. For example, compared with beneficiaries not receiving post acute care, those receiving postacute care were more likely to leave an ACO because they received a number of primary care services during this postacute care from someone other than their regular primary care physicians.[25,26]

Using data from Partners HealthCare, one of the largest Pioneer ACOs, we investigated the magnitude of turnover in the ACO population and differentiated among the following three types of changes: those resulting from beneficiaries leaving the traditional FFS Medicare program, those associated with physician turnover, and those resulting from the approach used to define the ACO population.

Study Data And Methods

Setting

Partners HealthCare is one of the largest Pioneer ACOs in terms of the number of aligned beneficiaries, with more than 82,000 aligned beneficiaries in the greater Boston area in the period 2012‒14. The Partners Institutional Review Board approved this study.

Definition Of The ACO Population

To define an ACO’s population, CMS first attributes beneficiaries to groups of providers (physicians or selected nonphysician providers) based on lists of participating providers submitted by ACOs for each year of the contract. The approach also considers potential beneficiary attribution to other groups of providers within Medicare, including the groups of providers associated with all other Pioneer ACOs, providers associated with non-ACO tax identification numbers, federally qualified health centers, rural health clinics, or critical access hospitals.

CMS aligns each beneficiary to the group of providers with the plurality of primary care spending for that beneficiary, as defined by evaluation and management codes.[1] At the beginning of each calendar year, beneficiaries receive a letter informing them of the Pioneer ACO program and permitting them to opt out of data sharing with the ACO. After this initial letter there is limited systematic communication with beneficiaries about the ACO.

ACOs have substantial flexibility in creating their list of providers: they choose which types of providers to include, and which specific providers are on the lists that they send to CMS. Partners HealthCare, the ACO we studied, included only physicians within primary care, defined as physicians with board certification in internal medicine or family practice, and nurse practitioners and physician assistants also working in the internal medicine or family practice departments. Other ACOs also included specialists and residents. CMS then aligns to each ACO those beneficiaries attributed to providers on the ACO’s list and sends an enrollment file of aligned beneficiaries to the ACO.

After alignment CMS sends the Medicare FFS claims (dating back through 2009) for all aligned beneficiaries to each ACO, to provide ACOs with data to use in improving care delivery. Other details about attribution and alignment are available online.[1] For this study, we used these annual lists. We defined three cohorts of ACO beneficiaries based on the year when they first joined the ACO population. Sixteen beneficiaries in 2012 (0.04 percent of cohort 1) and one in 2013 (0.01 percent of cohort 2) opted out of sharing their data, and thus we excluded them from the analysis.

Changes In ACO Eligibility

Medicare beneficiaries become ineligible for an ACO population if they switch to Medicare Advantage or move outside of the ACO service or catchment area. CMS assesses these reasons for ineligibility and provides a single variable for ACO ineligibility, which we describe as “leaving the local FFS program.” For this study, we have identified deaths, which also are related to turnover, using the ACO beneficiary table from CMS.

Factors Potentially Associated With Attribution Or Population Turnover

We were interested in the impact of beneficiaries’ clinical relationships with their primary care physicians, patients’ engagement with clinically based services, and changes in patient health status on the stability of patient alignment to the ACO. For example, we examined whether a beneficiary’s attributed primary care physician left the ACO, whether the beneficiary received supplemental services from the ACO (for instance, whether a high-risk patient engaged with the Partners’ integrated Care Management Program), and whether there were changes in the number of a beneficary’s medical conditions according to CMS’s Hierarchical Condition Categories (HCC) risk-adjustment algorithm.

The ACO’s integrated Care Management Program was an ancillary program for high-risk patients that supplemented the care that they received from their primary care physicians. The ACO identified potential candidates for the program at the ACO level and then had individual primary care physicians provide input on the initial list of candidates. For example, the physicians could exclude beneficiaries who were not clinically appropriate for the program.

We also included in our analyses a number of baseline beneficiary characteristics associated with leaving the ACO. These included age category, sex, race or ethnicity, eligibility for Medicaid (whether or not the beneficiary was a dual eligible), neighborhood socioeconomic status at the level of the census block group (using American Community Survey estimates), baseline and prospective HCC scores, the concentration of primary care received from a single primary care physician before being aligned with the ACO, and driving time between the home address and the clinic of the attributed primary care physician. To determine driving time for subjects living in the New England area, we geocoded both beneficiaries’ home addresses and physicians’ office addresses and used GIS software (ArcGIS) to calculate the minimum travel time by car.

Analyses

We first described the characteristics of each cohort of newly aligned beneficiaries and calculated the numbers of beneficiaries experiencing changes in the number of medical conditions, status of their attributed primary care physician, or engagement with the ACO. We then examined the numbers of beneficiaries who joined or left the ACO population in each year of the initial three-year contract.

After excluding those beneficiaries who ceased to be eligible for alignment because of death or departure from the local FFS Medicare program, we used logistic regression models to examine the association between the three factors of interest and leaving the ACO population—that is, ceasing to be aligned in the following calendar year. We focused on beneficiaries within cohorts 1 and 2, those for whom we had multiple years of complete ACO status data in the period 2012‒14.

We then estimated the predicted percentage of beneficiaries leaving the ACO population with respect to the following five factors of interest: changes in the attributed primary care physician’s participation in the ACO, participation in a care management program, primary care concentration with a physician before alignment with the ACO period, driving time, and changes in the number of medical conditions. For each of these factors, we used the “margins” command in Stata, version 14, to get the marginal effect of each covariate, standardized with respect to the other covariates. We used separate models for the five factors to exclude covariates that might exist on the causal pathway. Sensitivity analyses using all covariates yielded comparable findings.

To examine how the 2015 rules for defining ACO populations could have affected our findings, we also used the 2015 attribution and alignment rules to define the ACO population.[27] Our simulations differed slightly from CMS’s actual approach because we were unable to include groups of physicians associated with other Pioneer ACOs in the attribution process as CMS does, as year-specific national provider identifier lists were not publicly available.

Limitations

This study had a number of limitations. First, our data came from a single Pioneer ACO in a single market. Thus, our results may not be generalizable to other Pioneer ACOs or to other ACO programs. However, the ACO in our study is one of the largest ACOs, and complaints of beneficiary turnover appear to have been pervasive—a problem that CMS is now attempting to address.[28] Indeed, we expect that the extent of beneficiary turnover might be greater in ACOs that have different organizational structures than the ACO in our study—for example, without referral hospitals or with subspecialists in their lists of affiliated providers. Programs that use retrospective attribution (for example, the original Shared Savings Plan program) arguably would have more difficulty in planning and making investment decisions than programs that use prospective attribution.[29]

Second, we used only the data provided by CMS to ACOs, which excluded data about, for example, substance use disorder claims. However, the missing data were unlikely to alter the estimates of attribution to a primary care physician.

Third, we lacked information on the beneficiary’s perspective. Thus, we could not determine whether or not beneficiaries deliberately changed their primary care physician. However, including the beneficiary’s perspective is the motivation for several of our recommendations.

Fourth, our measures of changing health status were crude in that they relied on claims reports of inpatient and outpatient diagnoses. However, these are the data and approaches that CMS uses for payment in the Medicare Advantage program and that the health insurance Marketplaces use for risk-adjusted premium transfers. Moreover, we would expect that crudeness of the data would bias our findings to the null—that is, better data could result in a finding of more changes in medical use patterns.

Finally, we did not address the financial implications of the beneficiary changes, which are beyond the scope of this article.

Study Results

Baseline Characteristics And Year-To-Year ACO Population Turnover

There were 42,050 beneficiaries aligned with the ACO in 2012. With each program year, additional beneficiaries joined the ACO population either through new attribution to a primary care physician affiliated with the ACO or through the attribution to a primary care physician who was newly affiliated with the ACO. Among the beneficiaries newly aligned with the ACO in 2013 (n = 19,522) or 2014 (n = 21,202), 49 percent were newly attributed to an existing ACO primary care physician, and 51 percent were attributed to a primary care physician new to the ACO.

In 2014 (the third year of the program), only 45 percent of the beneficiaries had been aligned with the ACO since 2012 (Exhibit 1). Thirty-two percent of the beneficiaries who had been aligned with the ACO in 2012 had left by 2014. Of the 2012 beneficiaries, 1,948 (2 percent) left the ACO in 2013 and were realigned with it in 2014.

Exhibit 1.

Beneficiaries Aligned With Partners HealthCare, A Pioneer Accountable Care Organization (ACO), In 2012‒14, By Year Of Alignment

SOURCE Authors’ analysis of 2012–14 Pioneer ACO enrollment data. NOTES Beneficiaries are aligned, or linked, with an ACO if their primary care physician participates in the ACO. Beneficiaries in cohort 1 were aligned with the ACO in 2012, those in cohort 2 were aligned in 2013, and those in cohort 3 were aligned in 2014.

Among the beneficiaries aligned in 2012, 60.6 percent were female, 20.1 percent were dually eligible for Medicare and Medicaid, and 89.0 percent were white (Exhibit 2). The mean age was 72.8 years. Beneficiaries aligned in 2013 and 2014 were mostly similar to those aligned in 2012 in terms of sociodemographic characteristics. However, the mean risk scores (HCC) varied from year to year.

Exhibit 2.

Baseline Characteristics Of Beneficiaries Aligned With Partners HealthCare, A Pioneer Accountable Care Organization, By Year Of Initial Alignment

	Cohort number (date of initial alignment)
Characteristic	1 (2012)	2 (2013)	3 (2014)
Beneficiaries	42,050	19,522	21,202
Mean age (years)	72.8	72.2	71.0
Age range	21–109	20–106	21–104
Younger than 65	14.2%	14.5%	15.0%
Female	60.6	59.9	40.8
Dually eligible for Medicaida	20.1	21.9	19.8
Race or ethnicity
White	89.0	89.1	91.9
Black	5.2	4.5	3.0
Hispanic	1.8	1.8	1.3
Asian	2.3	1.7	1.3
North American native	0.05	0.06	0.0
Other	1.4	1.9	1.1
Unknown	0.3	1.0	1.3
Mean CMS-HCC Score	0.9	1.0	0.9
Range of CMS-HCC Scores	0.1–8.8	0.1–8.8	0.1–8.4
Travel timeb (minutes)
0–10	50.5%	54.0%	48.3%
>10–20	21.1	19.0	23.6
>20–30	8.8	7.5	8.1
>30–45	5.3	4.6	4.2
>45	4.4	3.7	3.2
Non-New England address	3.4	3.0	2.6
Unknown address	6.5	8.2	10.2
Census tract population below the federal poverty level
0–<5	34.8	33.9	36.0
5–<10	26.8	27.8	30.4
10–<20	19.0	20.3	14.7
20 or more	11.2	11.1	10.5
Unknown neighborhood income	8.3	6.9	8.5

SOURCE Authors’ analysis of Pioneer Accountable Care Organization enrollment and claims data. NOTES Beneficiaries are aligned, or linked, with an ACO if their primary care physician participates in the ACO. HCC is a Centers for Medicare and Medicaid Services’ Hierarchical Condition Category.

^aAs of 2014.

^bBetween beneficiary’s home and primary care physician’s office.

Nearly all beneficiaries appeared to obtain the majority of their primary care from a single physician, as defined by CMS. For example, 91 percent of beneficiaries aligned in 2012 saw their attributed primary care physician for more than half of their primary care visits (data not shown). Most beneficiaries appeared to live within a twenty-minute drive of their primary care physician’s office (Exhibit 2). However, during the study period there were 2,550 (3.1 percent) beneficiaries who lived outside of the New England area, according to Medicare’s records, and another 6,507 (7.9 percent) beneficiaries with addresses that we could not identify using our geocoding software (data not shown).

Primary Care Affiliations, Beneficiary Medical Conditions, And Program Engagement

Among the beneficiaries in each cohort who were aligned with the ACO at the beginning of the year, 4‒11 percent were attributed to a primary care physician who left the ACO in that year, and for 21‒22 percent, the number of HCCs decreased (Exhibit 3). The care management program identified 5‒10 percent of all beneficiaries across the cohorts as having high and potentially modifiable risks for elevated spending and enrolled between 1‒6 percent of all beneficiaries in the year of identification.

Exhibit 3.

Changes In Primary Care Physician Status, Beneficiary Medical Conditions, Or Care Management Eligibility And Participation In Partners HealthCare, A Pioneer Accountable Care Organization, 2012‒14

	Cohort number (date of initial alignment)
	1 (2012)		2 (2013)		3 (2014)
Changes	No.	%	No.	%	No.	%
Beneficiaries aligned in 2012	42,050	100
PCP left ACO during 2012	1,824	4.3
In 2012 beneficiaries whose number of HCCs:
Was unchanged	22,477	53.5
Decreased	9,169	21.8
Increased	10,404	24.7
Beneficiaries identified for care management in 2012	3,488	8.3
Beneficiaries starting care management in 2012	337	0.8
Beneficiaries aligned in 2013	34,588	100	19,522	100
PCP left ACO during 2013	1,370	4.0	2,143	11.0
In 2013 beneficiaries whose number of HCCs:
Was unchanged	18,227	52.7	10,221	52.4
Decreased	7,389	21.4	4,358	22.3
Increased	8,972	25.9	4,943	25.3
Cumulative beneficiaries identified for care management	5,552	16.1	—a	—a
Beneficiaries identified for care management in 2013	2,624	7.6	1,460	7.5
Beneficiaries starting care management in 2013	2,100	6.1	573	2.9
Beneficiaries aligned in 2014	30,577	100	14,388	100	21,202	100
PCP left ACO during 2014	1,675	5.5	1,510	10.5	2,084	9.8
Cumulative beneficiaries identified for care management	5,783	18.9	2,280	15.9	—a	—a
Beneficiaries identified for care management in 2014	1,624	5.3	1,099	7.6	2,140	10.1
Beneficiaries starting care management in 2014	512	1.7	290	2.0	271	1.3

SOURCE Authors’ analysis of Pioneer Accountable Care Organization (ACO) enrollment and claims data. NOTES Beneficiaries are aligned, or linked, with an ACO if their primary care physician participates in the ACO. A change in counts of Centers for Medicare and Medicaid Services’ Hierarchical Condition Categories (HCCs) is based on the number of HCCs in the current year (determined by inpatient and outpatient diagnoses) and the number in the previous year. Because of capacity constraints, not all beneficiaries identified as candidates for the high-risk care management program were able to start the program in the year they were identified. We used the month of care management assessment as the month when a beneficiary started the care management program. PCP is primary care physician.

^aNot applicable.

Changes In ACO Status

Of the 42,050 beneficiaries aligned with the ACO in 2012, 82.3 percent remained aligned in 2013, 2.5 percent died in 2012, 0.7 percent left the local traditional FFS Medicare program (Exhibit 4), and 14.6 percent remained in traditional Medicare but left the ACO. In 2014, 72.7 percent of the beneficiaries aligned in 2012 remained aligned (Exhibit 4). Similarly, of the 19,522 beneficiaries newly aligned in 2013 cohort, 73.7 percent remained aligned in 2014.

Exhibit 4.

Unadjusted Changes In Beneficiaries’ Accountable Care Organization (ACO) Eligibility Or Alignment With Partners HealthCare, A Pioneer ACO, 2012‒14

	Cohort 1		Cohort 2
	No.	%	No.	%
Aligned beneficiaries	42,050	100	19,522	100
Aligned in 2013	34,588	82.3
Not aligned in 2013	6,149	14.6
Died in 2012	1,029	2.5
Left local TM in 2012	284	0.7
Aligned in 2014	30,577	72.7	14,388	73.7
Not aligned in 2014	7,352	17.5	3,224	16.5
Died in 2012 or 2013	2,172	5.2	706	3.6
Died in 2013	1,143	2.7	706	3.6
Left local TM in 2012 or 2013	1,949	4.6	1,204	6.2
Left local TM in 2013	1,665	4.0	1,204	6.2

SOURCE Authors’ analysis of Pioneer ACO enrollment and claims data. NOTES Beneficiaries are aligned, or linked, with an ACO if their primary care physician participates in the ACO. Beneficiaries in cohort 1 were aligned with the ACO in 2012, and those in cohort 2 were aligned in 2013. Beneficiaries ceased to be eligible for the Pioneer ACO program if they died or left the local traditional fee-for-service Medicare program (TM), either by switching to Medicare Advantage or moving outside of the service area. We excluded cohort 3 because 2015 enrollment lists were not available at the time this study was conducted.

Simulations using CMS’s 2015 attribution and alignment rules did not substantially affect the percentage of beneficiaries remaining aligned with the ACO.[27]

Departure After One Year In the ACO

Beneficiaries whose primary care physician left the ACO were nine times more likely to leave the ACO in the next year, compared with those beneficiaries whose physician did not leave (Appendix Exhibit A2a).[30] Conversely, beneficiaries who started the ACO’s integrated Care Management Program were one-third less likely to leave the ACO population in the next year, compared with those beneficiaries who were identified for the program but who had not yet started it (for [INSERT], see Appendix Exhibit A2b).[30]

For all of our findings, sensitivity analyses using simulations of the proposed 2015 attribution and alignment rules did not result in substantial changes. That is, the direction of the associations between these factors and continued alignment with the ACO remained the same as the results discussed above (whose results are presented in greater detail in the Appendix).[30]

Discussion

To our knowledge, this is the first analysis of actual population turnover at a Pioneer ACO. We found substantial year-to-year turnover, which suggests that ACOs are faced with the prospect of addressing the medical needs of a fluid population. These findings highlight major challenges to designing thoughtful targeted programs to improve care delivery for a specific group of beneficiaries within a short time period.[31,32]

Some of the reductions in the size of the ACO’s population were because beneficiaries ceased to be eligible when they moved outside of the service area or died. After we removed these reasons for population turnover, we still found substantial numbers of beneficiaries who ceased to be aligned with the ACO. Not surprisingly, when a primary care physician left the ACO, his or her patients often left too. Indeed, several factors associated with population turnover reflected beneficiaries’ engagement with either individual physicians or the ACO as a whole. These associations persisted even after we adjusted for baseline level of illness (HCC score) and other individual traits as well as characteristics of the relationship between beneficiary and primary care physician. CMS’s 2015 attribution and alignment rules had little effect on the amount of population turnover.

The challenge for the Medicare Pioneer ACO program as a whole is that having unpredictable fluid populations of beneficiaries might have negatively affected and might continue to affect the investment decisions of or the development process for new programs to improve care, particularly for ACOs with limited access to capital or those in traditionally underresourced areas. The care management program in the ACO we studied represents one such program to improve care, but it is resource intensive, targeting beneficiaries with the greatest risk of future medical need and, arguably, with the most potential for improvement.

The optimal ACO program design also could vary as the composition of the beneficiary population changes. Conversely, reductions in access or use could affect medical spending very quickly, with potentially adverse effects manifesting themselves over longer periods of time—including after some beneficiaries leave the ACO.

To the extent that shifting ACO populations create perverse incentives for such skimping on care, careful monitoring is needed. For example, as the incentives to reduce overall annual spending increase, some ACOs could decide that redesigning care delivery for a fluid population is too risky an investment and choose instead to reduce access to expensive types of care, such as specialty services or drugs.

More important, the primary goal of CMS’s payment innovation efforts is to encourage sustainable changes in care delivery that improve the quality and efficiency of care. But such changes could be costly to develop, requiring substantial leadership attention and provider effort, and could require additional resources over long periods of time to have a positive impact. Decisions about such investments are made within the context of increasing financial austerity, not to mention the historical resource constraints that many providers face. In short, many organizations could have limited financial flexibility to innovate, which could be exacerbated by both uncertainty about the target population and the prospect that many patients might cease to be aligned with an ACO before any of its interventions could affect spending.

We also discovered a number of beneficiaries in the ACO population with home addresses in the Medicare data that were outside of the ACO’s region or were invalid, which raises additional concerns. For example, a large proportion of these out-of-region beneficiaries had home addresses in Florida, which could have represented second homes, but a number of beneficiaries had addresses as far away as Hawaii, which would have required considerable travel time to obtain primary care in Massachusetts. The algorithm that CMS used during the study period excluded beneficiaries who moved outside of the ACO’s service area but did not exclude those who were outside of the service area when they were aligned with the ACO and did not move. Excluding these beneficiaries from the ACO population could make sense from the perspectives of policy or incentive design and would represent a straightforward extension of the geographical restriction in the original population definition.

This issue also highlights the importance of such data as home addresses, which historically might not have been considered as clinically relevant or relevant under a FFS payment model but which are relevant under new alternative payment contracts such as those for ACOs.

Policy Recommendations To Create A Two-Sided Relationship

To date, the effort involved in fostering stable clinical relationships within ACOs has fallen mostly on providers.[33] In the future, in addition to making efforts to improve the technical aspects of linkage, CMS could require that beneficiaries choose to participate in the Pioneer ACO program through either opt-out or opt-in policies, instead of inferring the choice to participate based on visit patterns. For example, CMS has recently piloted an attestation process in which previously aligned beneficiaries identify their primary care physician, which would trump any linkage changes made through the attribution or alignment algorithms.[34] By 2017 beneficiaries will have the ability to attest that their physician is part of a particular ACO and be assigned to that ACO, but will not be required to do so.[35] For many years, commercial health maintenance organization insurance plans have used similar approaches of requiring beneficiaries to select a primary care physician.

There also could be tangible incentives for beneficiaries to adhere to their physician choices. For example, these choices could be linked with benefit design, such as permitting lower or no cost sharing for care within the ACO. Under such an arrangement, beneficiaries would be no worse off with respect to care outside of the ACO—that is, they would have no higher cost sharing for such care—than is now the case. Medicare Advantage and Medigap Select plans already use such approaches, and the latest Medicare ACO permutation, the Next Generation ACO model, is planning to use some permutation of these two approaches,[36] although details are not yet available. It also is too early to assess whether the attestation pilot within the Pioneer ACO program is having its intended effects, or whether the pilot could be extended to the initial attribution of patients to physicians.

Other more technical fixes could help refine the linkage process. For example, the minimum attribution strength could be increased, or the use of a maximum distance between ACOs and beneficiaries’ home addresses could be implemented. These types of changes could reduce the number of beneficiaries aligned with ACOs but could also improve the accuracy of the linkage. However, technical fixes might not have as large an impact as the policy recommendations above to align patient incentives with provider incentives.[37]

Conclusion

As Medicare payment reform enters its next stage, there is an opportunity to refine the reforms. Both improving the accuracy of the process for defining an ACO’s population and improving the stability of the relationships between beneficiaries and provider organizations represent important policy areas for consideration. During the previous stage of reform—our study period, 2012‒14—there was substantial Pioneer ACO population turnover from year to year. Even after removing turnover resulting from beneficiary deaths or departures from the traditional FFS Medicare program and adjusting for provider turnover, we found that the ACO studied had substantial losses of beneficiaries across years. Efforts to improve the process of defining an ACO’s population, such as requiring beneficiaries to select their primary physician or preferred health care delivery organization and providing financial incentives to adhere to this selection, could balance the incentives between patients and their health care providers and increase both the stability and the effectiveness of the payment innovations.

Biographies

John Hsu (jhsu2000@gmail.com) is director of the Clinical Economics and Policy Analysis Program, Mongan Institute, at Massachusetts General Hospital (MGH) and an associate professor in the Department of Medicine and in the Department of Health Care Policy at Harvard Medical School, both in Boston.

Mary Price is an analyst in the Mongan Institute, MGH.

Jenna Spirt is an analyst in the Mongan Institute, MGH.

Christine Vogeli is an assistant professor in the Mongan Institute, MGH.

Richard Brand is a professor emeritus of biostatistics at the University of California, San Francisco.

Michael E. Chernew is a professor in the Department of Health Care Policy at Harvard Medical School.

Sreekanth K. Chaguturu is vice president for Population Health at, Partners HealthCare, in Boston, a staff physician at MGH, and an instructor in medicine at Harvard Medical School.

Namita Seth Mohta is faculty at the Center for Healthcare Delivery Sciences and a hospitalist at Brigham and Women’s Hospital, in Boston.

Eric Weil is the senior medical director for population health, Partners Healthcare; associate medical director of the Massachusetts General Physicians Organization, in Boston; and associate chief of clinical affairs, Division of General Internal Medicine, MGH.

Timothy Ferris is the senior vice president, Population Health, Partners Healthcare and MGH, and an associate professor of Medicine at MGH and Harvard Medical School.

Appendix

Exhibit A1.

Odds of Leaving the ACO in the Next Year

	Model for Travel Time			Model for HCC Changes			Model for CMP/PCP Changes/Concentration
	OR	95%	CI	OR	95%	CI	OR	95%	CI
Quartile of Baseline HCC Score: Q1
Q2	0.82	0.76	0.88	0.83	0.77	0.89	0.82	0.76	0.89
Q3	0.73	0.67	0.78	0.75	0.69	0.81	0.73	0.68	0.79
Q4	0.66	0.61	0.72	0.74	0.69	0.81	0.68	0.62	0.73
Change in No. of HCCs: No Change
Decrease in HCC Count	1.11	1.04	1.19	1.17	1.09	1.25
Increase in HCC Count	1.02	0.96	1.09	1.05	0.99	1.12
Care Management Participation in Baseline Year: Identified, but did not start
Not Eligible	0.53	0.48	0.57				0.52	0.47	0.57
Started	0.56	0.44	0.71				0.56	0.44	0.71
Concentration of care with a single PCP (per 10 percentage point increase)	0.92	0.91	0.94	0.92	0.91	0.93	0.92	0.91	0.93
Travel Time: 0-10
>10-20	1.09	1.01	1.17	1.09	1.01	1.17	1.09	1.01	1.17
>20-30	1.19	1.08	1.31	1.18	1.07	1.31	1.19	1.08	1.31
>30-45	1.30	1.15	1.47	1.29	1.15	1.46	1.30	1.15	1.47
45+	1.90	1.68	2.13	1.87	1.67	2.11	1.90	1.68	2.13
Non-NE address	0.51	0.38	0.68	0.49	0.36	0.65	0.51	0.38	0.68
Unknown	0.29	0.23	0.38	0.29	0.22	0.37	0.29	0.23	0.38
PCP: Left ACO
Stayed in ACO	0.010	0.009	0.012	0.010	0.009	0.012	0.010	0.009	0.012
Age: <50
50-64	0.74	0.66	0.84	0.75	0.66	0.85	0.74	0.66	0.84
65-79	0.64	0.57	0.72	0.64	0.57	0.71	0.64	0.57	0.72
80+	0.72	0.63	0.82	0.72	0.63	0.81	0.72	0.63	0.82
Gender: Female
Male	1.03	0.97	1.08	1.03	0.97	1.08	1.03	0.97	1.08
Race: White
Black	0.92	0.81	1.04	0.92	0.82	1.05	0.92	0.81	1.04
Non-White & Non-Black	1.01	0.90	1.13	1.00	0.89	1.12	1.01	0.90	1.12
Non—dual
Dual eligible for Medicare and Medicaid	1.32	1.21	1.43	1.33	1.22	1.44	1.32	1.21	1.43
Year of ACO Entry: 2012
2013 Entry	0.88	0.83	0.94	0.88	0.83	0.94	0.88	0.83	0.94
% of Census Tract Population below FPL: 0-<5%
5-<10%	1.01	0.94	1.08	1.01	0.94	1.08	1.01	0.94	1.08
10-<20%	1.13	1.04	1.22	1.13	1.04	1.22	1.13	1.04	1.22
20%+	1.11	1.01	1.23	1.12	1.01	1.23	1.11	1.01	1.23
Unknown	5.70	4.35	7.47	5.83	4.45	7.64	5.70	4.35	7.47
Constant	66.33	51.16	86.01	36.17	28.26	46.30	69.38	53.64	89.74

Source: Authors’ analysis of Pioneer ACO enrollment and claims data

Exhibit A2. Predicted Percent Leaving the ACO.

A2a: Standardized % by Changes in PCP-ACO Affiliation

A2b: Standardized % by Care Management Participation

A2c: Standardized % by Pre-ACO PCP Care Concentration

A2d: Standardized % by Travel Time (in Minutes)

A2e: Standardized % by Change in HCC Count

Source: Authors’ analysis of Pioneer ACO enrollment and claims data