Association of Physician Group Participation in Accountable Care Organizations With Patient Social and Clinical Characteristics

IMPORTANCE Accountable care organizations (ACOs) may increase health care disparities by excluding physician groups that care for socially and clinically vulnerable patients. OBJECTIVE To estimate the association between the patient characteristics of a physician group and the group’s participation in a newly formed ACO. DESIGN, SETTING, AND PARTICIPANTS This retrospective cohort study investigated a 20% random sample of US Medicare fee-for-service beneficiaries attributed to physician groups identified in Medicare claims before ACO participation from January 1, 2010, through December 31, 2011. Physician groups that participated and did not participate in the Medicare Shared Savings Program (MSSP) from January 1, 2012, through December 31, 2014, were identified in the Medicare MSSP 2014 provider file. Data analyses were conducted from September 1, 2017, to March 30, 2018. EXPOSURES Using multivariable regression, the association between physician group participation in the MSSP and the group’s patients’ characteristics before ACO formation was estimated focusing on measures of the vulnerability of the group’s patients. All ACO-participating physician groups were compared with ACO-nonparticipating physician groups for reference, and estimates were made at the physician and patient level. ACO-participating physician groups cared for an equally large number of socially vulnerable patients compared with nonparticipating physician groups.


Introduction
Accountable care organizations (ACOs) are networks of health care practitioners that take on responsibility for managing the health care of a group of patients across the full continuum of health care settings and are held financially accountable for providing high-quality and low-cost care.
Accountable care organizations are an important and far-reaching recent change to health care with exponential growth over the past few years, covering more than 32 million lives by 2016. 1 Financial accountability is based on the sharing of savings, which is determined by whether the actual expenditures of the participating health care practitioners' assigned patient population are below a benchmark established by the prior expenditures of their assigned patients. Early evidence suggests that this ACO incentive structure can improve quality and constrain costs. [2][3][4][5][6][7][8][9][10] This incentive structure may also drive providers' choice to enter into an ACO arrangement.
Physician groups with patient panels that would most likely benefit from the coordination and management that the ACO model rewards may have the greatest incentives to enter into ACO contracts. These groups might include practices caring for a large number of patients with clinically complex conditions. If practices with more patients with complex conditions were more likely to participate in an ACO, documented improvements under the ACO model might appropriately be directed toward patients who would benefit the most.
However, in reality, the choice to be in an ACO may depend on the providers' capabilities to incorporate management tools to help manage and coordinate care for patients with complex needs.
These tools often require capital that might not be available to practices that serve poorly resourced populations, populations that often both have complex clinical needs and are socially vulnerable. On one hand, socially vulnerable groups (such as racial/ethnic minorities and patients living in impoverished neighborhoods) have much to gain from ACOs because they have a higher prevalence of many chronic health conditions, receive poorer quality of care, and experience worse health outcomes. 11,12 On the other hand, because care for vulnerable patients has historically been concentrated among relatively few providers who tend to have fewer financial and health care resources and perform worse on most traditional quality metrics, [13][14][15][16] ACOs may thus have little incentive to include providers that do not perform well on quality metrics and would require additional capital and resources to allow for the efficient management of population health. These realities raise concerns that ACOs could widen existing health care disparities. [17][18][19] Previous research has shown that ACOs have predominantly formed in geographic areas with a higher proportion of well-insured patients. 20 Additional research found that even after accounting for these geographic differences in where ACOs form, ACOs are less likely to include physicians working in areas that are more densely populated by low-income, low-educated, racial minority patients. 21 Thus, providers serving a disproportionate share of vulnerable patients may be systematically excluded from ACOs despite the fact that these patients may stand to gain the most from a well-functioning ACO model. However, the research to date has not examined whether the characteristics of physicians' patients are associated with which physician groups participate in ACOs. Thus, we estimated the association between physicians' patient characteristics and their likelihood of participating in a newly formed ACO. We focused on one of the best known and largest experiments with ACOs, the Medicare Shared Savings Program (MSSP), 22 which contracts with 561 ACOs and covers 10.5 million lives. 23

Study Cohort
We started with all Medicare fee-for-service beneficiaries enrolled in Part A and Part B, and following MSSP methodology, 24 we attributed beneficiaries to a physician taxpayer identification number, which approximates physician groups. We attributed beneficiaries with at least 1 primary care service assigns a point score to each based on mortality risk, and sums them to generate a score of disease burden. The final risk scores range between 0 and 41, with higher values associated with greater mortality risk over 10 years. 25 We included the following physician group-level variables from the SK&A Office Based Physician file as covariates: the number of physicians in the physician group, the percentage of physicians who were specialists, the percentage affiliated with a health system, and the percentage who were hospital based. In multivariable regression, physician groups with a higher proportion of dually enrolled patients and high-risk patients were more likely to participate in an ACO ( Table 3)  measures of patient characteristics and attribution to an ACO were not statistically different from zero.

JAMA Network Open | Health Policy
Finally, when testing whether these associations differed by the year a physician group became a part of an ACO, we found similar associations across ACO-participation years ( Table 4). Groups that began participating in an ACO in 2012 cared for 2 percentage points (95% CI, 0.3-3.7 percentage points; P = .02) more high-risk patients compared with non-ACO groups during the same period. At a patient level, this association remained, with 3.5 percentage points (95% CI, 0.8-6.2 percentage points; P = .01) more high-risk patients being attributed to ACO-participating physician groups compared with non-ACO groups. Physician groups that began participating in the ACO in 2013 and 2014 were also more likely to care for high-risk patients than non-ACO groups (by 1.

Discussion
We found little evidence that ACO-participating physician groups are less likely to care for socially vulnerable patients than nonparticipating ACO groups. Physician groups participating in the MSSP ACOs cared for patients who had more comorbidities than nonparticipating physician groups.  However, for other patient characteristics, including race, income, and dual eligibility, ACO-participating physician groups closely resembled nonparticipating physician groups.
Researchers and policy makers have raised concerns that ACOs may systematically exclude physicians who care for a disproportionate share of socially vulnerable patients. [17][18][19] Early research on ACOs found that patients attributed to Medicare ACOs tended to have higher incomes and were less likely to be black or enrolled in Medicaid 27 and that ACOs tended to be located in areas with lower poverty levels. 20 Other research found that even when controlling for the differences in the geographic location of ACOs, physicians working in zip codes with a higher share of vulnerable patients are less likely to participate in an ACO. 21 However, that research did not examine the patients attributed to ACO-participating groups, only the characteristics of patients living in the zip codes where the physician groups were located.
In this study, we extend that earlier work by looking at the characteristics of patients who were attributed to each physician group, and we found no evidence to support differences in patient characteristics by ACO participation. When we examined physician group-level characteristics, we found that those groups with higher proportions of dually enrolled patients were more likely to participate in the MSSP ACO program. At the same time, groups with higher proportions of patients who were black or living in high-poverty level zip codes were equally likely to participate. When we evaluated at the patient level, the association for dually enrolled patients was nonsignificant, suggesting that physician groups that care for fewer dually enrolled patients also care for fewer patients in general. Although we found no association at the patient level in social vulnerability and ACO attribution, there was an association between patient illness severity and attribution to an ACO, with patients who had more comorbidities being more likely to be attributed to an ACO-participating physician group. The ACO-participating physicians may be more focused on effectively managing chronic illnesses and better equipped to do so. 28 There are a number of reasons that practices caring for more chronically ill patients may be more likely to participate in an ACO. First, as noted, ACO practices may be better equipped to manage chronic diseases owing to the population-based emphasis of ACOs, which works to improve care for patients in the short term, the downside is that any efficiency gains from improving the treatment of patients with suboptimally managed chronic illness will likely only produce a 1-time gain in efficiency.
If practices are seeking opportunities to improve by targeting patients among whom efficiency gains will be easier to achieve, this may paradoxically protect vulnerable patients from exclusion from ACOs. That is, given the long history of low quality of care delivered to vulnerable patients, there are opportunities to improve that care and thus improve quality and efficiency of care.
Although our findings are encouraging with respect to the consequences of ACOs on disparities, whether vulnerable patients will equally benefit from being attributed to practices that are participating in the MSSP ACO program remains unknown. We examined baseline characteristics of practices that eventually participated in an ACO, that is, we examined patient characteristics approximately 1 year before ACO participation. After a physician group becomes part of an ACO, the makeup of its patients may change. Preferentially treating healthier or less vulnerable patients would be a strategy to improve performance under an ACO, although a nefarious one, and is an unintended consequence that must be monitored.
Furthermore, even in the absence of choosing healthier or less vulnerable patients, it is possible that physician groups that care for a disproportionate share of vulnerable patients do not see the same gains in quality that have been demonstrated among ACO-attributed patients more generally. [2][3][4][5][6][7][8][9] Practices that care for a disproportionate share of vulnerable patients are less likely to adopt Medicare's annual wellness visit, a strategy to target the treatment of high-risk patients that may be helpful to ACOs. 29 Thus, despite our findings, the broader question of whether ACOs might worsen disparities in care has not been answered.

Limitations
Our study results should be interpreted in the context of several limitations. First, our measures of social vulnerability are limited to those in the available data. Although race is accurately coded in Medicare claims, these data do not include ethnicity; thus, we could not examine the association between ethnicity and ACO attribution. In addition, our measure of poverty was necessarily measured at the zip code level rather than at the individual level. Second, we used taxpayer identification numbers to identify physician groups. Although this is standard practice, it does not fully represent the range of functional and economic relationships among physicians. Nonetheless, it is how the MSSP program defines and identifies groups of physicians. Our results are from a single ACO program. Although the MSSP is the largest ACO in the country, there has been an expansion of commercial and Medicaid ACOs; in 2016, the MSSP covered fewer than half of all lives attributed to an ACO in the United States. Thus, our results may not generalize beyond the MSSP ACO.

Conclusions
We provide the first evidence that we are aware of suggesting that physician groups that participated in the MSSP ACO program cared for an equally great number of socially vulnerable patients. Although this is a favorable outcome for ACOs, learning whether this translates into improved quality of care for such patients and thus narrows disparities requires continued monitoring.