Abstract
Background: The Centers for Medicare & Medicaid Services (CMS) has launched multiple alternative payment models (APMs) to address limitations with fee-for-service payments in traditional Medicare (TM), including challenges of TM in supporting high-quality primary care.
Methods: Using Medicare claims, CMS data on APM participation, and publicly available data, we examined the association between primary care physician (PCP) participation in TM APMs between 2017 and 2022 and the essential primary care features of accessibility, comprehensiveness, continuity, and coordination.
Results: 38.1% of PCPs in our analysis participated in at least 1 of 14 APMs during the study period. Based on a difference-in-differences analysis, we found that participation in APMs was significantly associated with improved outcomes on various key primary care dimensions. Of TM APMs assessed, hybrid payment models with no financial risk showed a significant association with improvement in all measures of the 4 essential features of primary care, both on measures that passed the parallel trends test as well as those that did not.
Conclusions: Overall participation in APMs is significantly associated with improved PCP outcomes on measures of various essential features of primary care. Of TM APMs assessed, only hybrid payment models with no financial risk showed a significant association with PCP improvement on all essential features of primary care. While methodological limitations preclude asserting a causal relationship, these findings support continued efforts to improve primary care for Medicare beneficiaries through hybrid payments to practices without imposing downside financial risk on PCPs.
- Alternative Payment Models
- Centers for Medicare and Medicaid Services (U.S.)
- Health Policy
- Medicare
- Primary Health Care
Introduction
Recent studies1 show that US primary care physicians (PCPs) perform poorly in achieving the core features of primary care (identified by Starfield2 and others3 as accessibility, comprehensiveness, continuity, and coordination). Over decades, studies have demonstrated that these primary care features are associated with improved outcomes for patients and society.1,3–8 These outcomes include enhanced patient satisfaction3 and adherence to recommendations,1 as well as reduced Emergency Department (ED) visits, hospitalizations, and overall costs of care.9,10
Recognizing the inadequacy of Traditional Medicare (TM) fee-for-service (FFS) payments for supporting high-quality primary care,11,12 the Centers for Medicare & Medicaid Services (CMS) and its Center for Medicare and Medicaid Innovation (CMMI) have implemented a variety of alternative payment models (APMs).13 Several of these models direct payments to PCPs or their practices, and focus on one or more of these core features of primary care through model requirements (eg, requirements for PCP accessibility in Comprehensive Primary Care Plus [CPC+]14) or incentives (eg, penalties for reduced PCP continuity in Primary Care First [PCF]15). More population-oriented TM APMs, such as the Medicare Shared Savings Program (MSSP) Accountable Care Organization (ACO) models, focus on a broader array of clinicians and provider entities but often use beneficiary assignment methodologies based on primary care services.16,17 Accordingly, observers have noted the potential importance of improved primary care for success of these population-based models as well,18,19 although their financial incentives are directed at larger entities rather than PCPs.
Research to date suggests PCPs in TM APMs may still show lapses in comprehensiveness, continuity, and coordination;6,20–22 and model participation may not lead to detectable improvements in these claims-based outcomes.22 Factors like community characteristics, practice composition, and health system affiliations also influence how PCPs in APMs deliver care.6,23 To explore which payment models promote better primary care, we categorized 14 PCP-relevant TM APMs and assessed the independent association of PCP participation in these models with improvements in the PCP’s delivery of the core features of primary care.
Data/Methods
Conceptual Framework
This observational study used a conceptual framework that describes factors contributing to variations in clinician practice, including the essential features of primary care.1,24–26 These include PCP and practice characteristics, the larger organizations with which they may be affiliated, the patients and communities they serve, and the payment model and policy contexts affecting primary care.
Data Sources
Through CMS’s Virtual Research Data Center we accessed Medicare FFS claims and data on participation in CMS initiatives to create claims-based outcomes and identify PCPs’ participation in specific APMs. We used Medicare Data on Provider Practice and Specialty (MD-PPAS) to identify PCPs and their characteristics, and other publicly available data to define characteristics of the areas where the PCPs were located.
PCP Sample
Using MD-PPAS we identified physicians with primary care specialties,27 excluding nurse practitioners (NPs) and physician assistants (PAs) because information in MD-PPAS does not allow a clear distinction between primary care and specialist NPs and PAs. We used combinations of Taxpayer Identification Numbers (TINs) and National Provider Identifiers (NPIs) to determine APM participation, identifying the NPI’s predominant TIN from MD-PPAS to create NPI-TIN combinations. We limited the sample to PCPs not in an APM in 2016, choosing this year as the baseline because it was the first year of claims submission with International Classification of Diseases Tenth Revision (ICD-10) diagnosis codes, used to construct several of our outcome measures. We further excluded PCPs who provided relevant services to fewer than 10 or more than 2500 TM beneficiaries, or served beneficiaries predominantly located outside the 50 states and District of Columbia. The intervention group consisted of PCPs who began participating in an APM between 2017 and 2022.
APMs Included in Analysis
We included APMs that qualified for the Merit-Based Incentive Payment System and operated in more than one state. Because we focused on models impacting the broad array of Medicare beneficiaries receiving primary care, we examine APMs that involve TM beneficiaries without regard to the presence of a specific medical condition (excluding, for example, the Comprehensive End-Stage Renal Disease Care Model). The resulting 14 APMs included 10 MSSP ACO tracks, the Next Generation ACO Model (NextGen), Global and Professional Direct Contracting, CPC+, and PCF.
We drew on frameworks in the literature to group these 14 APMs into 2 categories based on the entity to which payments are directed: primary care practices or ACOs.12,28,29 In another set of analyses, we grouped APMs using concepts from the National Academies of Sciences, Engineering, and Medicine’s (NASEM) consensus report on implementing high-quality primary care, based on each APM’s payment model (FFS, hybrid FFS with population-based payments, or global payments).12 Since the NASEM report also noted downside financial risk to primary care organizations as an added dimension of payment models, we considered upside-only versus both upside and downside risks as a separate feature cross-cutting with the payment model. This resulted in 6 reform type categories, shown in Figure 1 FFS upside, hybrid upside, FFS downside, hybrid downside, global, and entity-elected (since NextGen ACO entities selected from among FFS and population-based payment models).
APM groupings by reform type category. Abbreviations: ACO, Accountable Care Organization; APM, Alternative Payment Model; CPC, Comprehensive Primary Care; FFS, Fee-for-Service; MSSP, Medicare Shared Savings Program.
PCP Outcomes
We used previously validated FFS claims-based measures of PCP comprehensiveness, continuity, and coordination. Comprehensiveness measures included involvement in patient conditions, the extent to which a PCP was the physician most involved in treating the beneficiaries’ conditions;5 new problem management, the extent to which each PCP managed common primary care conditions for their beneficiaries;5 and range of services, a count of categories in which the PCP provided common services.6,9 For the continuity domain we used 2 widely-studied measures: first, fragmentation, based on the reversed Bice-Boxerman Continuity of Care Index which considers total ambulatory visits, total ambulatory providers (including primary care and specialists), and distribution of visits across providers, applying a validated threshold of ≥0.85 to quantify the proportion of the PCP’s beneficiaries with highly fragmented ambulatory care;20,21,23 and second, primary care continuity, which considers how many PCPs a patient saw and how many times they saw each PCP, determining for each PCP the proportion of beneficiaries meeting the validated primary care continuity threshold of 0.70.7 To evaluate coordination, we observed the proportion of a PCP’s beneficiaries with an eligible inpatient discharge who received a PCP follow-up visit within 7 days.27,30
For access, we developed a novel measure using information on beneficiaries’ index visits for new problems. This new first contact care measure quantified the extent to which a PCP was the first point of care for new problems arising among their beneficiaries.
We created these 7 outcomes for all TM beneficiaries seen by a PCP for primary care rather than retrospectively attributing beneficiaries to a single PCP; this allowed us to avoid misattributing primary care responsibility, and to examine outcomes across the full range of TM beneficiaries served by a PCP.
Statistical Analyses
We used a difference-in-differences analysis to identify the independent association of PCP participation in APM types during 2017 to 2022 with our outcomes of interest, comparing the change in mean outcomes in the intervention group (PCPs who join an APM during 2017 to 2022) from the 2016 baseline against the change in the comparison group (PCPs never in an APM during 2017 to 2022) over the same period. Because a PCP could join an APM in any of the years from 2017 to 2022, the intervention period varies across PCPs joining an APM. For example, a PCP who joined an APM in 2017 will be in the intervention group during 2017 to 2022, whereas another PCP who joined an APM for the first time in 2019 will be in the intervention group during 2019 to 2022. Models controlled for physicians’ characteristics including participation in other APMs, beneficiary attributes, and health system affiliation, and area-level characteristics. Because APM participants may differ from nonparticipants in innate motivation, institutional factors, or market characteristics,31 our models included TIN fixed effects; TIN identifies the entity that bills Medicare on behalf of, and distributes payment to, the PCP and thus is key to a PCP’s APM participation. Models clustered standard errors at the TIN level.
Because of the CMS transition to ICD-10 coding, we could not test for parallel trends in outcomes between APM participants and nonparticipants using data before 2016. Instead, we tested for parallel trends in outcomes during 2016 to 2018, comparing the subset of PCPs who first joined an APM in 2019 against those who did not join through 2019. Specifically, we examined whether there was a statistically significant divergence in trends between the 2 groups even before PCPs joined an APM. We implemented the parallel trends test by interacting indicators for years 2016 to 2018 with the APM treatment indicator and testing for joint significance of these interaction terms.
Given concerns about multiple comparisons across many models, we used a conservative 1% significance threshold for statistical testing.
Results
PCP Sample
Among 173,194 PCPs caring for TM beneficiaries in 2016, 48,148 were excluded for previous participation in a TM APM, 30,414 were excluded because they provided relevant services to fewer than 10 or more than 2,500 TM beneficiaries, and 1635 were excluded because their beneficiaries were predominately located outside the 50 states and the District of Columbia. The final sample contained 92,997 PCPs. Table 1 shows characteristics of the PCPs included in the analysis.
Selected Characteristics of Traditional Medicare PCPs, 2016–2022
PCP Participation in TM APMs
As shown in Table 1, 24.3% of PCPs observed in 2017 participated in a TM APM, increasing to 47.9% of observed PCPs in 2022; 38.1% of the 92,997 PCPs in the sample participated in an APM in at least 1 year. Among the 14 APMs observed in this analysis, the highest participation was in MSSP Track 1.
PCPs who participated in an APM were significantly more likely to be younger, in a metropolitan area, affiliated with a health system, and in a larger TIN than those who did not, and their beneficiaries were in higher-income areas and less likely to be high-risk.
PCP Outcomes
Table 1 also details PCP performance on access, continuity, comprehensiveness, and coordination across the study period. Of note, in the first year of the pandemic (2020), first contact care, involvement in patient conditions, and fragmentation improved; despite this, mean first contact care decreased by 12.1% over the study period, and mean fragmentation increased by 19.0%.
Parallel Trends
Parallel trends analysis suggested that the subset of PCPs entering APMs in 2019 were improving differentially on measures of comprehensiveness (involvement in patient conditions and new problem management) and continuity (fragmentation and primary care continuity) relative to PCPs not joining TM APMs; this pattern was seen among participants in ACO-directed payment APMs. The subset of PCPs joining primary care-directed payment APMs in 2019 had prior differential improvement only in the comprehensiveness measures of involvement in patient conditions and new problem management.
APM Association with Essential Features of Primary Care
Figure 2 summarizes findings on the association between any APM participation and essential features of primary care (Panel A). The outcome measures have different scales, so to aid interpretation we converted regression coefficients to a relative percent of the mean for comparison group PCPs. Participation in any APM was significantly associated (P < .001) with improvement in all outcomes. The magnitude of association ranged from 0.8% for new problem management and coordination to 6.8% for range of services.
TM APM association with improvement in primary care features, controlling for PCP characteristics, 2016 to 2022. Abbreviations: ACO, Accountable Care Organization; APM, Alternative Payment Model; CPC+, Comprehensive Primary Care Plus; PC, Primary Care; PCF, Primary Care First; PCP, Primary Care Physician; SSP, Shared Savings Program; TM, Traditional Medicare.
Figure 2 also summarizes findings from our analysis of PCP participation in APMs that direct payments to participating primary care (PC) practices or to an ACO entity (Panel B). For both APM types we observed a significant association in the expected direction with all primary care outcomes. PCPs in primary care-directed payment APMs performed significantly better than PCPs in ACO-directed payment APMs for first contact care, all 3 comprehensiveness measures, and primary care continuity (P < .001 for each).
Figure 3 summarizes findings for APMs based on the type of reform. For PCPs in hybrid upside-only APMs, we observed a significant association between APM participation and all 7 measures (P < .01). All other reform types were associated with significant improvement in some but not all measures.
TM APM with association improvement in primary care features, by reform type category, controlling for PCP characteristics, 2016 to 2022. Abbreviations: APM, Alternative Payment Model; FFS, Fee-for-Service; NS, Not Significant; TM, Traditional Medicare.
Excluding pandemic years from the models did not change our findings.
Discussion
Between 2017 and 2022, 38.1% of 92,997 PCPs serving TM beneficiaries participated in at least 1 of 14 APMs relevant to enhancing primary care. In our exploratory analyses, PCP participation in any of these APMs was associated with statistically significant improvement in claims-based measures of the core features of primary care: access, continuity, comprehensiveness, and coordination. However, these findings should be interpreted with caution because the modified parallel trends tests showed that PCPs entering APMs in 2019 were differentially improving on 2 measures of comprehensiveness and both continuity measures.
These 14 TM APMs varied considerably in payment structure as well as the types of organizations receiving these payments. We found payment models directed toward primary care practices were associated with significantly larger improvements on measures of access, comprehensiveness, and primary care continuity than APMs directed toward broader ACO entities.
This greater improvement among PCPs in primary care-directed APMs may not be surprising given commentaries suggesting enhanced payments in broader APMs like ACOs may be retained by the organization rather than transferred to practices.32 Interviews with PCPs and practice leaders echo concerns with allocation of resources by larger organizations to support practice transformation.27,33 Accordingly, the new ACO Primary Care Flex model has a mechanism to ensure model funds are directed to, and retained by, primary care practices.34
Prior research has also found that physician-led MSSP ACOs are more likely to produce savings than hospital-integrated MSSP ACOs,35 and reports suggest MSSP ACOs dominated by PCPs have achieved reduced specialist visits36 and greater savings,37 but we could not assess these associations with our data. We did find that PCPs in larger TINs had smaller improvements in some outcomes, but observed no relationship between changes in PCP outcomes and health system affiliation. Further research is required to better understand which settings are most conducive to using APM resources to enhance the key features of primary care. Such investigations may be particularly important given steadily increasing proportions of PCPs in health systems38 and increasing prominence of health systems in TM APMs.39
Only 1 of the 14 APMs studied was associated with large, consistently significant improvements in all 4 core primary care features. This model was the variation of the CPC+ model that combined FFS with per patient payments but had no downside financial risk to PCP practices. These findings are consistent with the NASEM report’s proposal that such “hybrid payment models” may be a particularly effective mechanism for supporting improvement across all the essential features of primary care. This finding is also consistent with commentators arguing for primary care-directed APMs that do not impose financial risk on practices.29
However, this finding was not observed in CPC+ evaluations22,40 which compared outcomes (including fragmentation, involvement in patient conditions, and new problem management) of CPC+ practice sites to matched comparison practice sites. Various differences between the analysis approach for the CPC+ evaluation and the analysis we conducted in this article were necessitated by CMMI's need to address model statutory requirements, limiting direct comparison of our findings with those in the CPC+ evaluation reports. Example differences include our focus on PCPs paid through the model rather than the CPC+ focus on participating practice sites, as well as our focus on all TM patients cared for by PCPs, not the subset attributed to practices.
Our observational study design also introduced methodological limitations. We attempted to account for selection bias by controlling for observable PCP, practice, beneficiary, and community characteristics. While we cannot control for unobserved characteristics such as appetite for improving performance on primary care features, we included TIN fixed effects in our models and tested whether PCPs who would participate in TM APMs in 2019 were differentially improving their primary care features from 2016 to 2018 versus those who would not.
We observe a sample of PCPs over time, but some are lost to follow-up as they retire, change specialty, or stop treating beneficiaries in TM. These PCPs may have performed differently from PCPs who remained in the sample, potentially biasing our results. Furthermore, aside from including year dummies in our models, we cannot account for differences in TM APM availability across both time and geography.
Because we limited to PCPs not already in an APM in 2016, we cannot generalize our findings to earlier adopters of APMs. In addition, we focused on outcomes for primary care physicians, excluding NPs and PAs because in TM these clinicians often bill “incident to” a PCP.41 Furthermore, it is difficult to identify if NPs or PAs are serving in a primary care role when they independently bill in a multispecialty TIN. Other data sources are needed to describe variations in the outcomes of these clinicians who comprise a growing share of the primary care workforce.
Another important limitation is that because of CMS’ 2015 change from ICD-9 to ICD-10 diagnosis coding we could not test for parallel trends on outcomes for our full sample of PCPs observed from 2016 to 2022. We explored this issue in a more limited sample of PCPs participating in TM APMs between 2019 and 2022 and found inconsistent patterns. PCPs joining primary care-directed APMs meaningfully failed this limited test for parallel trends for only 1 of 7 outcomes (new problem management), suggesting the association between primary care-directed APMs and our outcomes may be less attributable to selection bias than for ACO-directed APMs. Nonetheless we cannot definitively conclude that participation in any specific type of APM is causing the differential changes in primary care outcome measures that we observe.
Finally, CMS statutory and funding challenges have complicated implementing TM APMs that reliably enhance total payment amounts to PCPs.42 Thus, no TM APMs studied here are examples of these payment strategies, despite NASEM report emphasis on the importance of enhancing payments to primary care teams and of increasing overall spending devoted to primary care.12 Other policy changes will be required to achieve such reform of payment to primary care practices.
Conclusion
PCPs serving TM beneficiaries vary widely in the degree to which they deliver the essential features of accessible, comprehensive, continuous, and coordinated primary care. PCP participation in TM APMs was significantly associated with improved PCP delivery of these core primary care dimensions. Furthermore, only hybrid payment models with no financial risk showed a significant association with improvement in all essential features of primary care. While methodological limitations preclude asserting a causal relationship, these findings support continued efforts to improve primary care for Medicare beneficiaries through hybrid payments to practices without imposing downside financial risk on PCPs.
Acknowledgments
The authors would like to acknowledge Mathematica colleagues David J. Jones, Eli Michaels, Ann S. O’Malley, and Deliya Wesley for helpful suggestions in implementing the analysis and presenting findings; Mark Lee, Carol Razafindrakoto, Michelle Roozeboom-Baker, and Beny Wu for their assistance with the data and programming; and Adeline Wu for project management support. The authors also thank Robert L. Phillips, Jr., Executive Director, Center for Professionalism and Value in Health Care of the American Board of Family Medicine, for his feedback on the newly developed first contact care measure.
Notes
This article was externally peer reviewed.
Funding: This work was supported by the Commonwealth Fund (23-23721), a national, private foundation based in New York City that supports independent research on health care issues and makes grants to improve health care practice and policy, and Arnold Ventures (23-09788), a philanthropy dedicated to improving the lives of all Americans through evidence-based policy solutions that maximize opportunity and minimize injustice. The views presented here are those of the authors and not necessarily those of the Commonwealth Fund, Arnold Ventures, or their directors, officers, or staff.
Conflict of interest: The authors have no conflicts of interest to report.
- Received for publication March 28, 2025.
- Revision received May 12, 2025.
- Revision received July 21, 2025.
- Accepted for publication August 25, 2025.
