Impact of Hospital-Physician Integration on Medicare Patient Mix

ABSTRACT

Objectives: Hospital employment of physicians, often called hospital-physician vertical integration, has become widespread in health care delivery, but whether hospital employment tilts the case mix of physicians toward higher-complexity patients remains unknown.

Study Design: Cross-sectional and difference-in-differences analysis of 2014-2019 Medicare Standard Analytic Files.

Methods: We compared pre- and postemployment patient panels of primary care physicians who did and did not become hospital employees, analyzing changes in the prevalence of chronic conditions. We measured arthritis, depression, diabetes, hypertension, and ischemic heart disease. We also evaluated whether patients who were dropped from physician panels found alternative sources of primary care.

Results: Hospital-employed physicians treated patients of similar or better health; for instance, 54% of integrated physicians’ patients had 2 or more chronic conditions compared with 56% among independent physicians (P < .001). After becoming hospital employees, physicians treated approximately 10% fewer Medicare patients (–9.5%; 95% CI, –11.3% to –7.7%). Within physician panels, the prevalence of patients with 2 or more chronic conditions did not significantly change after employment relative to independent physicians (–1.1%; 95% CI, –2.3% to 0.2%). Approximately 37% of patients were dropped from physician panels after employment; these patients were less likely to find alternative primary care compared with those dropped from independent physician panels (P < .001).

Conclusions: Hospital employment of physicians resulted in neither a higher number nor a higher proportion of complex patients treated by integrated physicians, at least among traditional Medicare patients.

Am J Manag Care. 2025;31(4):In Press

Takeaway Points

• Advocates of hospital employment of physicians (often called hospital-physician vertical integration) argue that such physicians treat a higher-complexity case mix that justifies higher reimbursement, but evidence for the impact of employment on case mix remains thin.
• Cross-sectionally, hospital-employed physicians did not treat higher-complexity patients. Longitudinally, becoming a hospital employee did not increase physicians’ number or proportion of patients with multiple chronic conditions.
• Hospital-employed and independent primary care physicians appear to treat patients with comparable clinical complexity.

Hospital employment of physicians has rapidly accelerated in the past decade, with more physicians than ever working for hospital-owned practices.^1-5 Hospital employment of physicians, a term sometimes used interchangeably with hospital-physician vertical integration, entails hospital ownership of physician practices and/or direct employment of physicians. Hospital employment places physicians into care delivery settings where they will often have access to substantial clinical resources, including care management systems, information technology, and built-in networks of other clinicians.⁶

Certain patients might prefer care from health care systems featuring these integrated arrangements. For example, Mass General Brigham, a large integrated system in Massachusetts, is developing clinical centers to manage patients with chronic conditions such as atrial fibrillation. Such centers will coordinate appointments across specialty teams to improve patient care.⁷ Upon joining a hospital, employed physicians may find that the hospital’s centralized patient intake system assigns them more complex patients than when they worked in independent practice. In this way, the growing prevalence of hospital employment could help to meet the needs of the US population—and its growing prevalence of chronic health conditions—by placing patients with more complicated needs into the care of physicians with more resources. In addition, hospital advocates have often argued that their employed physicians treat more complex patients than independent physicians, justifying higher reimbursement.⁸ Yet evidence for the effect of employment on the case mix of physicians remains thin.

In this study, we compared pre- and postemployment panels of Medicare patients of physicians who became employed by hospitals with those of physicians who did not. We measured changes in the number and proportion of patients with hypertension, arthritis, ischemic heart disease, diabetes, and depression (5 highly common chronic conditions).⁹ Finally, we identified patients who stopped receiving care from their physician once their physician joined a hospital and determined whether these patients found alternative sources of primary care.

Our findings are relevant to the Federal Trade Commission, which is investigating hospital-physician integration to weigh the benefits and drawbacks of an increasingly consolidated provider system. This research provides evidence on an important potential benefit. Our findings are also relevant to patients, who may select providers based on how well physicians in different settings treat their conditions. This study can help other stakeholders better understand the potential winners and losers of the movement toward greater hospital employment of physicians.

METHODS

Data Sources and Study Sample

Our primary data sets were the Medicare Standard Analytical Files (SAF; available through a data use agreement with CMS), which contained claims, enrollment, and demographic information on a nationally representative 5% sample of Medicare patients. We also used the Doctors and Clinicians national downloadable data set and the Medicare Data on Provider Practice and Specialty data set (MD-PPAS), which were also available through a data use agreement. All data sets contained information from 2014-2019. We focused on Medicare patients specifically because this enables the use of nationally representative data among a population of patients who tend to have a high prevalence of chronic conditions. This provided the right setting to evaluate the clinical composition of patient panels across independent and hospital-employed physicians. Further, the policy conversation surrounding site-neutral payments (in which Medicare would pay for services in hospital-owned departments the same or nearly the same as it does for services in physician-owned practices) often centers on whether Medicare patients treated at hospital-owned vs physician-owned practices have different levels of clinical complexity that merit differential reimbursement (see further comment in the Discussion section). Taken together, these factors provided the rationale for our Medicare focus.

We extracted several key variables from these data sets. The granular claims data of the SAF, which included patient diagnosis codes, demographics, and the physician associated with each encounter, allowed us to calculate the proportion of patients in each physician’s panel with the 5 conditions listed earlier as well as whether each patient had 1, 2, or more than 2 of these chronic conditions. We identified specific patients by their unique beneficiary identifiers. We counted chronic conditions by applying Medicare’s Chronic Conditions Data Warehouse algorithms to the diagnosis codes in our claims data.¹⁰

Our exposure of interest, conceptually, was the economic primacy of a hospital (instead of a physician) in the delivery of physician care. We sought to assess any differences in case mix that resulted from physicians switching from independent practice to hospital employment. For our purposes, this encompassed outright hospital-physician vertical mergers in a traditional sense (eg, practice acquisitions in which physicians became employees) and individual physicians who left independent practice to become hospital employees. We determined whether physicians were hospital-employed using the claims-based algorithm of Neprash and colleagues¹¹ and supplemented this measure with a keyword search for hospital names in each physician’s primary organization drawn from MD-PPAS (eAppendix 1A [eAppendices available at ajmc.com]).¹²

To quantify compositional differences in panels of independent and hospital-employed physicians, we used the claims data to assess the prevalence of arthritis, depression, diabetes, hypertension, and ischemic heart disease among their attributed patients. We attributed each patient to the physician with whom they had the most primary care services in each year, comparable to the established attribution methodology that CMS uses for the Medicare Shared Savings Program.¹³ We could, therefore, trace differences over time with respect to the number of patients with chronic conditions (and with multiple chronic conditions) in each physician’s panel.

We defined exiting patients as those who were attributed to a physician in the year before the physician became hospital-employed but were not attributed the following year. We determined whether exiting patients appeared to find alternative sources of primary care (ie, had an encounter with another primary care physician [PCP]).

Our study population was PCPs, identified using provider specialty codes 01 (general practice), 08 (family practice), 11 (internal medicine), and 38 (geriatric medicine) from MD-PPAS data; hospitalists (specialty code C6) were excluded. In our main specification, physicians could enter or exit the sample; conclusions were unchanged if we enforced a balanced panel of physicians. We included patients 65 years and older living in any of the 50 US states or the District of Columbia with continuous Part A and Part B coverage (either 12 months or, among decedents, for all months the patient was alive). We did not exclude patients who died to avoid introducing bias if such patients were more common in one group or the other. This study was deemed exempt from review by Northeastern University’s institutional review board.

Statistical Analysis

First, we compared descriptive statistics between hospital-employed physicians and independent physicians. We hypothesized that we would observe higher rates of chronic conditions among patients of hospital-employed physicians. Second, we hypothesized that, beyond an unadjusted cross-sectional relationship, hospital employment would be associated with an increase in physicians’ proportion of patients with chronic conditions. This could occur in at least 2 ways: (1) patients with chronic conditions seek care from larger systems that are more likely to vertically integrate or (2) employed physicians might refer their less complex patients to other providers to focus their time on more clinically severe patients. To test this, we estimated a physician-level difference-in-differences model. Among physicians who began the study period as independent practitioners, some became hospital-employed (treatment group); others remained independent (comparison group). Recent methodological advances have highlighted the importance of avoiding problematic comparisons between early adopters and late adopters in difference-in-differences analyses.¹⁴ To avoid this pitfall, the comparison group included only physicians who remained independent throughout our study period (ie, never treated). In addition, we eschewed 2-way fixed-effects models and instead conducted our analysis using an event study design with a treatment indicator fully interacted with a set of pretreatment leads and posttreatment lags.¹⁵ We dropped the year in which a physician transitioned from independent to hospital-employed to avoid contaminating treatment and comparison time periods (ie, washout year). We used physician fixed effects. We included time-to-employment indicator variables as well as calendar-year indicator variables to control for secular trends in chronic condition prevalence. We implemented this analysis with ordinary least squares regression, clustering our SEs by physician. We also calculated an observation-weighted linear combination of postperiod coefficients to provide an average effect size. Our results were robust to a variety of sensitivity analyses (eg, excluding low-volume billers; a specification using stacked differences in differences) (eAppendices 1B-1C), and full regression specifications are available in eAppendices 2A-2C. Third, among the exiting patients mentioned above, we determined whether they found an alternative source of primary care. We assessed whether patients exiting the care of employed physicians exhibited different rates of access to primary care, which could be consistent with changes in patient panel following employment.

RESULTS

We identified 674,171 physician-years (152,435 unique physicians) that met inclusion criteria. Of the 112,651 unique physicians in 2014, 23,048 were hospital-employed; of the 110,737 in 2019, 25,475 were hospital-employed (Table 1). Across the full sample, 36% of patients had arthritis, 7% had depression, 32% had diabetes, 75% had hypertension, and 23% had ischemic heart disease. The mean patient age was 75 years and 59% were female.

Counter to expectations, hospital-employed physicians did not generally treat sicker patients. At both the beginning and end of our study period, a smaller proportion of patients of hospital-employed physicians had each of the chronic conditions we evaluated, except for depression. Among patients of hospital-employed physicians, approximately 54% had 2 or more chronic conditions; among patients of independent physicians, approximately 56% had 2 or more chronic conditions. In both groups, the mean patient age was 75 years and approximately 60% were female.

Physicians themselves were slightly different: Hospital-employed doctors were approximately 3 years younger on average, with 3 fewer years of experience; more were female (~40%-45% compared with ~30%-35%); and they had less overall Medicare volume (approximately half) compared with independent physicians. These differences underscored the importance of controlling for these factors in statistical models, which we did with physician fixed effects.

We next estimated our difference-in-differences models. We first tested the effect of integration on the total number of Medicare patients treated. We found that integration resulted in physicians treating fewer such patients (Figure 1). On average, physicians treated approximately 9.5% fewer Medicare patients post integration. Some hospital-employed physicians may work part time or take on management roles that result in less clinical care (or switch to commercial or Medicaid patients, thereby reducing the size of the Medicare panel).¹⁶ Correspondingly, post hospital employment, physicians also treated fewer Medicare patients—in sheer numbers—with each of the chronic conditions we studied; for instance, they treated 11% fewer patients with arthritis, 18% fewer patients with depression, and 15% fewer patients with diabetes (eAppendix 3).

Our main interest was whether physicians treated a higher percentage of patients with chronic conditions after becoming hospital employees. We found little evidence of this (Figure 2). Although there was a slight and nonsignificant increase in the proportion of patients with arthritis (average effect, 0.93–percentage point increase; 95% CI, –0.31 to 2.17), we found no effects among depression (–0.32; 95% CI, –0.99 to 0.35), diabetes (–0.79; 95% CI, –1.92 to 0.34), or ischemic heart disease (–0.60; 95% CI, –1.63 to 0.43). The proportion of patients with hypertension exhibited a small but statistically significant decrease (–1.69; 95% CI, –2.83 to –0.55; P = .038).

We also tested whether hospital employment changed physicians’ panels based on the proportion of their patients with multiple chronic conditions (Figure 3). In the year preceding hospital employment, 12% of patients had 0 chronic conditions. After hospital employment, this proportion increased (not significantly) by 0.47 percentage points (95% CI, –0.46 to 1.39; P = .32). In the year preceding hospital employment, 31% of patients had 1 chronic condition. After hospital employment, this proportion increased (not significantly) by 0.58 percentage points (95% CI, –0.59 to 1.75; P = .33). In the year preceding hospital employment, 57% of patients had 2 or more chronic conditions. After hospital employment, this proportion decreased (not significantly) by approximately 1 percentage point (–1.05; 95% CI, –2.31 to 0.22; P = .105). In the year preceding hospital employment, 24% of patients had 3 or more chronic conditions. After hospital employment, this proportion decreased (with marginal statistical significance) by approximately 1 percentage point (–1.11; 95% CI, –1.93 to 0.15; P = .092). In summary, hospital employment had little effect on the proportion of patients with chronic conditions; if anything, physicians treated fewer patients who had multiple chronic conditions following hospital employment.

Finally, we tracked the patients who stopped receiving care from their physician after their physician became hospital-employed and determined whether they found alternative sources of primary care. We compared their rates of finding alternative primary care with those of patients who stopped receiving care from independent physician panels (Table 2). In any given year, approximately 37% of patients exited their PCP’s panel (ie, had no encounter in the following year). This figure was the same to within a tenth of a percentage point across both hospital-employed and independent physicians. This implies a relatively high baseline churn among Medicare patients and their PCPs. Among those who exited hospital-employed physician panels, 79% had an encounter with a primary care clinician in the year after their physician became hospital-employed. This was significantly less (P < .001) than the rate among patients who had left independent physician panels, among whom 85% found alternative sources of primary care.

DISCUSSION

In this study, we found no cross-sectional evidence that hospital-employed PCPs treated patients with more complex conditions compared with independent physicians. In longitudinal analysis, PCPs who became hospital-employed treated fewer total Medicare patients and fewer patients with chronic conditions. After their primary care doctor became hospital-employed, approximately one-third of patients were no longer included in that physician’s panel, and among that third, 21% went without any primary care the next year. Taken together, these results imply that despite the potential for hospital employment of physicians to connect high-needs patients to high-resource providers, in practice, employment appeared to have little impact on patient mix.

Our results run counter to the prevailing wisdom that hospital-employed physicians treat more clinically complex patients. This idea is sometimes used as a rationale for higher payment for services provided at hospital-owned sites instead of being paid at the (lower) physician fee schedule rate. This rationale was raised, for example, at the policy panel on site-neutral payment that was hosted by KFF in June 2024: If hospital outpatient departments treat patients whose complexity induces costs that would go unrecognized by the physician fee schedule, it would be fair for hospitals to receive higher payment as compensation.¹⁷ The present study finds little to support that rationale, at least among these common conditions for Medicare fee-for-service patients. We caveat this conclusion by noting that our study accounts for medical conditions but not socioeconomic measures of complexity, which are also salient to discussions of fair reimbursement. Research conducted by the Medicare Payment Advisory Commission (MedPAC) suggested that hierarchical condition category risk scores (a measure of medical complexity) were higher among hospital outpatient department patients compared with physician office patients, but that the difference was small and payment adjustments for patient severity were unneeded.¹⁸ MedPAC’s study does not directly compare with ours; we focused on PCPs and specific chronic conditions, whereas the MedPAC sample inclusion criteria were unspecified and its variable of interest was risk score. Nevertheless, to the extent that MedPAC’s report suggests minimal differences between these 2 settings, our findings extend their results by showing that, in the primary care context, there was neither cross-sectional nor longitudinal evidence to suggest more complex patient panels among vertically integrated physicians.

Our results also extend the literature on hospital employment of physicians and quality. Work by previous investigators found mostly small or null effects of vertical integration on quality measures.^19-21 Our study adds further context that (1) hospital-employed physicians may be treating healthier patients in the first place, which should have (if anything) biased prior results toward better outcomes associated with hospital employment, and (2) hospital-employed physicians may treat fewer patients after becoming hospital-employed. Furthermore, some patients may be dropped from panels, and their quality of care should be considered as well. Researchers should try to account for which patients join and leave physician panels rather than using repeated cross-sectional designs, as they fail to account for patients who drop from panels. Converting a physician from independent to hospital-employed status affects 3 groups: patients who join, stay in, and exit a physician’s panel after the employment change. One might hypothesize that quality would improve in one group (eg, the stayers) while declining in another (eg, the exiters). Our results show that the exiters group could prove to be sizable—in our sample, it was approximately one-third of a physician’s panel. Studies that miss this nuance risk bias toward a spurious positive effect of hospital employment on quality.

Last, we found suggestive evidence of a disproportionate reduction in care among patients with chronic conditions (ie, a 9.5% decline in total Medicare patients but a 14% decline in patients with chronic conditions). One reason could be that hospitals may have incentives to phase out more complex patients if sicker patients hurt their scores on pay-for-performance programs or hospital value-based purchasing arrangements. Alternatively, patients with chronic conditions may need frequent access to care, and integration may reduce PCP availability.¹⁶ If one of the effects of integration is increased wait times, healthier patients may be able to handle the wait, whereas sicker patients may not. These are important areas for future work.

Limitations

Most physicians treat patients across multiple payers; given our study setting of traditional Medicare, our observation of patient panel dynamics is, by extension, incomplete. We could not account for turnover across payers (eg, among commercial insurers, Medicare Advantage, and Medicaid). In supplemental analysis, physician employment was not significantly associated with increases in the proportion of their patients who were dually eligible for Medicaid (eAppendix 2D). However, these results are nevertheless important, considering that 38 million beneficiaries depend on the traditional Medicare program²²; further, the traditional Medicare context enabled us to identify the upper bound of the benefit that we sought to study. Because we saw no increased access to primary care for patients with chronic conditions in traditional Medicare, in which all willing providers are in network, it is even less likely we would see an effect with patients covered by commercial plans, nearly all of which limit patients to a smaller network of providers. Our analysis did not include hospital characteristics. Unique hospital attributes could mediate the observed results (eg, for-profit and nonprofit hospitals may differ in their approach to assigning patient caseloads to their employed physicians).

CONCLUSIONS

Our findings illuminate an important dynamic in hospital-physician relations. If hospital employment of physicians shifted healthier patients to independent physicians while making room for patients with more severe health needs to be seen by hospital-based physicians, it could improve allocative efficiency in the health sector: Patients with more needs would be matched to physicians with more resources. These additional resources could potentially translate to better care. Policy makers keen to better understand and potentially limit such vertical integration, including MedPAC, the Federal Trade Commission, and the Department of Justice, should note that we find nothing to suggest such an improvement. Further research is needed to identify whether and how the increasingly consolidated health care system can supply enough physician services overall, for publicly insured patients, and for those with chronic conditions. 

Author Affiliations: Department of Health Sciences, Bouvé College of Health Sciences (BP, FA), Center for Health Policy and Healthcare Research (BP, GJY), and D’Amore McKim School of Business (GJY), Northeastern University, Boston, MA.

Source of Funding: Dr Post was supported by the Agency for Healthcare Research and Quality under award K01HS029278.

Author Disclosures: Dr Post reports receiving an Agency for Healthcare Research and Quality career development grant. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (BP, FA, GJY); acquisition of data (BP, FA, GJY); analysis and interpretation of data (BP, FA, GJY); drafting of the manuscript (BP, GJY); critical revision of the manuscript for important intellectual content (BP, FA, GJY); statistical analysis (BP, FA, GJY); provision of patients or study materials (GJY); and administrative, technical, or logistic support (GJY).

Address Correspondence to: Brady Post, PhD, Bouvé College of Health Sciences, Northeastern University, 336 Huntington Ave, 316H Robinson Hall, Boston, MA 02115. Email: b.post@northeastern.edu.

REFERENCES

1. Nikpay SS, Richards MR, Penson D. Hospital-physician consolidation accelerated in the past decade in cardiology, oncology. Health Aff (Millwood). 2018;37(7):1123-1127. doi:10.1377/hlthaff.2017.1520

2. Baker LC, Bundorf MK, Kessler DP. Vertical integration: hospital ownership of physician practices is associated with higher prices and spending. Health Aff (Millwood). 2014;33(5):756-763. doi:10.1377/hlthaff.2013.1279

3. Updated physician practice acquisition study: national and regional employment changes in physician employment 2012-2018. Physicians Advocacy Institute. February 2019. Accessed August 28, 2023. https://www.physiciansadvocacyinstitute.org/Portals/0/assets/docs/021919-Avalere-PAI-Physician-Employment-Trends-Study-2018-Update.pdf

4. Avalere Health. COVID-19’s impact on acquisitions of physician practices and physician employment 2019-2020. Physicians Advocacy Institute. June 2021. Accessed August 28, 2023. https://www.physiciansadvocacyinstitute.org/Portals/0/assets/docs/Revised-6-8-21_PAI-Physician-Employment-Study-2021-FINAL.pdf

5. Kocher R, Sahni NR. Hospitals’ race to employ physicians — the logic behind a money-losing proposition. N Engl J Med. 2011;364(19):1790-1793. doi:10.1056/NEJMp1101959

6. Burns LR, Goldsmith JC, Sen A. Horizontal and vertical integration of physicians: a tale of two tails. Adv Health Care Manag. 2013;15:39-117. doi:10.1108/s1474-8231(2013)0000015009

7. Bartlett J. This next step of integrating Mass General Brigham will be the toughest one yet. Boston Globe. September 27, 2022. Accessed September 28, 2022. https://www.bostonglobe.com/2022/09/27/metro/this-next-step-integrating-mass-general-brigham-will-be-toughest-one-yet/

8. Health subcommittee legislative hearing: “Lowering Unaffordable Costs: Legislative Solutions to Increase Transparency and Competition in Health Care.” US House Energy and Commerce Committee. April 26, 2023. Accessed April 8, 2024. https://energycommerce.house.gov/events/health-subcommittee-legislative-hearing-1

9. Medicare chronic condition charts. Chronic Conditions Data Warehouse. Accessed September 14, 2022. https://www2.ccwdata.org/web/guest/medicare-charts/medicare-chronic-condition-charts

10. Chronic Conditions Data Warehouse. Accessed November 21, 2022. https://www2.ccwdata.org/web/guest/home/

11. Neprash HT, Chernew ME, Hicks AL, Gibson T, McWilliams JM. Association of financial integration between physicians and hospitals with commercial health care prices. JAMA Intern Med. 2015;175(12):1932-1939. doi:10.1001/jamainternmed.2015.4610

12. Post B, Norton EC, Hollenbeck B, Buchmueller T, Ryan AM. Hospital-physician integration and Medicare’s site-based outpatient payments. Health Serv Res. 2021;56(1):7-15. doi:10.1111/1475-6773.13613

13. Medicare Shared Savings Program shared savings and losses and assignment methodology specification. CMS. February 2021. Accessed July 6, 2022. https://www.cms.gov/files/document/medicare-shared-savings-program-shared-savings-and-losses-and-assignment-methodology-specifications.pdf-0

14. Goodman-Bacon A. Difference-in-differences with variation in treatment timing. J Econom. 2021;225(2):254-277. doi:10.1016/j.jeconom.2021.03.014

15. Cunningham S. Causal Inference: The Mixtape. Yale University Press; 2021.

16. Post B, Hollenbeck BK, Norton EC, Ryan AM. Hospital-physician integration and clinical volume in traditional Medicare. Health Serv Res. 2024;59(1):e14172. doi:10.1111/1475-6773.14172

17. Levitt L, Cooper Z, Levinson Z, Thompson A. The health wonk shop: what’s behind the buzz about site-neutral payments? KFF. June 17, 2024. Accessed July 10, 2024. https://www.kff.org/other/event/june-17-web-event-the-health-wonk-shop-whats-behind-the-buzz-about-site-neutral-payments/

18. Report to the Congress: Medicare and the Health Care Delivery System. Medicare Payment Advisory Commission. June 15, 2023. Accessed September 10, 2024. https://www.medpac.gov/wp-content/uploads/2023/06/Jun23_MedPAC_Report_To_Congress_SEC.pdf

19. Cuellar AE, Gertler PJ. Strategic integration of hospitals and physicians. J Health Econ. 2006;25(1):1-28. doi:10.1016/j.jhealeco.2005.04.009

20. Scott KW, Orav EJ, Cutler DM, Jha AK. Changes in hospital–physician affiliations in U.S. hospitals and their effect on quality of care. Ann Intern Med. 2017;166(1):1-8. doi:10.7326/M16-0125

21. Colla C, Yang W, Mainor AJ, et al. Organizational integration, practice capabilities, and outcomes in clinically complex Medicare beneficiaries. Health Serv Res. 2020;55(suppl 3):1085-1097. doi:10.1111/1475-6773.13580

22. Total number of Medicare beneficiaries by type of coverage. KFF. Accessed June 15, 2021. https://www.kff.org/medicare/state-indicator/total-medicare-beneficiaries/