Physician mergers involve 38% of doctors, substantial health system participation, and frequent serial acquisition

Abstract

We describe the incidence of mergers between US physician firms. We considered the role of health systems in physician merger activity and the extent to which individual firms engage in multiple acquisitions, or “roll-up” behavior. We used a unique collection of subpoenaed data from 6 insurers in 15 states, which account for approximately 80% of commercially insured members in those states between 2015 and 2020. Over 6 years, we observed 2019 mergers between physician firms that spanned from small single-specialty practices to large groups affiliated with health systems. Approximately 20% of the roughly 15 000 firms were involved in a merger. Of the nearly 400 000 doctors in our sample, 38% were employed by firms involved in a merger, although only 3.8% of the doctors were part of an acquired group. Sixty-three firms made more than 5 acquisitions, and 3 made more than 20 acquisitions. Health care systems, responsible for approximately 40% of mergers, frequently engaged in serial acquisitions. Acquired physician firms are typically small (<10 doctors), and much smaller than the acquiring firm. We document that many physician mergers occurred in 15 states over 5 years. A significant fraction of the physician mergers we identified, especially those involving health systems, exhibited characteristics that could indicate the potential for competitive harm. Our results suggest that researchers should examine the degree to which these competitively suspect mergers have affected competition.

Introduction

After the Federal Trade Commission (FTC) and Department of Justice collectively lost 7 consecutive hospital merger challenges, the then-FTC Chairman Muris launched a project to reinvigorate US hospital merger enforcement. A major component of that effort was the FTC's Hospital Merger Retrospective Project, which demonstrated the harmful effects of hospital mergers on hospital prices, both theoretically and in practice, while also providing evidence that the studied mergers did not increase the quality of care at the merging hospitals (see volume 18, Issue 1 of the International Journal of the Economics of Business for a special issue devoted to the topic). Subsequent studies reinforced the potentially negative consequences of mergers of competing hospitals.¹

Although a large literature has demonstrated that mergers of competing hospitals can increase health care prices and diminish health outcomes, there is comparatively little literature examining the impact of mergers of physician practices on competition despite evidence suggesting substantial consolidation over the last decade.^2-4 A small, but growing, literature has begun to examine these acquisitions. Capps et al⁵ identified many acquisitions using commercial claims data, and found that a number of those deals occurred in already-concentrated areas. Koch and Ulrick⁶ found that a merger of orthopedists in southeastern Pennsylvania led to substantial price increases following the merger. Zhang et al⁷ found that the acquisition-induced transfer of primary care patients from small firms to large firms decreased primary care visits and spending overall for the Medicare population.

Another stream of research has focused on the effects of hospital and private equity acquisitions of physician practices.^8,9 Private equity acquisitions have also been shown to increase prices and change how care is provided.^10-13

The goal of this study is to describe recent physician merger activity. We used data collected using an FTC subpoena from 6 major insurers from 15 states with more than 60 million covered members each year (including an estimated 80% of commercially insured members' claims data) to identify physician mergers.¹⁴ We observed 2019 mergers between 2015 and 2020. We found that 3.8% of doctors in our sample were employed by practices that were acquired, and that 38% of doctors worked for a firm that acquired 1 or more practices during the study period.

We next provide some descriptive information about these acquisitions. First, we examined how much acquisition activity is undertaken by health care systems making use of a non-public dataset from the Agency for Healthcare Research and Quality (AHRQ) that links health care systems to physician practice firms. Next, we measured the relative size of the acquired and acquiring firms. Third, we examined how the prevalence of mergers varies by medical specialty.

We also conducted an analysis to measure the potential competitive significance of physician mergers using the estimated diversion ratios between merging firms. A diversion ratio is a commonly used measure of competition that measures the closeness of competition between 2 firms. Diversion ratios between the merging firms have been shown to be predictive of merger price effects in studies examining consummated hospital mergers.^15-17

Finally, we examined the extent to which firms acquiring physician practices engage in serial acquisition. While we found that most acquiring firms engaged in a single merger during our sample period, we did find evidence that a significant fraction of firms engage in serial acquisitions.

Taken together, our findings suggest that many consummated physician mergers may have raised competitive concerns. Some mergers combined firms with relatively high diversions. Moreover, there is evidence that some firms are combining many firms in the same medical specialty.

Data and methods

Data

Our dataset consists of professional, outpatient, and inpatient claims data from 6 insurance companies—Aetna, Anthem, Cigna, Florida Blue, Health Care Services Corporation, and UnitedHealthcare. The data contain claims information for up to 63 million members (including more than 51 million with commercial insurance) in 15 states over 6 years (2015–2020). The plans include those that are individual, employee-sponsored, Medicare Advantage, and Medicaid claims from these firms. The study states include Colorado, Florida, Georgia, Illinois, Indiana, Kentucky, Maine, Missouri, Montana, Nevada, New Hampshire, New Mexico, Ohio, Oklahoma, and Texas. We estimate that our sample includes approximately 82% of the commercially insured population in the sample states during this time. The states and payers were chosen to create a sample with broad geographic scope, including states from each of the 4 primary US Census divisions (namely, the Midwest, Northeast, South, and West), as well as states having substantial rural and urban populations.

These data include claims for all commercial plans; thus, they account for employer-sponsored plans and individual market plans. The data also include government-sponsored plans, such as Medicare Advantage and Medicaid Managed Care, insofar as the payers offer these plans in the states and time we studied.

The claims data report a variety of information required to perform our analysis: a unique identifier for each patient and where they live (5-digit ZIP code), an identifier of the person or firm providing a medical service (National Provider Identifier [NPI]), a code that identifies the firm that billed the insurer (Tax Identification Number [TIN]), the services provided (Current Procedural Terminology [CPT] codes), the underlying medical diagnoses that led to this course of care (International Classification of Diseases, Tenth Revision [ICD-10], codes), the amount paid by the patient and their insurer, and in what kind of facility were the services rendered (eg, a medical office, an ambulatory surgical center, or an inpatient hospital facility). Additional details can be found in the online Appendix.

We supplemented the claims data with AHRQ's Health System Compendia for 2016 and 2018 and the Centers for Medicare and Medicaid Services’ (CMS's) Medicare Data on Provider Practice and Specialty (MD-PPAS). The AHRQ data allow us to designate certain TINs as being part of a health system, which we used in our analysis. The 2016 Compendium was matched to claims data for 2015 through 2017; the 2018 Compendium was used for data for 2018 and later.

The MD-PPAS assigns providers to specialties, so we can distinguish between providers who provide similar services (eg, cardiologists) and those who don’t (primary care and orthopedists). For non–primary care providers, we grouped providers according to their 2-digit Provider Enrollment, Chain and Ownership System (PECOS) specialty code. We coded all primary care providers together into 1 specialty, except for pediatrics. Pediatricians are of particular interest, because they serve a different patient population for primary care than other primary care providers, and because they are very unlikely to be enrolled in Medicare, we can provide new measurement on their role in the US health care system and the acquisition trends therein.

Methodology

We created a panel that identifies every doctor's (identified by a provider NPI) primary employer (TIN) in a quarter using all of the claims data from all plans. When we characterize the relationships between providers and their employers, we are primarily referring to the relationships between NPIs and TINs. That is, both small independent, single-specialty practices as well as large multispecialty practices are both firms, even though their scope and scale may differ substantially. In some of our analyses, we make use of AHRQ's Health System Compendia to identify the multiple TINs controlled by a health system. In this case, the “firm” corresponds to the health care system that controls the affiliated TINs.

We used this doctor/employer panel, which includes approximately 15 000 TINs and 400 000 NPIs, to identify TINs that exit the panel, and labelled these exiting TINs as potentially acquired firms. We limited the sample of potentially acquired firms to those with a median number of doctors of more than 2 during the sample period to avoid mismeasurement of mergers. For example, a transition of a single-doctor firm (TIN) to a larger practice likely represents a change of employer rather than a practice acquisition. We then examined the TINs that doctors move to after their previous employer (TIN) exits the panel. When we observed that at least 80% of doctors moved to a single TIN, we concluded that a merger had taken place. In this instance, we assumed that the destination TIN has purchased the firm associated with the TIN that exited the panel. Finally, we used information about the doctor's previous employer (the acquired TIN) and current employer (the acquiring TIN) to describe the mergers that have taken place. This enables us to identify those acquisitions where the acquiring firm absorbs the other and operates the acquired firm using its TIN. There may be other changes in ownership where the acquiring firm continues to operate the subsidiary using its TIN. Our methodological approach cannot identify these types of transactions.

We also calculated diversions between merging firms for our analysis. As described in the 2023 Merger Guidelines, diversion ratios are a common tool used to measure customer substitution (or closeness of competition) between 2 goods or 2 firms.¹⁸ The diversion ratio is the fraction of unit sales lost by the first product due to a change in terms, such as an increase in its price, that would be diverted to the second product. Larger diversions between 2 practices imply that the practices are closer substitutes. We conducted this calculation for every merger (separately by medical specialty) for both sides of an acquisition; we took the larger of the 2 diversions as the relevant diversion for each merger. Because the acquiring firm is typically much larger than the acquired firm, in most cases (75%) the diversion we used is that of the acquired to the acquiring firm. The minimum TIN size restriction we used when identifying mergers was not used when we calculated diversions; that is, TINs with fewer than 2 doctors are included when calculating diversions.

We estimated diversions between firms based on share calculations at the ZIP code level. For example, if 3 firms (A, B, and C) each provided 2 units of care to beneficiaries in a particular ZIP code, the diversion from A to B would be B/(B + C) = 2/(2 + 2) = 1/2, because the relative share of B and C reflects the relative attractiveness of B and C for patients who are no longer receiving care from A. The diversions we report in the paper are the average diversion in the ZIP codes where the practices provide service, weighted by the amount of care they provide in each ZIP code (eg, if group A provides 2 units of care in ZIP code Y with diversion A to B of 0.6, and 1 unit of care in ZIP code Z with diversion of A to B of 0.9, then the diversion of A to B is 2/3 × 0.6 + 1/3 × 0.9 = 0.7.

The output for each TIN/specialty used in calculating diversions was set to the total number of Relative-Value Units (RVUs) that the TIN/specialty provides in a quarter, except for those specialties such as anesthesiology, which do not use an RVU-based reimbursement system. For those specialties, we used total revenue in calculating diversions. We only included output measures for employer-sponsored and individual plans in calculating diversion; that is, we excluded Medicare Advantage and Medicaid claims. Antitrust complaints have focused on competition among providers for participation in nongovernment plans since the prices for nongovernment plans are set as a result of market-based negotiations.¹⁹

Diversions are a key input used by many merger simulation tools to forecast a merger's predicted price increases. For example, in the hospital context, if we assume that both hospitals and insurers are equally sophisticated negotiators, all hospital prices are the same pre-merger, and that hospitals earn a margin of 50% (eg, for every $100 of revenue a hospital generates, it incurs $50 in costs), then widely used models of hospital–insurer negotiations would predict a price increase to be one-fourth of the diversion between the merging hospitals. That is, a merger between 2 hospitals with 20% diversion would be predicted to increase prices by 5%.²⁰

Limitations

The claims data we used cover only 15 states. While we have substantial coverage of physicians in those states (drawn from claims data for >80% of all commercial covered lives), our sample does not include some large states, such as California and New York. We also observed specialties not covered extensively in commonly used data sources (Medicare), such as pediatrics or obstetrics. However, our sample has relatively less coverage of patients with government-provided insurance (Medicare and Medicaid). As a result, since certain physicians exclusively provide care to the elderly or poor populations, we may not be capturing those physicians' movements across firms.

Our definition of mergers may be under-inclusive for at least 3 reasons. First, there may be instances where firms merge and fewer than 80% of the doctors in an original TIN move on to an acquiring TIN in the next quarter. We adopted the 80% threshold to be consistent with the firm dynamics literature in economics.²¹ Alternatively, other researchers have used 60% of doctors moving into a new TIN as a threshold.¹⁰ As we have defined it, transactions where between 60% and 80% of the doctors move to a single firm (TIN) would not qualify as a merger.

Second, we used TINs to measure the boundaries of the firm—which providers are part of a company and which are not. As mentioned above, TINs are routinely used for this purpose. However, TINs do not perfectly capture the boundaries of firms since firms may operate multiple TINs. It is possible that some firms engaging in serial acquisitions may continue to operate the acquired firm using the acquired firm's TIN. If this practice is common, our method will undercount mergers. That said, some research has shown that, while not perfect, TINs can be used to identify practice ownership appropriately, albeit with some possibility of error.²²

Third, we made the choice to exclude practices with 2 or fewer doctors from our count of mergers. We made this choice because transitions of very small practices may reflect changes in an employment relationship rather than traditional merger activity. If, however, we included these transitions, we would likely observe more mergers. For instance, a recent American Medical Association (AMA) report estimates that the fraction of doctors in solo practices (presumably billing to a single TIN) fell from 16.5% to 14% between 2016 and 2020.²³

Results

In Table 1, we see that mergers occur frequently: we observed 2019 mergers in the study states. The number of mergers occurring each year has been relatively constant between 2015 and 2020. Overall, 20% of approximately 15 000 firms (TINs) were involved in a merger. A much larger fraction of doctors, however, were employed by practices that were involved in a merger. Approximately 38% of the doctors were employed by firms that acquired other doctors, while 3.8% of doctors were employed by an acquired firm. Because doctors could be in acquired firms or acquiring firms multiple times, the results over the entire sample are not the average of the yearly samples. Finally, we found that 40% of mergers taking place in the study states were associated with a health care system.

Table 1.

The number of physician mergers by year and share where the acquiring firm was a health system.

	Mergers		Doctors
Year	Total	System	Acquiring	Acquired	Share firms
2015	318	45%	20%	1.0%	4%
2016	323	40%	20%	1.0%	4%
2017	355	37%	16%	1.0%	5%
2018	372	37%	16%	1.1%	5%
2019	377	40%	16%	1.1%	5%
2020a	274	39%	17%	0.6%	4%
Total	2019	40%	38%	3.8%	20%

Source: Authors' analysis of commercial claims data. We also calculated the share of doctors in our sample states that were either in an acquiring firm or an acquired firm, as well as the share of firms overall in a merger during the year. Because doctors can both be in an acquired firm and an acquiring firm within the same year and can be in a firm that is either acquired or acquiring multiple times in different years, the shares reported here do not necessarily add up across rows or down columns.

^aThe dip in 2020 is likely attributable to the fact that we cannot identify mergers that took place in the fourth quarter, since we do not have claims data for 2021.

We found that the acquiring firm is typically much larger than the acquired firm. We show this using 2 calculations: first, by measuring the number of doctors employed by each firm in the period before the merger, and second, by comparing the total revenues (payments made by insurers in our sample and Medicare Fee-for-Service reimbursement as reported in MD-PPAS) in the calendar year prior to the acquisition. By both measures, the acquiring firms are much larger than the acquired firms: the average number of doctors in acquiring firms is 232, with a median of 53, suggesting a heavily skewed distribution of very large firms acquiring other firms. The average number of doctors in acquired firms is only 9, with a median of 5. Similarly, the average revenue of an acquiring firm is $70 M (million), with a median of $21 M, compared with a mean of $4.3 M and median $1.8 M for acquired firms.

We next assessed the extent to which mergers combine providers that compete by calculating a standard measure of horizontal competition, the diversion ratio, for each merger. As mentioned above, we focused on the larger of the 2 diversion ratios for each acquisition (acquirer to acquired, and acquired to acquirer).

In Table 2, we report the number of horizontal overlaps across all physician specialties with diversions in 4 separate categories: diversions from 0% to 5%, 5% to 10%, 10% to 15%, and 15% or larger. (See Appendix Table A1 for a list of all of the medical specialties included in this analysis.) Over half (1260) of the horizontal overlaps had diversions of less than 5%, while one-fifth (434) had diversions of more than 15%.

Table 2.

The number of horizontal overlaps with diversions in 4 categories: less than 5%, 5% to 10%, 10% to 15%, and greater than 15%.

		Diversion								Health system
Specialty	Total	<5%		5% to 10%		10% to 15%		>15%
All	2184	1260	0.58	302	0.14	188	0.09	434	0.20	950	0.43
Primary care	582	369	0.63	78	0.13	53	0.09	82	0.14	246	0.42
Pediatric medicine	240	168	0.70	20	0.08	12	0.05	40	0.17	110	0.46
Emergency medicine	187	119	0.64	27	0.14	9	0.05	32	0.17	66	0.35
Anesthesiology	157	94	0.60	20	0.13	8	0.05	35	0.22	46	0.29
Obstetrics/gynecology	150	67	0.45	29	0.19	17	0.11	37	0.25	76	0.51
General surgery	117	53	0.45	18	0.15	16	0.14	30	0.26	68	0.58
Diagnostic radiology	107	67	0.63	16	0.15	10	0.09	14	0.13	31	0.29
Orthopedic surgery	98	41	0.42	19	0.19	9	0.09	29	0.30	49	0.50
Cardiology	74	36	0.49	13	0.18	7	0.09	18	0.24	41	0.55
Gastroenterology	54	24	0.44	8	0.15	5	0.09	17	0.31	20	0.37
In a system	950	436	0.46	167	0.18	108	0.11	239	0.25
Not in a system	1234	824	0.67	135	0.11	80	0.06	195	0.16

Source: Authors' analysis of commercial claims data. These diversion groupings are reported only for the specialties with at least 50 mergers (see Appendix Table A1 for a list of specialties included in the All category but not separately listed here). The share of mergers in each diversion category within a specialty row is shown in italics next to the number of mergers.

In Table 2, we also report these numbers separately by specialty, for which we observed at least 50 mergers. There were 582 horizontal overlaps in primary care; 369 had diversion of less than 5%, but 82 were larger than 15%. For many of these specialties (anesthesiology, orthopedic surgery, cardiology, and gastroenterology), dozens of horizontal overlaps had diversions greater than 15%.

Forty-three percent (950 out of 2184) of the specialty overlaps involved a health system. Health system acquisitions are of policy interest: research has found that, when doctors are acquired by health systems, reimbursement for commercial payers and Medicare increases, while not improving the health outcomes for the latter group.²⁴ Health system acquisitions tended to have diversions that were larger than non-system acquisitions. The raw number of system deals in the 3 largest diversion groups is larger for the system deals vs the non-system deals, even though there are fewer system deals overall. In other analysis (not reported in these tables), we found that system deals have higher diversion for all specialties that we report in Table 2, with the exception of Gastroenterology and Orthopedic Surgery.

Recent research has studied the presence of serial acquirers of providers in some specialties.^25,26 To help generalize these findings, we measured the extent to which physician practices engage in serial acquisitions for all specialties in our sample. Figure 1 shows the number of acquisitions made by acquirers. Most acquirers make only a single acquisition. A significant number of firms, however, are engaged in serial acquisitions. In some cases, serial acquisition involves a single-specialty practice purchasing rivals operating in the same specialty: 20 single-specialty practices purchased between 2 and 5 other practices. Health systems were also involved in roll-ups, with 24 making more than 10 and 4 acquiring more than 20 (see Appendix Table A2 for more details).

Figure 1.

The number of firms, separately counted by the number of acquisitions they make. Source: Authors' analysis of commercial claims data. These firm counts are reported separately by whether the acquirer is an individual firm or part of a health system.

As noted above in our discussion of the study's limitations, our methodology only identifies mergers where the acquiring firm stops using the acquired firm's TIN. If firms engage in serial acquisitions and continue using the acquired firm's TIN, our methodology will not identify these transactions and will understate the degree of serial acquisitions.

Discussion

In this study, we provide a descriptive analysis of the more than 2000 mergers of physician firms in the study states that took place between 2015 and 2020. Many of these mergers are between firms that are competitors, with an estimated diversion of greater than 5%, 10%, or 15%. Mergers among non–primary care specialties and acquisitions where health systems are involved tend to have larger diversions. Most acquiring firms purchased a single firm during our sample period; however, there is some evidence of “roll-up” behavior, where a firm acquires several physician firms over time. Health systems are particularly active in “rolling-up” physician practices.

Conclusion

US health care markets have been undergoing a dramatic restructuring over the last few decades. A large literature has documented these changes in the hospital sector and shown that horizontal consolidation frequently leads to substantial increases in provider prices and, in many cases, reductions in health quality.

These results provide a first look at the extent of physician merger acquisitions and their varying potential harm to competition. More research is needed to precisely detect when and where enforcement resources should be spent to identify acquisitions most likely to harm competition. Further, understanding the effects of these acquisitions, in terms of price, quality, and other outcomes of concern, will be crucial to understanding the effects of these acquisitions on the provision and quality of health care in the United States.

Supplementary material

Supplementary material is available at Health Affairs Scholar online.

Funding

None.