Is Bigger Better? A Closer Look at Small Health Systems in the United States

Abstract

Background:

Research on U.S. health systems has focused on large systems with at least fifty physicians. Little is known about small systems.

Objectives:

Compare the characteristics, quality and costs of care between small and large health systems.

Research Design:

Retrospective, repeated cross-sectional analysis.

Subjects:

Between 468–479 large health systems, and between 608–641 small systems serving fee-for-service (FFS) Medicare beneficiaries, yearly between 2013–2017.

Measures:

We compared organizational, provider and beneficiary characteristics of large and small systems, and their geographic distribution, using multiple Medicare and Internal Revenue Service administrative data sources. We used mixed-effects regression models to estimate differences between small and large systems in claims-based HEDIS quality measures and HealthPartners’ Total Cost of Care measure using a 100% sample of Medicare FFS claims. We fit linear spline models to examine the relationship between the number of a system’s affiliated physicians and its quality and costs.

Results:

The number of both small and large systems increased from 2013–2017. Small systems had a larger share of practice sites (43.1% versus 11.7% for large systems in 2017) and beneficiaries (51.4% versus 15.5% for large systems in 2017) in rural areas or small towns. Quality performance was lower among small systems than large systems (−0.52 standard deviations of a composite quality measure) and increased with system size up to approximately 75 physicians. There was no difference in total costs of care.

Conclusions:

Small systems are a growing source of care for rural Medicare populations, but their quality performance lags behind large systems. Future studies should examine the mechanisms responsible for quality differences.

Background:

The organization of health care markets in the U.S. has shifted considerably in recent decades, with one key feature being the formation and growth of health systems. A health system is commonly defined as a group of affiliated health care delivery organizations (e.g., hospitals, physician practices) connected through common ownership or contractual relationships.^{1, 2} The number of physicians and practices that participate in health systems has risen markedly: between 2012 and 2018, the number of hospital-owned practices more than doubled; by the end of 2018, 31 percent of physician practices were owned by hospitals and 44 percent of physicians were employed by hospitals.^{3, 4} These figures likely underestimate the share of practices and physicians affiliated with systems, as they are based on ownership relationships – there are other forms of system affiliation, including clinically integrated networks and joint ventures.^{5, 6} Estimates that include ownership or joint management suggest that as of 2018, 51 percent of physicians were affiliated with health systems.⁷

The literature describing U.S. health systems and their cost and quality performance has focused primarily on larger systems, typically with 50 or more physicians.^7–12 Yet there are many smaller health systems about which little is known.⁸ Research describing individual physician practices finds that small practices differ from large practices in the types of providers and specialties they include, the patient populations they serve, and their structural capabilities (e.g., use of health information technology [HIT]).^{13, 14} Whether these differences persist when physician practices affiliate to form small or large health systems is unknown.

System size might also influence clinical quality and cost performance. Large systems may have greater access to capital to support investments in quality improvement infrastructure and processes that may translate into higher performance.^{15, 16} Also, if large systems offer a more comprehensive range of services that allow patients to receive the majority of their care within the system, this may create opportunities for more effective case management and care coordination, potentially leading to higher quality and reduced utilization of high-cost care, such as hospitalizations.^17–19 Conversely, large systems could face challenges standardizing care processes and building a unified culture across multiple practice sites and providers, or inertia in adopting and scaling-up innovations to improve quality and lower costs.^{20, 21} Finally, differences in patient populations, the types of providers (e.g., specialty mix) or payers between small and large systems could further influence quality, utilization and costs.

Understanding the characteristics and performance of small health systems is important to more fully characterize the national health care delivery landscape and the diversity of health systems. This study addresses this gap in the knowledge base by describing the features of small U.S. health systems (i.e., with fewer than 50 affiliated physicians) that served fee-for-service (FFS) Medicare beneficiaries between 2013 and 2017 and comparing them to large systems. We also compare the performance and costs of care for small and large systems, using measures of ambulatory, emergency department and inpatient clinical quality, and total cost of care. To our knowledge, this is the first study examine the characteristics of small systems.

Methods

Data:

To identify U.S. health systems serving Medicare FFS beneficiaries between 2013 and 2017, we used data from the Medicare Provider Enrollment, Chain, and Ownership System (PECOS) and the Internal Revenue Service (IRS) Form 990 database for the same years. PECOS contains information about ownership relationships between physician organizations (POs) (e.g., medical groups, faculty practices, independent practice associations) and health systems that serve Medicare FFS beneficiaries. The IRS Form 990 database contains tax documents filed by not-for-profit organizations, including health systems, hospitals, and POs. For a given organization, the Form 990 Schedule R lists its affiliated organizations, which we used to supplement PECOS where necessary, to determine whether a PO was affiliated with a health system (and what system). We used 2013 to 2017 Medicare Data on Provider Practice and Specialty (MD-PPAS) to identify physicians affiliated with specific POs, and we used PECOS and Physician Compare data to determine the location of each PO’s practice sites. Finally, to attribute Medicare FFS beneficiaries to POs, describe beneficiary characteristics, and measure quality and costs of care we used a 100 percent sample of Medicare FFS enrollment and claims data for beneficiaries age 65 and older covering the same time period.

Sample:

We used MD-PPAS data to identify POs based on their unique taxpayer identification numbers (TINs), and identified their affiliated physicians using the National Provider Identifiers (NPIs) associated with each TIN. We combined some TINs to form single POs if they were part of the same academic group, or if there was a large degree of overlap in their affiliated physicians. We excluded POs that did not include at least one physician from a specialty that routinely provides direct patient care (e.g., single-specialty groups consisting only of radiologists or pathologists). We then used PECOS and IRS Form 990 data to identify which POs were affiliated with health systems and group POs and hospitals with common ownership or management relationships into systems. We defined health systems as entities that included at a minimum one short-term general acute care hospital and at least one PO. We classified as “large” systems those with 50 or more physicians in total across all their affiliated POs, of whom at least 10 were PCPs (i.e., general internists or family physicians), and as “small” systems those with between 5 and 49 physicians in total, of whom at least 1 was a PCP. Our size criteria for large systems closely match the criteria used by other research teams to define health systems.^{7, 12} The derivation of our size criteria for small systems is described in the Supplement. We also conducted sensitivity analyses using a minimum size of 10 physicians of whom at least 1 is a PCP (see Supplement). Our resulting sample consisted of repeated cross-sections of large and small health systems, for each year from 2013 to 2017.

We attributed Medicare FFS beneficiaries to the health systems where they received the plurality of their ambulatory visits to PCPs in that year. Beneficiaries who did not visit a PCP during the year were attributed to the system where they received the plurality of their visits to internal medicine subspecialists (i.e., cardiology, endocrinology, gastroenterology, hematology/oncology, infectious disease, nephrology, pulmonology/critical care, rheumatology/immunology, and physical medicine/rehabilitation). Of all attributed beneficiaries, 93 percent were attributed based on visits to PCPs, and only 7 percent based on subspecialty visits.

Outcome Measures:

Quality was measured using a composite of Healthcare Effectiveness Data and Information Set (HEDIS) measures addressing ambulatory care quality (i.e., chronic disease management, preventive care, avoidance of overuse), readmissions and select inpatient and emergency room quality measures (see Supplement for measures). Individual HEDIS measures were constructed using Medicare FFS claims data. For each year from 2013 to 2017, we derived system-level scores for each individual HEDIS measure, where the value of each score was the percent of eligible beneficiaries in the system who satisfied the corresponding measure. These system-level scores were case-mix adjusted for within-provider differences in beneficiary age, sex, race, disabled status, dual Medicare-Medicaid eligibility, and rurality (as measured by category of Rural-Urban Commuting Area [RUCA] code).²² This approach is recommended by the National Quality Forum (NQF) to adjust performance measures for within-system differences in risk between groups of beneficiaries.^{23, 24} Some beneficiaries, as a group, tend to have lower quality scores, even when receiving care from the same providers as other groups. Systems that serve more beneficiaries who, as a group, tend to have lower quality scores therefore receive a positive adjustment to their performance measures. Of note, this approach preserves between-system differences in quality, such that systems that achieve lower performance across all groups of beneficiaries receive a lower score. We then used a Bayesian Rasch model to aggregate the individual system-level scores into a composite quality measure.²⁵ This approach has been shown to produce valid and reliable rankings of health system quality that are stable over time.²² We conducted sensitivity analyses using quality composites that were not adjusted for case-mix (see Supplement).

Total costs of care were calculated annually for each beneficiary using HealthPartners’ NQF-Endorsed Total Cost of Care measure (see Supplement).²⁶ Importantly, this measure purges price effects, such that estimated differences in the total cost of care reflect differences in resource use alone. We log-transformed the total cost of care measure because the distribution was heavily right-skewed.

We also examined the following descriptive characteristics: physician specialties, derived from MD-PPAS; beneficiary disabled status and dual eligibility for Medicare and Medicaid, derived from Medicare enrollment data; beneficiary race, imputed using the Medicare Bayesian Improved Surname Geocoding 2.0 methodology, which combines Medicare administrative data with U.S. census data to calculate probabilities of membership in each of six racial/ethnic groups (White, Black, Hispanic, Asian/Pacific Islander, American Indian/Alaska Native, and multiracial);²⁷ beneficiary Hierarchical Condition Category (HCC) score; and both beneficiary and practice site rurality (i.e., the percent of a system’s beneficiaries/practice sites residing in rural areas or small towns), using 2010 RUCA codes to identify beneficiaries.

Statistical Analysis:

We described each system’s organizational characteristics (e.g., number of practice sites), provider characteristics (e.g., number of physicians, specialist mix) and attributed Medicare FFS beneficiary characteristics (e.g., race/ethnicity, disabled status, dual status, HCC score). Within each of the two system size categories (i.e., small versus large), we then calculated the mean value of each characteristic across systems within a given year, or in the case of characteristics with highly skewed distributions, the median, 25^th and 75^th percentiles.

We examined differences in the number and type of subspecialty providers by computing, for each specialty, the percent of small and large systems with at least one physician in that specialty. To compare the geographic distribution of small and large systems, we mapped systems after assigning each one a location corresponding to the hospital referral region (HRR) containing the plurality of its practice sites.

We estimated differences in quality of care and total costs of care between small and large systems using the following multivariable regression model:

$$Y_{it}={\beta }_0+{\beta }_1{Small}_{it}+{\beta }_2{Rural}_{it}+{Year}_t+{HRR}_{it}+{System}_i+{\epsilon }_{it}$$

Y_it is either the case-mix adjusted quality composite score, or the log-transformed total cost of care for system i in year t; Small_it is a binary indicator that has a value of 1 if system i is a small system in year t and 0 otherwise; Rural_it is the percent of system i’s practices located in rural areas or small towns; and ε_it is an error term. The models include year fixed effects (Year_t) to control for time trends, HRR fixed effects (HRR_it) to control for differences in outcomes Y_it by HRR, and system random effects (System_i) to account for repeated measures on the same system across time. We examined the associations between system size and each of the 20 individual HEDIS measures that compose the quality composite with logistic mixed models and created a forest plot of these individual associations (see Supplement, Figure S1). We estimated alternative specifications omitting Rural_it and/or HRR_it, and using the untransformed total costs of care (see Supplement).

In addition to this model, we applied a piece-wise linear function to describe the relationship between the number of physicians in a system and that system’s composite quality rating, because we hypothesized the effect of increasing size on a system’s quality would differ depending on system size. We fit a series of linear models that regressed the quality composite measure on a linear spline representation of the total number of physicians in a system, with a single knot (change-point) (see Supplement for details). All models were estimated on the pooled data for all years 2013 through 2017, including only systems for which the reliability of the quality composite measure was at least 0.7.

Results:

Between 2013 and 2017, the number of both small and large systems increased, reaching 623 small systems and 478 large systems in 2017 (Table 1). As expected, small systems had far fewer practice sites than large systems. A striking difference between small and large systems is in the percent of practice sites in rural areas and small towns: in 2017, on average 43.1 percent of small systems’ practice sites were located in rural areas or small towns, compared to only 11.7 percent of large systems’ practice sites. Figure 1, which maps the locations of small and large systems in 2017 to HRRs, shows a higher number of large systems close to major population centers, particularly in the Northeast, mid-Atlantic and upper Midwest, as well as urban centers in California. Small systems are more common in less densely-populated areas of the Midwest and more rural Western states.

Table 1:

Number and Characteristics of Large and Small Systems

System Numbers and Characteristics		2013	2014	2015	2016	2017	Change from 2013–2017
Number of Large Systems		468	475	479	479	478	+2.1%
Number of Small Systems		608	615	618	641	623	+2.5%
Organizational Characteristics
Number of Practice Sites, median [25th, 75th %’ile]	Large	52 [26, 108]	57 [27, 118]	57 [27, 119]	56 [28, 122]	55 [27, 126]	+5.8%
	Small	6.5 [3, 12]	7 [3, 12]	7 [3, 12]	6 [4, 12]	6 [4, 12]	−7.7%
Percent of System’s Practice Sites in Rural Areas/Small Towns, mean, (sd)	Large	10.5 (14.5)	10.2 (14.5)	11.1 (15.3)	11.2 (15.2)	11.7 (15.8)	+1.3 pp
	Small	41.5 (40.6)	42.7 (40.7)	42.1 (40.4)	42.5 (40.7)	43.1 (40.7)	+1.6 pp
Provider Characteristics
Number of Physicians median [25th, 75th %’ile]	Large	210.8 [94.5, 557.5]	219 [99, 593]	229 [99, 616]	242 [93, 637]	233.5 [97, 667]	+10.8%
	Small	12 [7, 23]	13 [8, 23]	12 [7, 22]	12 [7, 21]	12 [7, 22]	0%
Number of Non-Physician Providers median [25th, 75th%’ile]	Large	54 [26, 114]	60 [29, 133]	71 [34, 156]	80 [35, 195]	90 [39, 221]	+66.7%
	Small	3 [1, 7]	4 [1, 8]	4 [1, 9]	4 [2, 9]	4 [2, 10]	+33%
Number of Specialists median [25th, 75th %’ile]	Large	121 [50.5, 348]	126 [53, 371]	130 [55, 374]	158 [61, 443]	150.5 [63, 490]	+24.4%
	Small	6 [3, 12]	6 [3, 12]	6 [3, 12]	6 [3, 13]	7 [3, 13]	+16.7%
Number of Systems with No Specialists	Large	0	0	0	0	0	0%
	Small	26	36	34	28	31	+19.2%
Beneficiary Characteristics
Number of Beneficiaries in Systems, median [25th, 75th %’ile]	Large	7580.5 [3930, 15801.5]	7825 [3950, 17862]	8038 [4361, 19623]	8567 [4295, 20358]	8685.5 [4316, 20151]	+14.6%
	Small	466.5 [126.5, 1053.5]	511 [180, 1162]	513 [160, 1147]	739 [285, 1311]	729 [316, 1306]	+56.3%
Percent of Total FFS Beneficiaries Served, mean (sd)	Large	31.5	33.9	35.2	37	38.3	+6.8 pp
	Small	2	2.2	2.2	2.6	2.4	+0.4 pp
%Living in Rural Areas/Small Towns, mean (sd)	Large	13.4 (15.6)	13.6 (16.1)	14.9 (17)	14.8 (17.1)	15.5 (17.6)	+2.1 pp
	Small	45.5 (35.5)	47.1 (35.9)	46.8 (36.1)	49.5 (36.5)	51.4 (36.5)	+5.9 pp

Table presents characteristics of large and small systems each year from 2013–2017. Where median values are presented, the 25^th and 75^th percentile values are shown in square brackets in the row below. Where mean values are presented, the standard deviation is shown in round brackets in the row below. For a comparison of an expanded set of beneficiary characteristics, see Table S3 in the Supplement. Abbreviations: %’ile – percentile; sd – standard deviation; pp – percentage point.

Figure 1: Geographic Distribution of Large and Small Systems, 2017.

For each hospital referral region (HRR), Figure shows the number of large (left side of figure) and small (right side of figure) health systems located in that HRR in 2017. Each health system’s assigned location was the HRR containing the plurality of its practices; practice location was determined using PECOS and Physician Compare data.

The median number of Medicare FFS beneficiaries attributed to small systems and large systems has grown considerably over time (Table 1). The percentage of all Medicare FFS beneficiaries attributed to small systems increased slightly during the study period, from 2.0 percent to 2.4 percent; in contrast, the percent of beneficiaries attributed to large systems grew more substantially, from 31.5 percent in 2013 to 38.3 percent in 2017. The starkest differences in beneficiary characteristics between small and large systems were in rural versus urban residence: in 2017, on average 51.4 percent of beneficiaries attributed to small systems lived in rural areas or small towns, compared to only 15.5 percent of beneficiaries attributed to large systems. This finding mirrors the differences in system and practice locations reported in Figure 1 and Table 1. For small systems, the percent of beneficiaries living in rural areas or small towns increased from 45.5 percent in 2013 to 51.4 percent in 2017. Large systems saw a smaller increase.

The median number of physicians, non-physician providers and specialists in small systems was stable from 2013 to 2017 (Table 1). In large systems, however, the median number of providers in each category increased substantially over time. All large systems had at least one affiliated specialist in all years of our study, as did a large majority of small systems (95 percent in 2017). The number of providers from specific specialties, however, differed considerably, and all specialties were more commonly found in large systems than small systems (Table 2). Interestingly, the prevalence of medical subspecialists in small systems was lower than for other types of specialists (e.g., surgical, radiology). Findings for the number and types of specialties in health systems were similar for the other years. We note that the absence of a particular specialist within a health system does not necessarily imply that beneficiaries were unable to access that type of specialty care elsewhere.

Table 2:

Specialty Mix in Large and Small Systems, 2017

Specialty	% Large Systems with Specialist	% Small Systems with Specialist
Medicine Subspecialties
Cardiology	89.7	16.2
Emergency Medicine	85.1	48.3
Gastroenterology	76.2	8.7
Geriatric Medicine	45.8	2.1
Hematology/Oncology	29.1	0.5
Neurology	86.4	14
Pulmonology	83.9	11.6
Surgical Specialties
Cardiac Surgery	51.9	1.6
General Surgery	96.4	57
Neurosurgery	64.4	3.9
Obstetrics/Gynecology	91.2	40
Orthopedic Surgery	81.8	38.7
Other Specialties
Anesthesiology	62.1	15.9
Diagnostic Radiology	51.5	10.8
Psychiatry	84.1	20.5

For select specialties, Table compares the percent of large and small systems with at least one physician from that specialty category.

Estimated differences in the quality and total costs of care between small and large systems are reported in Table 3. We estimate a negative, statistically significant association between classification as a small system and the composite quality measure across all model specifications (columns 1–3). The magnitude of this effect (equivalent to 0.52 standard deviations [SD] of the composite quality measure, column 1) decreases only slightly when adjusting for the percent of a system’s practices located in rural areas or small towns (column 2) and for HRR fixed effects (column 3). We find no statistically significant relationship between classification as a small system and total costs of care across all model specifications (columns 1–3). We obtained similar findings in our sensitivity analyses modeling the relationship between untransformed total costs of care and classification as a small system; using the unadjusted composite quality measure; and using an alternative definition of small systems as requiring at least 10 physicians in total of whom one is a PCP (see Supplement).

Table 3:

Differences in the Quality of Care and Total Cost of Care by System Size

	Model 1	Model 2	Model 3
Quality Composite a
Small System	−0.094 (−0.106, −0.082)***	−0.083 (−0.096, −0.07)***	−0.077 (−0.091, −0.064)***
Percent of Practices in rural areas or small towns		−0.062 (−0.084, −0.039)***	−0.061 (−0.087, −0.036)***
Number of systems in analysis	1230	1228	1207
Number of records in analysis	4922	4913	4808
Log-transformed Total Cost of Care b
Small System	−0.005 (−0.014, 0.005)	−0.007 (−0.016, 0.003)	−0.009 (−0.019, 0.001)
Percent of Practices in rural areas or small towns		0.01 (−0.005, 0.026)	−0.004 (−0.022, 0.015)
Number of systems in analysis	1312	1309	1288
Number of records in analysis	5390	5378	5249

Table presents model coefficient estimates and their 95% confidence intervals from linear mixed effects regression models of the quality composite and of the log-transformed total cost of care. All models included fixed-effect adjustment for year and system random effects to account for correlation within system. Model 3 additionally included fixed-effect adjustment for Hospital Referral Region (HRR).

^aRange = [−1.29, 0.65], Mean= 0.036, Standard Deviation= 0.18

^bRange = [7.94, 9.45], Mean= 8.57, Standard Deviation= 0.13

^***p-value < 0.001

^**p-value <0.01

^*p-value<0.05

Figure 2 displays the relationship between system size and quality across the range of 5 to 500 total physicians. Blue shaded data points are contributed by small systems, while red shaded data points are contributed by large systems. The black curve is a locally weighted smoothing function²⁸ that flexibly summarizes the relationship between system size and quality. The plot suggests that on average, quality is increasing with system size up to an inflection point between 50 and 100 total physicians, consistent with our linear spline modeling result (see Supplement). Thereafter, quality improvements are much smaller on average as system sizes increase further.

Figure 2: Relationship Between Total Number of System Physicians and Performance.

Figure plots the adjusted quality composite score against the total number of MDs for each system-year observation from 2013 to 2017. Small systems are represented by blue circles, and large systems by red circles. System-year observations were included only if the reliability of the quality composite measure was at least 0.7. For exposition, plot shows systems in the range of 5 to 500 total physicians only (the maximum number of total physicians in a system was over 6,000 but greater than 75 percent of systems had fewer than 1000 physicians). Line represents smoothed lowess curve describing the relationship between the adjusted quality composite score and physician count.

Discussion:

Although existing research on U.S. health systems has focused on large systems with at least 50 affiliated physicians, this study shows that there are also a substantial number of small systems. Our study provides one of the first examinations of small systems, describing their features and comparing quality and costs of care between small and large health systems.

Our analysis identified several distinctive features of small health systems. First, while only 2.4 percent of Medicare FFS beneficiaries overall receive the plurality of their medical care in small systems, these small systems are nonetheless an important – and growing – source of care in rural communities, particularly in the Midwest and in Western states. Challenges in ensuring access to health care in rural areas are well documented, and they include health care provider shortages and hospital and nursing home closures.^29–33 It is estimated that nearly half of rural hospitals are operating with negative margins and are therefore at risk of closure.³⁴ These concerning trends have focused policy attention on ensuring access to health care in rural areas and our findings suggest that small health systems are a growing source of care for rural populations.

Second, while the median number of attributed Medicare FFS beneficiaries has increased for both small and large systems, the pace of growth in large systems is far greater. These disparate patterns could reflect geographic variations in health care consolidation.¹⁰ Shortages in health care professionals in the more rural settings where small systems operate could also constrain their ability to attract increasing numbers of patients.^29–31

Third, while we find no difference in the costs of care between small and large health systems, we find lower performance by small systems on a composite measure of clinical quality derived from individual measures across a diversity of settings – ambulatory, emergency department, and inpatient. This performance gap of moderate magnitude persists even after adjusting for within-system differences in quality by beneficiary case-mix, the share of a system’s practices in rural or small-town settings, and unobserved features of the HRR that are constant during our study period. On the other hand, we also find that bigger is not always better: health care quality increases as system size grows from 5 to 75 physicians, but then the relationship plateaus. The reasons for this pattern are uncertain. There may be unobserved differences in the skill or quality of providers who practice in small versus large systems. Another possibility is that as systems increase in size, their larger revenue base facilitates investments in quality improvement, HIT or provider support, and their larger size facilitates peer learning. Alternatively, if large systems offer a broader scope of services, this may enable them to provide a larger share of their patients’ care and facilitate care coordination.³⁵ Yet our findings suggest diminishing returns to quality once systems reach a size threshold of approximately 75 physicians, raising the question of whether greater size creates countervailing challenges to both quality improvement and coordination of care. For example, in systems that have grown through mergers, administrators report that differences in culture can create barriers to standardizing care processes.^{21, 36} The complex relationship between size and quality is an important area for future research.

Our study has limitations. First, though PECOS is considered to be the best available source of information about ownership relationships between health care delivery organizations serving Medicare beneficiaries, it does not capture organizational affiliations that are not based on ownership (e.g., clinically integrated networks).³⁷ Further, it distinguishes health care organizations based on TINs, which do not always represent distinct entities.³⁸ We supplemented PECOS with IRS Form 990 data to identify additional relationships between not-for-profit health care organizations and systems. Nonetheless, gaps may persist in our ability to identify relationships between for-profit entities using PECOS alone.

Second, in our analyses we assigned each system to a single HRR based on where the largest number of its affiliated practices were located. Large systems, however, may have broad geographic reach across multiple HRRs. While our approach identifies areas with the highest concentration of a system’s practices, it does not fully characterize each system’s geographic scope. Third, we identify differences in quality performance by system size, but we are unable to determine the mechanisms that are responsible. Fourth, although we adjusted for a number of differences between the local environments of small and large systems – e.g., rurality and time-invariant characteristics of the HRR – we were unable to adjust for systematic, time-varying differences between the localities where small and large systems operate, such as policy changes. Fifth, while our clinical quality composite measure includes a broad set of nationally-endorsed, commonly-used quality measures, it is only a partial accounting of all patient care. Finally, we examine Medicare FFS beneficiaries only, which may limit the generalizability of our findings to other patient populations.

Notwithstanding these limitations, to our knowledge our study is the first to characterize small systems in detail. Our findings have implications for rural health policy in particular, given that small systems are a growing source of care for rural Medicare populations. We identify opportunities for performance improvement among small systems, and evidence of a positive association between quality and system size up to a limit. Small systems operating in rural areas, however, may be constrained in their growth due to limitations in the availability of health care professionals, or insufficient catchment population to support a larger enterprise. Identifying alternative approaches to improving quality in these settings is therefore important. Large systems, on the other hand, have increased in size in recent years – even though we observe diminishing returns to quality improvement at sizes above 75 physicians in total. Greater than 75 percent of large systems in our data already exceed this size threshold. Policymakers might therefore consider weighing the trade-offs associated with health care system growth and consolidation differently for small versus large systems.