Health Equity and Hospital Markets

Abstract

Background:

As hospital markets become increasingly consolidated, whether regulators should account for consolidation’s impacts on health equity has become a key policy question. We assess the association of hospital market concentration with quality of care and examine differences by patient race/ethnicity and payer.

Methods:

We analyzed linked 2017 American Hospital Association Annual Survey and Healthcare Cost and Utilization Project State Inpatient Data from 14 US states. Market concentration was measured using the Herfindahl-Hirschman Index (HHI) at the county level, and quality was assessed using the Prevention Quality Indicators (PQI). We assessed the relationship of HHI, patient race/ethnicity, and payer with having any PQI admission, controlling for patient and hospital characteristics. We used interaction terms for race-HHI and payer-HHI to assess differential associations of concentration by race/ethnicity and payer using linear probability models.

Results:

In adjusted analyses, minoritized racial/ethnic group status and having a noncommercial primary payer were associated with a higher probability of having a PQI admission. Differences between Hispanic adults and White adults decreased in more competitive markets but increased for Asian/Pacific Islander adults versus White adults. Differences in the probability of a PQI admission between adults covered by Medicaid and self-pay/no-pay adults versus commercially insured adults increased, while differences for adults covered by Medicare decreased.

Conclusions:

Hospital market concentration may have heterogeneous effects on the quality of care by patient race/ethnicity and payer. Because market concentration may impact equity, regulators should consider accounting for health equity impacts in merger reviews.

BACKGROUND

Hospital market structure (ie, the number and types of hospitals in a given market) and competitive dynamics represent an important topic for health services research. Highly concentrated hospital markets have been previously associated with higher health care prices^1–4; however, it is less clear whether such consolidation is associated with better quality of care. Evidence suggests a complex relationship between consolidation and quality of care in the United States,^5–7 whereby increased market concentration may have positive, negative, or null effects on measures of quality of care depending on the quality measures being studied. In the past few years, consolidation of hospital markets has continued to increase,⁸ spurring interest in identifying policies needed to mitigate potential negative impacts.^1,9–11

Recent works have argued for the need for policy change to improve competition in health care markets.^12–14 In July 2023, the Federal Trade Commission (FTC) and the Department of Justice’s (DoJ) antitrust division released draft guideline updates.¹⁵ Updating the prior set of guidelines released in 2010, the proposed changes will likely increase the number of potential mergers that may be scrutinized; for example, the Herfindahl-Hirschman Index (HHI) required for a market to be considered highly concentrated is 1800, down from 2500 in the 2010 guidelines.¹⁶ Such thresholds reflect the DoJ and FTC’s experience and considerations in classifying markets when deciding whether antitrust enforcement action is needed.

While there are several measures of quality, the Agency for Health care Research and Quality (AHRQ) Prevention Quality Indicators (PQIs) have been central to performance improvement initiatives. PQIs measure potentially preventable hospitalizations and are a commonly accepted indicator of health care system quality and population health.¹⁷ Hospital consolidation can influence these outcomes through various mechanisms, including reducing incentives to improve quality, decreasing access with potential closure/consolidation of facilities, and deterring patients from seeking care due to price increases. For example, a prior study found that increasing hospital consolidation increased insurance coverage disparities for both members of minoritized racial/ethnic groups and low-income households.¹⁸ Additional work has suggested increasing market concentration leads to reduced access to nonprofit hospitals for Medicaid patients in New York.¹⁹ Because the concentration of hospital markets may influence access to care, assessing changes in quality of care using PQI-based measures can provide information on how consolidated hospital markets may have a broader impact beyond the use of a measure limited to inpatient care quality.

In recent years, there has been increasing interest in health equity across different levels of health, including in the provision and outcomes of health services. For example, racial and ethnic disparities in access and receipt of care,²⁰ including inpatient care,^21,22 continue to be a key area of focus in health services research. A similar focus has examined disparities faced by patients enrolled in public insurance programs (eg, Medicaid and Medicare) compared with patients with commercial insurance, as Medicaid and Medicare patients experience worse outcomes across a variety of measures.²³ Ultimately, little is known about the association between hospital market concentration and disparities in quality of care, including which populations receive worse quality of care in more concentrated hospital markets. Given this, we use data from fourteen states to answer three research questions:

1. What is the association between market concentration and preventable hospitalizations, controlling for patient and hospital-level covariates?

2. To what degree does race/ethnicity modify the association between market concentration and preventable hospitalizations, controlling for the same patient and hospital-level covariates?

3. To what extent does payer status modify the association between market concentration and preventable hospitalizations, controlling for the same patient and hospital-level covariates?

METHODS

Data and Study Sample

We analyzed data from the 2017 Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SID), all-payer hospital discharge data.²⁴ Fourteen states were included based on data availability and ability to link such data to the American Hospital Association’s (AHA) Annual Survey: Arkansas, Arizona, Colorado, Florida, New Jersey, New York, Oregon, Rhode Island, Vermont, Washington, Kentucky, Maryland, Nevada, and Wisconsin. The AHA Annual Survey is a census of all hospitals in the United States and has an ~85% response rate; survey responses are supplemented with information from the US Census Bureau, hospital accrediting bodies, and other organizations that provide comprehensive information for each hospital.²⁵ In this study, we focused on adult patients (eg, 18 y of age or older) because the PQI does not typically apply to pediatric populations.

Outcome Variable

We used each admission’s ICD-10 codes to identify 10 AHRQ-defined potentially preventable admissions (PQIs): admissions related to diabetes short-term complications, diabetes long-term complications, chronic obstructive pulmonary disease (COPD) or asthma in older adults, hypertension, heart failure, community-acquired pneumonia, urinary tract infection, uncontrolled diabetes, asthma in younger adults, and lower-extremity amputation among patients with diabetes.²⁶ We aggregated the ten individual PQIs into a single dichotomous measure, where 1 indicated a PQI admission, and 0 indicated a non-PQI admission. See Appendix 1, Supplemental Digital Content 1, http://links.lww.com/MLR/C956 for more details on the PQIs included in our composite measure.

Independent Variables

The main independent variables were market concentration, race/ethnicity, payer, and the interactions between: (1) race/ethnicity and concentration; and (2) payer and concentration. Hospital market concentration was measured by the Herfindahl-Hirschmann index, a widely used measure of hospital market consolidation.^8,27,28 In this study, HHI was calculated using inpatient services provided by hospitals at the county level using data from the 2017 AHA Annual Survey. A higher HHI indicates a market that is more concentrated. We defined a dichotomous variable for market concentration that defined a market as “highly concentrated” (or noncompetitive) if the market concentration was >2500, and “moderately concentrated/competitive” (reference) if the market concentration was 2500 or less based on the 2010 Horizontal Merger Guidelines. The Horizontal Merger Guidelines govern the processes by which federal authorities scrutinize and challenge potential mergers across different markets, including health care, and provide key HHI cutpoints that can be used to classify markets as competitive or not. We use the 2010 guidelines (defining a highly concentrated market as >2500) instead of the 2023 guidelines (defining a highly concentrated market as 1800) because the former were the guidelines in place at the time the admissions were made. 5.1% of admissions in our sample were in hospital markets with an HHI less than or equal to 1800 (377,372), compared with 18.4% with an HHI of ≤2500.

Patient race/ethnicity was documented by the hospital when providing care for patients. The default combined race/ethnicity categories in HCUP data included non-Hispanic White, Black, Hispanic, Asian/Pacific Islander (API), Native American, and Other. Adults identified as non-Hispanic White served as the reference group.

Primary payer included the categories Medicare, Medicaid, commercial insurance, self-pay/no pay, and other. Private insurance served as the reference group. Patients covered by both Medicare and Medicaid (“dual eligibles”) are coded as Medicare patients for the main analyses given secondary payers are only available for roughly ~60% of admissions in our sample. In states where the secondary payer is identified in the data, 8% of admissions were dual eligibles.

Other Covariates

Both admission-level and hospital-level covariates were determined following prior research on racial and ethnic health disparities and quality of care^17,23 and are included to control for confounding the relationship between race, competition, and potentially preventable inpatient admissions. We included sex, age, and zip income quartile (with the highest income quartile as referent) as admission-level covariates. Hospital covariates included hospital ownership (public, for-profit, not-for-profit, and government-owned [referent]), major teaching hospital status, joint commission accreditation, membership in a health system, accountable care organization status, bed size [up to 99 beds, 100–299 beds (referent), 300+ beds], and US Census region [West, Midwest, South, and East (referent)].

Statistical Analyses

Univariate and bivariate analyses were conducted to summarize the analytic sample characteristics. In our analyses, we used multilevel linear probability models to estimate the probability of having a potentially preventable inpatient admission. For all models, we included hospital-level random effects as part of the multilevel structure. We used linear models over logistic models given the challenges of interpreting interaction terms for logistic regression models.^29,30

First, we estimated a model without covariates (null model) to determine the portion of variance that was within and between individual hospitals. Second, we estimated the difference in potentially preventable inpatient admission based on hospital market concentration, controlling for patient and hospital covariates. We then extended the base model by adding interaction terms between market concentration and each racial/ethnic category when estimating the association of market concentration with a preventable admission, controlling for admission and hospital-level covariates. Similarly, we examined the interaction between market concentration and payer category when estimating the association of market concentration on preventable inpatient care use, controlling for admission, and hospital-level covariates. We show the results for the “other” category for both race/ethnicity and payer but avoid drawing inferences, given the expansiveness of such categories.

All analyses were conducted using Stata 17. We also transformed estimates from our model into predicted probabilities using Stata’s margins command, which aids in interpretation of complex regression results by holding covariates at specific values or means of values. To aid in understanding the difference between groups across the binary HHI variable, we calculated percentage point (pp) differences from the point estimates. Because all analyses used de-identified patient data, the research study was not considered human subjects research, and IRB approval was not required.

Sensitivity Analysis

First, we were interested in whether our HHI results were consistent when HHI is measured continuously rather than using a binary measure based on the 2500 HHI cutpoint. While the use of this cutpoint is used by both government agencies and researchers to study how market concentration is associated with outcomes, considering market concentration as a spectrum may yield different inferences. As a sensitivity analysis, we re-estimated models using a standardized and continuous HHI. We generated a standardized version of the HHI measure such that the interpretation of the coefficient from the regression model is the effect of a 0.1 SD change in the predictor on the outcome. We then constructed models using this measure similar to our analytical procedure using the binary measure (ie, with a base model with no interactions, a model introducing the race/ethnicity and standardized HHI interaction, and a model introducing the payer and standardized HHI interaction).

Next, we also analyzed results for our payer analyses when dual eligibles are split out (for the subset of states identifying secondary payers), repeating the analytical procedures listed above. We analyze results using both the binary exposure and the standardized, continuous HHI.

RESULTS

Descriptive Analyses

The analytic sample included complete patient and hospital information across all covariates for ~7.3M admissions across 1059 hospitals (average admissions: 6878; minimum: 14; maximum: 99,292). Table 1 displays our sample’s characteristics. 82% of our sample was in a highly concentrated/ noncompetitive market (HHI>= 2,500), in line with prior research showing a steadily rising trend of concentration across hospital markets in the United States.^8,31 Sixty-eight percent of our sample was non-Hispanic White, and Medicare was the primary payer for the majority of admissions (54.5%).

TABLE 1.

Visit and Hospital Characteristics, by Hospital Market Concentration

	Total sample (N=7,305,850) (%)	Herfindahl-Hirschman Index <2500 (N=1,348,518) (%)	Herfindahl-Hirschman Index ≥2500 (N=5,957,332) (%)
		18.4	81.5
Visit characteristics
Race
Non-Hispanic White	68.3	46.6	73.2
Black	14.3	22.3	12.5
Hispanic	10.5	19.1	8.6
Asian or Pacific Islander	2.0	3.1	1.8
Native American	0.4	0.2	0.5
Other	4.4	8.7	3.4
Female	51.3	50.0	51.6
Mean age (SD)	62.5 (18.4)	61.4 (19.0)	62.7 (18.2)
ZIP code income quartile
1st quartile	29.5	26.3	30.2
2nd quartile	25.7	18.6	27.3
3rd quartile	23.7	22.5	24.0
4th quartile	21.1	32.6	18.5
Primary payer
Medicare	54.5	49.7	55.6
Medicaid	16.3	20.0	15.5
Commercial	22.4	24.0	22.1
Self or no pay	4.1	4.8	4.0
Other	2.9	2.0	3.1
PQI admission	13.6	12.3	13.9
Hospital characteristics (by visit
Member of the larger health system	79.7	79.1	79.9
Ownership
Public (county, state, other gov’t)	10.5	15.8	9.3
Not-for-profit (non-gov’t)	73.8	69.1	74.8
For-profit	15.7	15.1	15.9
Bed size
<100	7.0	0.8	8.4
100–299	35.1	28.1	36.6
300+	57.9	71.1	54.9
Joint Commission Accreditation	87.2	97.2	85.0
Major teaching hospital status	21.4	40.7	17.1
US Census region
Northeast	32.1	51.5	27.8
Midwest	5.0	0	6.1
South	41.6	38.9	42.2
West	21.3	9.6	23.9

PQI indicates Prevention Quality Indicators.

The null model’s intraclass correlation of 0.07 means that among our unmeasured variance for PQI admissions, 7% can be attributed to the hospital level.

Adjusted Analyses

Table 2 shows results from the multilevel linear probability models with the binary competition measure. After controlling for patient-level and hospital-level covariates, increased market concentration was associated with an increased probability of having a PQI admission. After controlling for all covariates, adults identified as a member of a minoritized racial/ethnic group had an increased probability of having a PQI admission compared with non-Hispanic White adults. Similarly, adults covered by Medicare and Medicaid, as well as self/no-paying adults, had a higher probability of a PQI admission compared with commercially insured patients.

TABLE 2.

Multilevel Regression Results for Probability of Having a PQI Admission Using HHI = 2500 Cutpoint

	Base model with no interactions	Model with race × competition interaction	Model with payer × competition interaction
Key exposures
Market concentration
Highly concentrated/ noncompetitive hospital market	0.0169*	0.0180*	0.0159*
Moderately concentrated or competitive	REF	—	—
Race/ethnicity
Non-Hispanic White	REF	—	—
Black	0.0484***	0.0484***	0.0484***
Hispanic	0.0187***	0.0221***	0.0187***
Asian or Pacific Islander	0.00341***	−0.00208	0.00346***
Native American	0.0163***	0.0178**	0.0160***
Other	−0.0000121	0.00674***	0.000117
Payer status
Commercial	REF	—	—
Medicare	0.0510***	0.0510***	0.0533***
Medicaid	0.0386***	0.0385***	0.0331***
Self-pay/no pay	0.0280***	0.0280***	0.0160***
Other	0.00572***	0.00572***	−0.00994***
Concentration × race/ethnicity
Non-Hispanic White	—	REF	—
Black	—	0.000271	—
Hispanic	—	−0.00444***	—
Asian or Pacific Islander	—	0.00789***	—
Native American	—	−0.00169	—
Other	—	−0.0100***	—
Concentration × payer
Commercial	—	—	REF
Medicare	—	—	−0.00277***
Medicaid	—	—	0.00699***
Self-pay/no pay	—	—	0.0150***
Other	—	—	0.0179***
Random intercept	0.00433***

Models control for patient-level covariates (sex, age, primary payer, zip code income quartile) and hospital-level covariates (ownership, major teaching hospital status, bed size, joint commission accreditation, health system membership, and region) and use hospital random effects.

^* P<0.05.

^** P < 0.01.

^*** P < 0.001.

HHI indicates Herfindahl-Hirschman Index; PQI, Prevention Quality Indicators.

Table 3 shows the predicted probabilities from the models with the interaction terms for both race/ethnicity and payer. In the race/ethnicity analyses, all groups had a higher predicted probability of a PQI admission in highly concentrated markets compared with moderately concentrated markets. Adults identified as Asian or Pacific Islander had a slightly greater difference in predicted probabilities compared with adults identified as non-Hispanic White (2.2 percentage points vs. 1.9 percentage points). Adults identified as Hispanic (1.3 percentage points), Native American (1.7 percentage points), and Black (1.8 percentage points) had slightly lower differences in predicted probabilities compared with adults identified as non-Hispanic White (1.9 percentage points).

TABLE 3.

Predicted Probability of Having a PQI Admission Using HHI = 2500 Cutpoint for the Interaction Models

	Model with race × competition interaction [95% CI]			Model with payer × competition interaction [95% CI]
	<2500 HHI × race/ethnicity	≥2500 HHI × race/ethnicity	Difference in point estimates	<2500 HHI × payer	>=2500 HHI × payer	Difference in point estimates
Non-Hispanic White	0.118 [0.105, 0.132]	0.137 [0.131, 0.143]	0.019	—	—	—
Black	0.167 [0.154, 0.180]	0.185 [0.179, 0.191]	0.018	—	—	—
Hispanic	0.141 [0.127, 0.153]	0.154 [0.148, 0.160]	0.013	—	—	—
Asian or Pacific Islander	0.117 [0.103, 0.130]	0.139 [0.133, 0.145]	0.022	—	—	—
Native American	0.136 [0.118, 0.154]	0.153 [0.146, 0.160]	0.017	—	—	—
Other	0.125 [0.112, 0.140]	0.133 [0.127, 0.140]	0.008	—	—	—
Commercial	—	—	—	0.0941 [0.0810, 0.107]	0.110 [0.104, 0.116]	0.0159
Medicare	—	—	—	0.147 [0.134, 0.160]	0.161 [0.154, 0.167]	0.014
Medicaid	—	—	—	0.127 [0.114, 0.140]	0.150 [0.144, 0.156]	0.023
Self-pay/no pay	—	—	—	0.110 [0.0968, 0.123]	0.141 [0.135, 0.147]	0.031
Other	—	—	—	0.0842 [0.0704, 0.0978]	0.118 [0.111, 0.124]	0.0338

Differences in point estimates are used to illustrate differences across the HHI variable for each racial/ethnic and payer group but do not represent the products of significance tests across the categories.

CI indicates confidence level; HHI, Herfindahl-Hirschman Index; PQI, Prevention Quality Indicators.

In the payer analyses, all groups had a higher predicted probability of a PQI admission in highly concentrated markets compared with moderately concentrated markets. Adults identified as self-pay or no pay had a greater difference in predicted probabilities when compared with commercially covered adults (3.1 percentage points vs. 1.59 percentage points). Similar dynamics existed for Medicaid-covered adults (2.3 percentage points vs. 1.59 percentage points). Adults covered by Medicare had lower differences in predicted probabilities compared with commercially covered adults (1.4 percentage points vs. 1.59 percentage points). Plotted predicted probabilities by racial and ethnic group are available in Appendix 2, Supplemental Digital Content 1, http://links.lww.com/MLR/C956 and similar content is available by payer in Appendix 3, Supplemental Digital Content 1, http://links.lww.com/MLR/C956.

Sensitivity Analysis

Table 4 shows results from the multilevel regression models with the continuous HHI measure. In the base model, the standardized, continuous measure of market concentration was associated with a higher probability of having a PQI admission. Similar to results from the binary measure, after controlling for all covariates, each group of adults identified as a member of a minoritized racial/ethnic group had a higher probability of having a PQI admission compared with non-Hispanic White adults. In addition, adults covered by both Medicare and Medicaid, as well as self/no-paying adults, had a higher probability of a PQI admission compared with commercial patients. In the model with the interaction between race/ethnicity and market concentration, the interaction term’s coefficient was negative for adults identified as Black, Hispanic, and Native American, indicating narrowing differences in the probability of a PQI admission compared with non-Hispanic White adults as the standardized HHI measure increased. The interaction term’s coefficient for adults identified as Asian or Pacific Islander was not significant. In the model with the interaction between primary payer and market concentration, the interaction term’s coefficient was positive for adults covered by Medicare and self/no-paying adults. The interaction term’s coefficient for adults covered by Medicaid was not significant.

TABLE 4.

Multilevel Regression Results for Probability of Having a PQI Admission Using Standardized, Continuous HHI

	Base model	Model 5 with race × competition interaction	Model 6 with payer × competition interaction
Key exposures
Market concentration (standardized, continuous)	0.0112***	0.0120**	0.00891***
Race/ethnicity
Non-Hispanic White	REF	—	—
Black	0.0484***	0.0481***	0.0484***
Hispanic	0.0187***	0.0181***	0.0187***
Asian or Pacific Islander	0.00343***	0.00454***	0.00336***
Native American	0.0163***	0.0168***	0.0163***
Other	0.0000776	−0.00134***	−0.00000628
Payer status (vs. commercial)
Commercial	REF	—	—
Medicare	0.0510***	0.0514***	0.0511***
Medicaid	0.0386***	0.0384***	0.0386***
Self-pay/no pay	0.0279***	0.0280***	0.0281***
Other	0.00571***	0.0186***	0.00535***
Interaction between concentration × race/ethnicity
Non-Hispanic White	—	REF	—
Black	—	−0.00332***	—
Hispanic	—	−0.00387***	—
Asian or Pacific Islander	—	0.00139	—
Native American	—	−0.00445*	—
Other	—	−0.00722***	—
Interaction between concentration × payer
Commercial	—	—	REF
Medicare	—	—	0.00328***
Medicaid	—	—	0.000773
Self-pay/no pay	—	—	0.00272***
Other	—	—	0.00686***
Random intercept	0.00420 (0.00383, 0.00459)***

Models control for patient-level covariates (sex, age, primary payer, zip code income quartile) and hospital-level covariates (ownership, major teaching hospital status, bed size, joint commission accreditation, health system membership, and region) and use (hospital random effect).

^* P < 0.05.

^** P < 0.01.

^*** P < 0.001.

HHI indicates Herfindahl-Hirschman Index; PQI, Prevention Quality Indicators.

Table 5 shows results from the dual eligible subanalysis using both the binary and continuous HHI measures. The coefficient for the interaction term for dual eligibles was comparable to adults covered by Medicare using the binary competition exposure. However, the coefficient for the interaction term for dual eligibles was incrementally higher compared with adults covered by Medicare when using the standardized and continuous form of HHI. Plotted predicted probabilities with dual eligibles split out are available in Appendix 4, Supplemental Digital Content 1, http://links.lww.com/MLR/C956.

TABLE 5.

Multilevel Regression Results for Dual Eligible Sensitivity Analysis Using Both HHI Exposure Definitions

	Model with binary exposure	Model with standardized, continuous HHI
Key exposures
Market concentration (standardized, continuous)	0.0167	0.00731**
Payer status (vs. commercial)
Medicare	0.0406***	0.0439***
Dual eligible	0.0630***	0.0665***
Medicaid	0.0309***	0.0324***
Self-pay/no pay	0.0208***	0.0257***
Other	−0.0167***	−0.00109
Interaction between concentration × payer
Medicare	0.00407***	0.00399***
Dual eligible	0.00438**	0.00522***
Medicaid	0.00167	0.00182**
Self-pay/no pay	0.00639*	0.00041
Other	0.0197***	0.0101***
Random intercept	0.00470***	0.00460***

Models control for patient-level covariates (sex, age, primary payer, zip code income quartile) and hospital-level covariates (ownership, major teaching hospital status, bed size, joint commission accreditation, health system membership, and region) and use hospital random effects.

^* P<0.05.

^** P<0.01.

^*** P<0.001.

HHI indicates Herfindahl-Hirschman Index.

DISCUSSION

This study examined the association between hospital competition, race/ethnicity payer coverage, and individual interactions with the PQI in a multilevel model. Using a binary specification based on the Horizontal Merger Guidelines (Tables 2 and 3), we found that higher market concentration may be associated with a narrowing of disparities for some groups (adults identified as Hispanic compared with White, and adults covered by Medicare compared with commercial insurance) but a potential widening of disparities for others (adults identified as Asian or Pacific Islander versus White; adults covered by Medicaid and self/no-paying versus commercially covered adults) reflects the complex relationship between quality and market concentration documented in prior research. Adding to this nuanced relationship are the results of the models using the continuous formulation of market concentration (Table 4), which yielded a few different results from the binary specification. This may mean that the threshold for HHI from the 2010 Horizontal Merger Guidelines may have been more relevant for some groups and not for others.

Taken together, our main results and sensitivity analyses underscore the complexity of the relationship between competition and equity and how choices in measures may influence the identification of relationships. Part of the challenge for both researchers and regulators alike is driven by the limitations of the simplistic HHI measure, which ultimately precludes a holistic consideration of dynamics in a given market. In practice, some patients may benefit from the potential efficiencies in care that are an opportunity when markets are more concentrated, while others may experience worse care due to the lack of competitive incentives. Importantly, these dynamics may not be homogeneous for all groups within a given market; advantages for some and disadvantages for others can coexist within markets and have different implications for members of different racial/ethnic groups and with different insurance coverage. Ultimately, however, the coefficients for the interaction term were relatively small, suggesting that market concentration may only be associated with small differences (if any) in PQI admissions by race/ethnicity and payer status.

Under the current state of antitrust enforcement, how potential mergers may impact health equity is not a current consideration, as the FTC and courts have focused mostly on the impact of mergers on higher prices for hospital services for commercially insured patients.¹⁴ These findings suggest that, given regulators’ inability to consider such issues, mergers that produce disparate impacts on quality across different groups of patients may be allowed. While the FTC may face substantial difficulties in challenging mergers due to equity concerns,¹⁴ state agencies are also able to challenge mergers based on a broader range of authority given by state legislatures,^1,10,14 with the possibility of expansion of authority to explicitly allow consideration of equity. Beyond antitrust regulation, states may also have additional regulatory authority that can be extended to incorporate equity concerns. For example, effective June 2023, New York “requires a Health Equity Impact Assessment (HEIA) to be filed with a Certificate of Need (CON) application for the establishment, ownership, construction, renovation, and change in service of health care facilities across New York State, which will provide information on whether a proposed project impacts the delivery of or access to services for the service area, particularly medically underserved groups.”³² While these are example policies that may be impactful, more comprehensive evaluations of how equity in quality of care may vary by hospital market concentration are needed to inform policy action.

There are many areas for future research on this topic. While this study’s methodology used data across markets to examine disparities, it is very likely that specific hospital markets and market changes (ie, closure, mergers, and acquisitions) will have their own unique impacts on the quality of care based on existing hospital practices and the patient populations served. Further research is needed to parse out the market and/or merger conditions that yield such heterogeneity in these dynamics, including identified characteristics of markets where market concentration produces the largest differences (both positive and negative) in disparities between racial/ethnic groups and by payer coverage, as well as isolating specific comparisons that focus on hospitals in danger of closure and the impacts on the market when the hospital is acquired vs closes.³³ Other dynamics, like separations of previously integrated hospitals in a health system, also represent topics worth further inquiry.³⁴ The significance of the random intercept suggests that there are hospital-level drivers of the outcome that are not included in the fixed effects; identification of these characteristics in future work may be informative for policy development.

Limitations

First, all analyses are cross-sectional using a single year of data and should not be interpreted as causal. Investigating the interaction between market concentration and both race/ethnicity and payer status in a more causally aimed study would be worthwhile. Second, this study is not nationally representative and contains a select handful of states given data availability. Third, we examine only a single quality measure using a single way of defining HHI. Other quality measures, including AHRQ’s Patient Safety Indicator and Inpatient Quality Indicator,¹⁷ may potentially be more directly tied to hospital incentives/behaviors that may change based on market concentration and should be examined. Similarly, we were limited to a county-based HHI measure and did not leverage other market definitions (for example, hospital service area or hospital referral region⁶). Fourth, we consider hospital competition’s association with disparities in quality of care but do not consider how competition in other health care markets (ie, specialist/primary physician¹⁰ and insurance markets) may influence these dynamics.^35,36 Alongside hospital consolidation, these changes in competitive dynamics may interact in ways that can strengthen or weaken their individual impacts.

CONCLUSIONS

As hospital markets continue to consolidate, it will be imperative for states to consider how to leverage regulatory authority to ensure minimal impacts on equity of care. Further research would be well suited in longitudinally assessing how market concentration may impact equity in care, as well as distinguishing between the potential effects of mergers vs closures. While prior work has leveraged mergers as shocks in quasiexperimental studies focused on quality,⁷ added evidence on heterogenous impacts across different groups (ie, race/ethnicity, payer) is needed. Such work can help inform the type and extent of policy action to address the preservation of quality services for potentially vulnerable patient populations.