State-Funded Health Insurance Expansion for Noncitizen Immigrants and Hospital Uncompensated Care

Key Points

Question

What is the association between expansion of state-funded health insurance for undocumented and noncitizen adults (Health Benefits for Immigrant Adults [HBIA] and Health Benefits for Immigrant Seniors [HBIS]) in Illinois and hospitals’ uncompensated care costs?

Findings

In this study of 161 Illinois hospitals and 1902 control hospitals, HBIA and HBIS were associated with a significant reduction in unadjusted and adjusted hospital bad debt, ranging from 8% to 15%. Reductions were larger among hospitals in counties with higher proportions of immigrants.

Meaning

These findings suggest that the expansion of health insurance to undocumented and noncitizen immigrants can improve hospitals’ financial stability while remedying significant health care disparities.

This cohort study examines the association between implementation of the Health Benefits for Immigrant Adults and Health Benefits for Immigrant Seniors programs and changes in uncompensated care in Illinois hospitals.

Abstract

Importance

Undocumented and many noncitizen immigrants are largely excluded from federally funded health insurance programs, creating substantial gaps in health coverage and financial strain on safety-net hospitals that disproportionately serve these populations through uncompensated care. Several states have implemented state-funded health coverage expansions for immigrants, but the financial implications of these policies remain poorly understood.

Objective

To examine whether Health Benefits for Immigrant Seniors (HBIS) and Health Benefits for Immigrant Adults (HBIA) programs in Illinois were associated with changes in hospitals’ charity care and bad debt.

Design, Setting, and Participants

This cohort study of nonfederal general medical and surgical hospitals from January 1, 2017, to December 31, 2023, used Healthcare Cost Report Information System data. Difference-in-differences regression compared hospital financial outcomes in Illinois with hospitals in comparison states. Difference-in-differences-in-differences models examined variation by county share of immigrant populations. Analysis was performed from October 1, 2024, to June 12, 2026.

Exposure

Illinois HBIA and HBIS programs.

Main Outcomes and Measures

Hospital-reported bad debt and charity care.

Results

The study included 161 hospitals in Illinois and 1902 hospitals in comparison states. Illinois hospitals were located in counties with a higher share of Hispanic population (mean [SD], 11.4% [10.1%] vs 9.1% [10.6%]; standardized mean difference [SMD], 0.23) and higher median income ($58 121 [range, $34 982-$100 325] vs $56 102 [range, $23 968-$134 050]; SMD, 0.17), but a lower share of uninsured population (mean [SD], 7.8% [2.0%] vs 9.1% [3.5%]; SMD, −0.68). Illinois (treatment) and control hospitals were largely comparable on hospital-level attributes. In adjusted difference-in-differences models, the HBIA and HBIS programs were associated with declines in bad debt (−0.08 [95% CI, −0.11 to −0.04), adjusted bad debt (−0.13 [95% CI, −0.17 to −0.09), charity care (−0.19 [95% CI, −0.26 to −0.13), and adjusted charity care (−0.19 [95% CI, −0.27 to −0.12). Results were robust in entropy-balanced models, which showed reductions for adjusted bad debt (−0.14 [95% CI, −0.20 to −0.09]) and adjusted charity care (−0.15 [95% CI, −0.23 to −0.07). Estimated reductions in charity care were more sensitive to potential violations of the parallel trends assumption than reductions in bad debt.

Conclusions and Relevance

In this cohort study of hospitals in Illinois and comparison states, the HBIA and HBIS programs were associated with significant reductions in hospital bad debt and suggestive reductions in charity care. State-funded health insurance expansions for undocumented and noncitizen immigrants were associated with improved hospital financial stability and may help address persistent gaps in insurance coverage among immigrants.

Introduction

The Patient Protection and Affordable Care Act (ACA) substantially expanded health insurance coverage in the US and improved access to care.^1,2,3 However, undocumented immigrants and some noncitizen immigrants were explicitly excluded from ACA expansions.⁴ As of 2023, half of likely undocumented adults and 18% of lawfully present immigrant adults reported being uninsured compared with 6% of naturalized and 8% of US-born citizens.⁵ The exclusion of immigrants from coverage has resulted in more uncompensated care in safety-net systems serving this community.

Recently, several states have created state-funded insurance programs for immigrants who remain ineligible for federal programs.⁶ Illinois was among the first through the implementation of Health Benefits for Immigrant Seniors (HBIS)⁷ and Health Benefits for Immigrant Adults (HBIA) programs, in 2020 and 2022, respectively.⁸ HBIS covered low-income adults 65 years and older, while HBIA expanded eligibility to adults aged 42 to 64 years. These programs provided Medicaid look-alike coverage, including hospital, primary, behavioral health, dental, and pharmacy services. Enrollment grew to 70 000 participants by 2023 before the state paused enrollment in 2024, and sunset the HBIA program in 2025 due to budget constraints.

Despite their policy significance, there is limited empirical evidence on the benefits or tradeoffs of these state-funded immigrant coverage programs. Understanding their impact on hospitals’ uncompensated care is critical because these costs can strain health system finances and performance.^9,10 Evidence from previous ACA and Medicaid expansions found that increased insurance coverage reduced unpaid care,¹¹ freeing hospital resources and improving financial stability. Effects were particularly pronounced for rural and public hospitals that shared a disproportionate level of patients with Medicaid coverage. Insurance expansions may influence multiple components of uncompensated care through distinct mechanisms. By reducing the number of uninsured patients, expanded coverage may decrease hospital bad debt from unpaid bills, while also reducing charity care by lowering the need for hospital financial assistance among previously uninsured patients. Whether immigrant-inclusive coverage policies yield similar benefits remains unknown.

Hospitals’ financial stability has direct implications for patients, clinicians, and the communities hospitals serve.¹² Those facing sustained financial strain may reduce service lines, limit staffing, or delay investments, constraining access to care, particularly in safety-net communities and those with a high percentage of immigrants.^13,14 Financial strain, including uncompensated care, has also been linked to system performance, quality of care, and patient outcomes.^15,16,17 Understanding how expansion of insurance to immigrants affects uncompensated care is therefore not only a fiscal question, but one tied to broader system quality and outcomes.

This study addresses this gap by examining changes in hospitals’ uncompensated care following the implementation of HBIS and HBIA. Our study provides early evidence on the fiscal and health-system impacts of Illinois’s immigrant health coverage expansions and the broader implications for states pursuing inclusive health policies for immigrants.

Methods

Data Collection

This cohort study was determined to be exempt from review and informed consent by the institutional review board of the University of Chicago Biological Sciences Division and University of Chicago Medical Center because it did not involve human participants. We adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

We obtained hospital-level data of uncompensated care from RAND Hospital Data from January 1, 2017, to December 31, 2023, which standardizes Centers for Medicare & Medicaid Services hospital cost report data.¹⁸ These data include measures of hospital revenues, expenses, uncompensated care, and service volumes, harmonized to the calendar year. We restricted our analysis to general medical and surgical hospitals in states like Illinois that expanded Medicaid and did not implement other major health insurance expansion shocks during the study period. States were excluded if they had not expanded Medicaid by 2017, implemented Medicaid expansion during the study period, or adopted state-funded health insurance programs for undocumented immigrants. We excluded hospital-year observations that had missing outcomes and missing covariates. To ensure hospitals contributed prepolicy information for assessment of parallel trends and entropy balancing, we excluded hospitals that were not observed in 2019. Missing cases analysis comparing the characteristics of hospitals included and excluded from the study sample is provided in eTable 1 in Supplement 1.

We incorporated county-level data from the American Community Survey 5-year estimates (2018-2022)¹⁹ and County Health Rankings²⁰ to capture population size, income, poverty, self-reported race and ethnicity, nativity (using percentage Hispanic and percentage non-Hispanic White residents as proxies), and insurance coverage for Illinois and comparison counties. To account for pandemic-related shocks, we included county-level measures of COVID-19 inpatient bed utilization, emergency department visit rates, and admission rates from US Department of Health and Human Services facility-level data from 2020 to 2023.²¹ In the prepandemic period (2017-2019), all values were coded as zero.

Variables

Uncompensated Care

We examined 2 components of uncompensated care: bad debt and charity care. Bad debt captures services rendered for which payment is unlikely or unattainable. Charity care reflects the value of services provided at free or discounted rates to patients who qualify for financial assistance. While other studies often include unreimbursed and total uncompensated care,^11,12 we focused on bad debt and charity care because the HBIA and HBIS programs likely shifted many patients from uninsured to covered status under state-funded insurance, whereas unreimbursed or total uncompensated care primarily capture partial payments under Medicaid.

To ensure comparability across hospitals of varying size and billing practices, we also calculated cost-adjusted measures of bad debt and charity care by multiplying reported charges by each hospital’s cost-to-charge ratio (CCR). The CCR converts hospital charges into estimated costs, providing a more comparable estimate of the actual resources expended in delivering uncompensated care. Hospital charges vary substantially across institutions and often do not reflect underlying costs, therefore cost-adjusted outcomes complement the unadjusted charge-based measures by accounting for differences in hospital pricing practices. To provide additional context regarding the magnitude of uncompensated care relative to hospital financial resources, we also examined adjusted bad debt and adjusted charity care as a share of total hospital operating expenses. To address skewed distributions, all uncompensated care measures were log-transformed. The RAND Hospital Data processing procedures replaced a small number of implausible or zero values with interpolated values; therefore, the analytic variables did not contain zero values at the time of log transformation.

Noncitizen Immigrant Composition

We additionally examined county-level variation in immigrant density, measured by the share of residents who were foreign-born noncitizens relative to the total population. Counties were categorized into those with high and low percentages of immigrants based on whether their ratio of foreign-born noncitizens to population fell below or at and above the statewide median.

Controls

We include hospital and county-level characteristics that might be related to uncompensated care, consistent with prior literature. Time-invariant controls include hospital ownership type and teaching status; time-varying controls include hospital capacity, staffing, county age structure, Hispanic population share, White population share, uninsured rate, unemployment rate, population density, child poverty rate, and median household income. We also accounted for potential disruptions to hospital finances by including county-level measures of COVID-19 case and mortality rates. We estimated both adjusted and unadjusted models because some variables, such as the uninsured rate, may partially mediate the effects of HBIA and HBIS; results were substantively similar across specifications.

Statistical Analysis

We conducted analysis from October 1, 2024, to June 12, 2026. We first compared baseline hospital and county characteristics between hospitals in Illinois and comparison states using descriptive statistics and standardized mean differences (SMDs). To improve comparability between groups, we used entropy balancing to construct weights that balanced the distribution of pretreatment hospital- and county-level characteristics between Illinois hospitals and hospitals in comparison states.²² Entropy balancing weights were estimated once at the hospital level using prepolicy baseline characteristics and subsequently applied to all hospital-year observations in the longitudinal panel. Balancing variables included baseline hospital characteristics (eg, bed size, ownership, teaching status) and county demographic and socioeconomic characteristics. Covariate balance before and after weighting was assessed using SMDs, with absolute SMD values less than 0.10 indicating adequate balance.

To estimate the association between Illinois’ HBIA and HBIS programs and hospital financial outcomes, we used a difference-in-differences (DiD) design with 2-way fixed effects comparing changes in outcomes among Illinois hospitals following program implementation with contemporaneous changes among hospitals in comparison states. Models were estimated using both the unweighted and the entropy-balanced samples. Charge-based bad debt and charity care outcomes were estimated using all hospitals in the analytic sample with nonmissing outcome data. Cost-adjusted bad debt and charity care outcomes required hospital-specific CCRs; therefore, analyses of these outcomes were restricted to hospital-year observations with nonmissing CCRs.

Our primary analysis examined the combined postpolicy period (2020-2023), capturing the implementation of both HBIS and HBIA, relative to the prepolicy period (2017-2019). To account for the phased implementation of Illinois’ immigrant coverage expansions, we conducted sensitivity analyses examining the HBIS-only period (2020-2021) separately. We also estimated models excluding the COVID-19 years (2020-2021) to assess the robustness of findings to pandemic-related disruptions. Next, we estimated a triple difference-in-differences-in-differences (DDD) model to assess whether program effects varied by local immigrant context, comparing changes in uncompensated care among hospitals in high-immigrant counties vs those in low-immigrant counties in Illinois relative to control states.

Models included hospital-level and county-level covariates, hospital fixed effects, and year fixed effects to account for observable and time-invariant differences across hospitals. Standard errors were clustered at the state level. Because treatment was assigned at the state level and Illinois was the only treatment state, inference based on conventional clustered SEs may be unreliable and P values may be overly optimistic. To address this concern, we additionally conducted wild bootstrap randomization inference, which combines cluster-level bootstrap resampling with state-level placebo treatment assignments and is designed for settings with a small number of treated clusters. Wild bootstrap randomization inference P values are reported alongside conventional state-clustered P values.

We conducted several sensitivity analyses to assess the robustness of our findings to sample selection and hospital composition. Analyses of the unadjusted bad debt and charity care outcomes included all hospitals with nonmissing outcome data, while analyses of the cost-adjusted bad debt and charity care outcomes were restricted to hospitals with nonmissing CCRs, as these outcomes cannot be calculated in the absence of a CCR. To evaluate whether differences in analytic samples influenced the results, we reestimated adjusted uncompensated care models using imputed CCRs for hospitals with missing values with the mean CCR for hospitals of the same state, hospital type, and size category, consistent with recommendations from the Agency for Healthcare Research and Quality²³ (eTable 2 in Supplement 1). Second, we reestimated all unadjusted and adjusted models using the common sample of hospitals with complete data for both sets of outcomes to ensure that comparisons between charge-based and cost-adjusted estimates were not driven by differences in the analytic sample (eTable 3 in Supplement 1). In addition, we reestimated models including all states regardless of Medicaid expansion timing or other insurance expansions (eTable 4 in Supplement 1) and restricted analyses to a balanced panel of hospitals observed in all study years (eTable 5 in Supplement 1).

To assess the robustness of our findings to potential violations of the parallel trends assumption, we implemented the sensitivity analysis proposed by Rambachan and Roth²⁴ in both the original and entropy-balanced samples. Rather than requiring parallel trends to hold exactly, this approach allows for deviations from parallel trends by restricting the magnitude of posttreatment trend differences relative to those observed during the pretreatment period. Specifically, we used the relative magnitude restriction, where M̅ = 1 corresponds to exact parallel trends and allows posttreatment violations to be as large as the largest pretreatment violation. In the HBIA-HBIS setting, this approach addresses the possibility that unobserved state-specific, hospital-market, or macroeconomic shocks may have differentially affected Illinois hospitals relative to comparison hospitals. We report conditional least-favorable confidence intervals across a range of M̅ values; confidence intervals that continue to exclude zero at larger M̅ values indicate greater robustness to violations of the parallel trends assumption.

All statistical tests were 2 sided, and statistical significance was defined as P < .05. Analyses were conducted using R, version 4.5.1 (R Project for Statistical Computing).

Results

Descriptive Statistics

The study sample comprised 161 Illinois hospitals in the treated group and 1902 hospitals in the control group states (eFigure 1 in Supplement 1). Table 1 summarizes the differences between Illinois (treatment) and control hospitals. Baseline differences in hospital characteristics, including bed counts, number of full-time employees, and hospital ownership, between Illinois and comparison hospitals were generally small, with all SMDs below 0.10, a commonly used threshold for acceptable balance. However, at the county level, Illinois hospitals were located in counties with a higher mean (SD) share of Hispanic population (11.4% [10.1%] vs 9.1% [10.6%]; SMD = 0.23), higher median income ($58 121 [range, $34 982-$100 325] vs $56 102 [range, $23 968-$134 050]; SMD = 0.17), but a lower mean (SD) share of uninsured population (7.8% [2.0%] vs 9.1% [3.5%]; SMD = −0.68). Entropy balancing improved the comparability of hospital and county-level covariates in Illinois and control hospitals.

Table 1. Baseline Characteristics^a.

Characteristic	Study group		SMDc
	Treatment status (Illinois) (n = 976)	Control status (n = 11 337)b
Before entropy balancing
Bed count	148.9 (158.1)	150.7 (194.7)	−0.01
No. of full-time employees)	1011.5 (1411.8)	1133.6 (1903.2)	−0.09
Hospital type, No. (%)
For-profit	85 (8.7)	1258 (11.1)	0.09
Government or nonprofit	891 (91.3)	10 079 (88.9)
Teaching status, No. (%)
Nonteaching	697 (71.4)	7732 (68.2)	−0.07
Teaching	279 (28.6)	3605 (31.8)
Age <18 y, %	22.1 (1.7)	21.9 (2.7)	0.10
Age ≥65 y, %	16.7 (3.0)	17.7 (4.0)	−0.33
Hispanic population, %	11.4 (10.1)	9.1 (10.6)	0.23
Uninsured population, %	7.8 (2.0)	9.1 (3.5)	−0.68
Unemployed population, %	5.4 (0.7)	4.8 (1.5)	0.87
Population density, persons per square mile	1605.9 (2282.9)	1525.6 (6725.0)	0.04
White population, %	72.9 (20.2)	75.8 (19.1)	−0.15
Children living in poverty, %	19.2 (5.3)	19.6 (7.8)	−0.08
Median income (range), $	58 120 (34 982-100 325)	56 102 (23 968-134 050)	0.17
Log bad debt	15.4 (1.2)	15.1 (1.5)	0.23
Log charity care	13.7 (1.9)	13.0 (2.2)	0.34
Log-adjusted bad debt	14.4 (0.9)	14.3 (1.2)	0.11
Log-adjusted charity care	12.9 (1.5)	12.3 (1.9)	0.42
Log-adjusted ratio of bad debt to operating expense ratio	−4.5 (0.7)	−4.5 (0.8)	0.03
Log-adjusted ratio of charity care to operating expense	−6.0 (1.0)	−6.5 (1.3)	0.54
After entropy balancing
Bed count	148.9 (158.1)	148.9 (168.8)	<0.001
No. of full-time employees	1011.5 (1411.8)	1011.5 (1285.2)	<0.001
Hospital type, No. (%)			<0.001
For-profit	85 (8.7)	85 (8.7)	<0.001
Government or nonprofit	891 (91.3)	891 (91.3)
Teaching status, No. (%)			<0.001
Nonteaching	697 (71.4)	697 (71.4)	<0.001
Teaching	279 (28.6)	279 (28.6)
Age <18 y, %	22.1 (1.7)	22.1 (2.8)	<0.001
Age ≥65 y, %	16.7 (3.0)	16.7 (3.2)	<0.001
Hispanic population, %	11.4 (10.1)	11.4 (15.0)	<0.001
Uninsured population, %	7.8 (2.0)	7.8 (3.2)	<0.001
Unemployed population, %	5.4 (0.7)	5.4 (2.8)	<0.001
Population density, persons per square mile	1605.9 (2282.9)	1605.9 (5913.1)	<0.001
White population, %	72.9 (20.2)	72.9 (21.6)	<0.001
Children living in poverty, %	19.2 (5.3)	19.2 (8.3)	<0.001
Median income (range), $	58 120 (34 982-100 325)	58 120 (34 982-100 325)	<0.001
Log bad debt	15.4 (1.2)	15.2 (1.4)	0.20
Log charity care	13.7 (1.9)	13.0 (2.0)	0.33
Log-adjusted bad debt	14.4 (0.9)	14.3 (1.1)	0.15
Log-adjusted charity care	12.9 (1.5)	12.3 (1.8)	0.47
Log-adjusted ratio of bad debt to operating expense	−4.5 (0.7)	−4.5 (0.7)	0.04
Log-adjusted ratio of charity care to operating expense	−6.0 (1.0)	−6.5 (1.2)	0.58

Abbreviation: SMD, standardized mean difference.

^a Data are presented as mean (SD) unless noted otherwise.

^b Control group values after entropy balancing are weighted. Categorical counts are reported as weighted counts.

^c Absolute SMD values below 0.10 are commonly interpreted as indicating adequate balance.

DiD Models

Estimates from the 2-way fixed effects DiD models indicate that Illinois’ HBIA and HBIS programs were associated with meaningful reductions in uncompensated care. In the full sample from 2017 to 2023 (Table 2), the programs were associated with a 7.5% decline in bad debt (ATT, −0.08 [95% CI, −0.11 to −0.04]) and a 19.4% decline in charity care (ATT, −0.19 [95% CI, −0.26 to −0.13]) in covariate-adjusted models. Results were similar using cost-adjusted measures. The programs were associated with a 13.0% reduction in adjusted bad debt (ATT, −0.13 [95% CI, −0.17 to −0.09]) and a 19.4% reduction in adjusted charity care (ATT, −0.19 [95% CI, −0.27 to −0.12]). When scaled relative to hospital operating expenses, HBIA and HBIS were associated with a 10.1% reduction in adjusted bad debt as a share of operating expenses (ATT, −0.10 [95% CI, −0.14 to −0.06]) and a 16.6% reduction in adjusted charity care as a share of operating expenses (ATT, −0.17 [95% CI, −0.24 to −0.09]).

Table 2. DiD and Entropy Balanced–DiD Estimates Using 2-Way Fixed Effects Model^a.

Measure	2-Way fixed-effects DiD estimates with covariates
	Bad debt	Charity care	Adjusted bad debt	Adjusted charity care	Adjusted bad debt per operating expense	Adjusted charity care per operating expense
Full sample
ATT (95% CI)	−0.08 (−0.11 to −0.04)	−0.19 (−0.26 to −0.13)	−0.13 (−0.17 to −0.09)	−0.19 (−0.27 to −0.12)	−0.10 (−0.14 to −0.06)	−0.17 (−0.24 to −0.09)
Clustered P value	<.001	<.001	<.001	<.001	<.001	<.001
WBRI P value	.40	.15	.15	.26	.26	.21
Fixed effects	Year and hospital	Year and hospital	Year and hospital	Year and hospital	Year and hospital	Year and hospital
No. of observations	12 313	12 313	9342	9342	9336	9336
R 2	0.931	0.923	0.883	0.909	0.802	0.824
Within R2	0.012	0.006	0.013	0.006	0.009	0.006
Entropy-balanced sample
ATT (95% CI)	−0.09 (−0.15 to −0.04)	−0.170.17 (−0.25 to −0.09)	−0.14 (−0.20 to −0.08)	−0.15 (−0.233 to −0.07)	−0.12 (−0.18 to −0.06)	−0.12 (−0.20 to −0.04)
Clustered P value	.002	<.001	<.001	<.001	<.001	.006
WBRI P value	.12	.06	.11	.23	.08	.22
Fixed effects	Year and hospital	Year and hospital	Year and hospital	Year and hospital	Year and hospital	Year and hospital
No. of observations	12 313	12 313	9342	9342	9336	9336
R 2	0.911	0.928	0.863	0.910	0.761	0.832
Within R2	0.039	0.015	0.038	0.016	0.028	0.013

Abbreviations: ATT, posttreatment coefficient from a 2-way fixed effects regression with hospital and year fixed effects; DiD, difference-in-differences; WBRI, wild bootstrap randomization inference.

^a All outcomes are log-transformed. All models include hospital-level controls for total bed count, number of full-time equivalent employees, ownership type, and teaching status; county-level controls for age structure, percentage of Hispanic population, uninsured rate, unemployment rate, population density, percentage of White population, child poverty rate, and median household income; and COVID-19 bed utilization, emergency department visits, and admission rates. The 95% CIs and clustered P values were calculated using robust SEs clustered at the state level.

However, treatment and comparison hospitals differed on several baseline characteristics; therefore, we estimated entropy-balanced models to improve covariate balance while retaining the full analytic sample. Results from the entropy-balanced sample were generally consistent with those from the unweighted sample. Covariate-adjusted models indicated a 9.5% reduction in bad debt (ATT, −0.09 [95% CI, −0.15 to −0.04]) and a 17.1% reduction in charity care (ATT, −0.17 [−95% CI, −0.25 to −0.09]) (Table 2). Cost-adjusted outcomes displayed similar patterns, with a 14.5% decline in adjusted bad debt (ATT, −0.14 [95% CI, −0.20 to −0.09]) and a 14.7% decline in adjusted charity care (ATT, −0.15 [95% CI, −0.23 to −0.07]). HBIA and HBIS were also associated with declines in adjusted bad debt and adjusted charity care as a share of hospital operating expenses at 11.9% (ATT, −0.12 [95% CI, −0.18 to −0.06) and 12.3% (ATT, −0.12 [95% CI, −0.20 to −0.04), respectively.

To assess the robustness of our findings, we estimated alternative specifications that varied the definition and timing of policy exposure. During the early HBIS period (2017-2021) (eTable 6 in Supplement 1), estimated effects were smaller but remained statistically significant for most outcomes. In the main sample, HBIS was associated with an 8.9% reduction in adjusted bad debt (ATT, −0.09 [95% CI, −0.14 to −0.04), a 9.5% reduction in charity care (ATT, −0.09 [95% CI, −0.16 to −0.03]), and a 9.3% reduction in adjusted charity care (estimate, −0.09 [95% CI, −0.17 to −0.02]). In the entropy-balanced sample, declines were also observed for adjusted bad debt (ATT, −0.09 [95% CI, −0.15 to −0.04]), charity care (ATT, −0.10 [95% CI, −0.19 to −0.02]), and adjusted charity care (ATT, −0.09 [95% CI, −0.18 to −0.004]). However, in both the full and entropy-balanced samples, declines in bad debt were indistinguishable from zero (ATT, −0.02 [95% CI −0.06 to 0.02] and −0.03 [95% CI, −0.08 to 0.03], respectively). When excluding COVID-19 years (eTable 7 in Supplement 1), program effects become slightly larger but consistent with the primary results. Adjusted models indicated a 9.4% reduction in bad debt (ATT, −0.09 [95% CI, −0.13 to −0.05]) and a 23.5% reduction in charity care (ATT, −0.24 [95% CI, −0.33 to −0.14]). Cost-adjusted outcomes showed similarly large reductions in adjusted bad debt (14.6%; ATT, −0.15 [95% CI, −0.19 to −0.10]) and adjusted charity care (23.8%; ATT, −0.24 [95% CI, −0.33 to −0.15]). Results from the entropy-balanced sample were comparable, with reductions in bad debt (12.0%; ATT, −0.12 [95% CI, −0.19 to −0.05]), charity care (20.4%; ATT, −0.20 [95% CI, −0.30 to −0.11]), adjusted bad debt (17.1%; ATT, −0.17 [95% CI, −0.25 to −0.09]), and adjusted charity care (18.1%; ATT, −0.18 [95% CI, −0.28 to −0.08]).

Results were also robust to alternative sample definitions. Estimated reductions in uncompensated care were substantively similar when all states initially excluded because of Medicaid expansion timing or immigrant coverage policies were included in the comparison group (eTable 4 in Supplement 1) and when analyses were restricted to hospitals observed in all study years (eTable 5 in Supplement 1). Results using imputed CCRs (eTable 2 in Supplement 1) were somewhat attenuated for adjusted bad debt outcomes but remained statistically significant in both the main sample (ATT, −0.05 [95% CI, −0.09 to −0.01]) and entropy-balanced sample (ATT, −0.07 [95% CI, −0.13 to −0.01]). Despite modest variation in effect sizes across specifications, the direction and overall conclusions remained unchanged.

DDD Model: Immigrant Context

To examine whether the outcomes associated with HBIA and HBIS differed by local concentration of immigrant population, we estimated DDD models comparing hospitals located in counties with high vs low concentrations of non–US-born noncitizen residents. The coefficient on the DDD estimator is defined by the interaction among the postpolicy period, Illinois treatment status, and high immigrant concentration. A negative coefficient indicates that uncompensated care declined more sharply among counties with a high concentration of immigrants than among those with a low concentration in Illinois, relative to corresponding differences observed in comparison states.

In the full sample with covariate adjustment (Table 3), the DDD coefficient indicated larger reductions in bad debt among hospitals located in counties with high immigrant populations. Specifically, HBIA and HBIS were associated with a 35.0% greater reduction in bad debt (ATT, −0.35 [95% CI, −0.42 to −0.28]) and a 25.5% greater reduction in adjusted bad debt (ATT, −0.26 [95% CI, −0.34 to −0.17]) in counties with a high relative to a low concentration of immigrants. Similar patterns were observed for measures scaled to operating expenses, with a 27.9% greater reduction in adjusted bad debt as a share of operating expenses (ATT, −0.28 [95% CI, −0.36 to −0.20]).

Table 3. DDD Estimates by County-Level Immigrant Concentration^a.

Measure	Bad debt	Charity care	Adjusted bad debt	Adjusted charity care	Adjusted bad debt per operating expense	Adjusted charity care per operating expense
Full sample
Main DDD effects (95% CI)
Post policy × treatment status	0.10 (0.05 to 0.15)	−0.24 (−0.32 to −0.17)	0.03 (−0.03 to 0.09)	−0.34 (−0.42 to −0.26)	0.07 (0.01 to 0.13)	−0.30 (−0.38 to −0.22)
Post policy × high immigrant concentration	0.01 (−0.06 to 0.07)	−0.01 (−0.05 to 0.04)	−0.03 (−0.10 to 0.04)	−0.05 (−0.13 to 0.04)	−0.01 (−0.07 to 0.06)	−0.02 (−0.11 to 0.06)
Post policy × treatment status × high immigrant concentration	−0.35 (−0.42 to −0.28)	0.10 (0.04 to 0.16)	−0.26 (−0.34 to −0.17)	0.25 (0.15 to 0.34)	−0.28 (−0.36 to −0.20)	0.22 (0.12 to 0.32)
Clustered P value	<.001	.002	<.001	<.001	<.001	<.001
WBRI P value	.08	.54	.34	.40	.18	.49
Baseline controls	Yes	Yes	Yes	Yes	Yes	Yes
Fixed effects	Year and hospital	Year and hospital	Year and hospital	Year and hospital	Year and hospital	Year and hospital
No. of observations	12 313	12 313	9342	9342	9336	9336
R 2	0.932	0.923	0.883	0.910	0.803	0.825
Within R2	0.016	0.006	0.016	0.007	0.012	0.006
Entropy-balanced sample
Main DDD effects (95% CI)
Post policy × treatment status	0.07 (−0.01 to 0.14)	−0.22 (−0.30 to −0.15)	−0.01 (−0.10 to 0.07)	−0.29 (−0.37 to −0.22)	0.02 (−0.06 to 0.11)	−0.26 (−0.34 to −0.17)
Post policy × high immigrant concentration	−0.002 (−0.11 to 0.11)	0.01 (−0.05 to 0.08)	−0.07 (−0.17 to 0.03)	−0.03 (−0.13 to 0.06)	−0.05 (−0.15 to 0.04)	−0.02 (−0.13 to 0.09)
Post policy × treatment status × high immigrant concentration	−0.31 (−0.41 to −0.22)	0.10 (0.03 to 0.16)	−0.19 (−0.30 to −0.09)	0.25 (0.16 to 0.34)	−0.22 (−0.33 to −0.11)	0.22 (0.13 to 0.31)
Clustered P value	<.001	.005	.001	<.001	<.001	<.001
WBRI P value	.013	.11	.24	.09	.16	.09
Baseline controls	Yes	Yes	Yes	Yes	Yes	Yes
Fixed effects	Year and hospital	Year and hospital	Year and hospital	Year and hospital	Year and hospital	Year and hospital
No. of observations	12 313	12 313	9342	9342	9336	9336
R 2	0.912	0.928	0.864	0.910	0.764	0.833
Within R2	0.055	0.016	0.048	0.020	0.039	0.016

Abbreviations: DDD, difference-in-differences-in-differences; WBRI, wild bootstrap randomization inference.

^a Data are from 2013 to 2023. Coefficients are estimated from DDD specifications including post policy × treatment status × high immigrant concentration and all lower-order interaction terms. All outcomes are log-transformed. All models include hospital and year fixed effects and baseline hospital- and county-level controls, including hospital capacity, number of full-time equivalent employees, ownership type, teaching status, county demographic and socioeconomic characteristics, and COVID-19 bed utilization, emergency department visits, and admission rates. The 95% CIs and clustered P values were calculated using robust SEs clustered at the state level. All P values correspond to the DDD interaction coefficient.

Results from the entropy-balanced sample (Table 3) show a consistent pattern in declines in bad debt. Counties with high concentrations of immigrants experienced a 31.3% greater reduction in bad debt (ATT, −0.31 [95% CI, −0.41 to −0.22]), a 19.5% greater reduction in adjusted bad debt (ATT, −0.19 [95% CI, −0.30 to −0.09]), and a 22.1% greater reduction in adjusted bad debt as a share of operating expenses (ATT, −0.22 [95% CI, −0.33 to −0.11]) relative to counties with a low concentration of immigrants.

In contrast, the corresponding DDD coefficients for charity-care outcomes across both the full sample and entropy-balanced sample were less consistent across models. In the full sample, the coefficient for charity care was positive (ATT, 0.10; 95% CI, 0.04-0.16), while those for adjusted charity care and adjusted charity care as a share of operating expenses were also positive (ATT, 0.25 [95% CI, 0.15-0.34] and 0.22 [95% CI, 0.12-0.32], respectively). Similar patterns were observed in the entropy-balanced sample. These findings suggest that, unlike bad debt, reductions in charity care were not disproportionately concentrated among hospitals located in counties with high concentrations of immigrants.

Sensitivity to Violations of Parallel Trends

The Figure and eFigure 2 in Supplement 1 present the HonestDiD sensitivity analysis proposed by Rambachan and Roth,²⁴ which evaluates the robustness of the estimated policy effects to potential violations of the parallel trends assumption. The estimated reductions in adjusted bad debt remained statistically distinguishable from zero across the full range of M̅ values examined, indicating substantial robustness to violations of the parallel trends assumption. Unadjusted bad debt estimates were also relatively robust, remaining statistically distinguishable from zero under moderate departures from parallel trends. In contrast, charity care outcomes exhibited greater sensitivity to departures from parallel trends, with 95% CIs including zero at smaller M̅ values. Results from the entropy-balanced sample were substantively similar, with adjusted bad debt estimates remaining robust across the range of M̅ values examined and charity care outcomes exhibiting greater sensitivity to potential violations of parallel trends (eFigure 2 in Supplement 1). Overall, these findings suggest that the most robust evidence pertains to reductions in hospital bad debt following implementation of the HBIA and HBIS programs.

Figure. Bar Graphs Showing HonestDiD Sensitivity Analysis for Hospital Uncompensated Care Outcomes.

Each panel displays bars representing 95% CIs for the estimated treatment effects across increasing values of the sensitivity parameter M̅. Blue bars indicate the original difference-in-differences estimate; orange bars show the 95% CIs under increasing values of M̅. Any 95% CIs that remain below zero indicate that the estimated treatment effect is robust to increasingly severe departures from the parallel trends assumption. C-LF indicates conditional least-favorable.

Discussion

In this cohort study of hospitals in Illinois and comparison states, we examined the association between Illinois’s HBIS and HBIA programs and hospitals’ uncompensated care. Across multiple specification, including entropy-balanced models that achieved improved covariate balance between treatment and control hospitals, we found that these coverage expansions were consistently associated with meaningful reductions in hospital bad debt. Reductions in charity care were also observed, although the evidence was less consistent and robust. These findings provide early empirical evidence on the financial consequences of state-funded health insurance expansions for undocumented and noncitizen immigrants.

Our findings extend the literature on insurance expansions and uncompensated care by showing that HBIA and HBIS yields fiscal benefits similar to those observed following Medicaid expansion under the ACA.^{11,25,26,27,28,29} Prior studies of Medicaid expansion have documented reductions in uncompensated care, typically ranging from 24% to 35%.^30,31,32,33 In our preferred approach using an entropy-balanced sample, we find that hospital bad debt declined by approximately 9.4% after the implementation of HBIA and HBIS. Cost-adjusted bad debt declined by 14.6%, and adjusted bad debt as a share of hospital operating expenses declined 11.9%, suggesting that the programs reduced uncompensated care both in absolute terms and relative to hospitals’ overall financial resources.

Although HBIA and HBIS targeted a substantially smaller population than ACA expansion, several findings support the plausibility of these effect sizes. First, reductions were observed for raw and adjusted outcomes, indicating that findings are not driven solely by hospital size or accounting practices. As shown by the triple-difference models, declines were even larger in counties with larger noncitizen immigrant populations, where the policy would be expected to have the greatest impact due to the greater concentration of potentially eligible immigrants. In addition, although smaller in magnitude, reductions in bad debt were already evident in the early HBIS-only period and became larger following implementation of HBIA, consistent with the expansion of eligibility to a larger population.

At peak enrollment, the Illinois programs covered approximately 70 000 individuals, representing roughly 8% of the state’s uninsured population. Although HBIA and HBIS covered a smaller share of the uninsured population than Medicaid expansion under the ACA, prior studies^33,34 show that reductions in uncompensated care do not always increase in direct proportion to the number of individuals newly insured and vary substantially across states and programs. HBIA and HBIS enrollees were older and more likely to have experienced prolonged periods without insurance, suggesting a population with potentially high baseline hospital utilization and uncompensated care. As a result, coverage for a comparatively small share of the uninsured population in Illinois could still have generated sizable reductions, rendering our estimates plausible and consistent with prior evidence.

Importantly, the most robust evidence pertains to bad debt rather than charity care. Reductions in charity care were observed in the full and entropy-balanced DiD models, ranging from 14.7% to 19.4%. However, HonestDID sensitivity analysis suggested that charity care outcomes were more sensitive to potential violations of the parallel trends assumption than bad debt outcomes. In addition, the DDD analysis by county immigration composition did not demonstrate larger declines in charity care among hospitals located in counties with high concentrations of immigrants. The stronger findings for bad debt are consistent with the mechanisms by which HBIA and HBIS would be expected to affect hospital finances. Bad debt reflects care delivered with the expectation of payment that ultimately goes unpaid, making it directly responsive to insurance coverage expansions that convert uninsured encounters into reimbursed services. Charity care, by contrast, is governed by hospital-specific financial assistance policies that vary widely across institutions³⁵ and may therefore be less sensitive to changes in insurance coverage, particularly in the short term.

These findings have important implications for states considering the adoption, expansion, or retrenchment of health coverage for undocumented and noncitizen immigrants. As of 2019, 6 other states and Washington, DC, had implemented some form of state-funded health coverage for income-eligible adults regardless of immigration status. Fiscal concerns are frequently cited as a barrier to these policies and weighed heavily on Illinois’s decision to end the HBIA program in 2025. Our results indicate that such programs may partially offset their costs by reducing hospitals’ uncompensated care burdens, particularly in regions with large immigrant populations. These findings also have implications beyond hospital accounting. By mitigating uncompensated care burdens, immigrant insurance expansions may contribute to hospital financial stability in ways that protect service capacity, support clinician workforce stability, and sustain access to care in immigrant-serving communities.³⁶ In this sense, policies that reduce uncompensated care may function not only as coverage expansions but also as investments in the resilience of local health systems.

Limitations

This study has several limitations. First, although we used multiple quasi-experimental approaches, unobserved differences between Illinois and control hospitals may remain. Because HBIA and HBIS were implemented in a single state, we cannot fully rule out the influence of unmeasured state-level factors or differential state-specific trends that coincided with the policy implementation. Consistent with this concern, inference was more conservative under wild bootstrap randomization inference. Our outcome measures rely on hospital cost reports, which are known to vary in reporting practices and may contain measurement error.

Conclusions

This cohort study provides evidence that Illinois’ state-funded health coverage for undocumented and noncitizen immigrants was associated with reductions in hospitals’ bad debt, with effect sizes that scaled with local immigrant eligibility. These findings have important implications for health equity, given persistently high rates of uninsurance among undocumented and noncitizen immigrants, driven by long-standing exclusion from federally funded health programs. Such exclusions contribute not only to unequal access to care but also to disproportionate financial strain on safety-net hospitals serving immigrant communities. By reducing hospitals’ exposure to unpaid care in these settings, immigrant-inclusive coverage policies may help stabilize safety-net providers. As states continue to debate the scope and sustainability of immigrant-inclusive health policies, our findings underscore the role of coverage expansions as mechanisms for advancing equity within health systems.

Supplement 1.

eFigure 1. Study Sample Construction and Exclusion Criteria for Hospitals and States Included in the Analysis

eFigure 2. Event-Study Estimates of the Effects of HBIA and HBIS on Uncompensated Care Outcomes: Entropy-Balanced Sample

eTable 1. Missing Case Analysis: Characteristics of Hospitals Included in and Excluded From the Study Sample

eTable 2. DiD and Entropy-Balanced DiD Estimates Using Imputed Cost-to-Charge Ratios (CCRs) (2017-2023)

eTable 3. DiD and Entropy-Balanced DiD Estimates Using Common Sample (2017-2023)

eTable 4. DiD and Entropy-Balanced DiD Estimates Using All States as Comparators (2017-2023)

eTable 5. DiD and Entropy-Balanced DiD Estimates in a Balanced Hospital Sample (2017-2023)

eTable 6. Difference-in-Differences (DiD) and Propensity Score Matched DiD Estimates of HBIA Effects (2017-2021)

eTable 7. DiD and Entropy-Balanced DiD Estimates Excluding COVID-19 Years (2017-2019 vs 2022-2023)

Supplement 2.

Data Sharing Statement