Health Care Costs and Immigration Status Among US Adults

Study Design, Data Source, and Population

This serial cross-sectional study was conducted using pooled data from the 2010-2023 Integrated Public Use Microdata Series Medical Expenditure Panel Survey (MEPS).¹⁵ MEPS is a nationally representative survey of the US civilian noninstitutionalized population that provides detailed data on sociodemographic characteristics, insurance coverage, health care use, and medical expenditures.¹⁵ All adult respondents aged 19 to 64 years with data available on immigration status and total health care expenditures were included in the study cohort. Analyses were restricted to this age group because insurance eligibility pathways and coverage structures differed substantially for children and Medicare-eligible adults.⁸ Respondents with missing data on age or family income required to estimate poverty level were excluded. Because this study used publicly available, deidentified data, it was considered to be exempt from ethics review by The Ohio State University Institutional Review Board; informed consent was obtained by MEPS at enrollment.¹⁶ The study followed applicable data-use requirements and was reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

Exposures, Covariates, and Outcomes

The primary exposure was policy era, categorized as pre-ACA expansion (2010-2013: reference), post-ACA expansion (2014-2018), and post-PCRP (2019-2023).^4,11,17 The post-PCRP period began in 2019, the first full calendar year after publication of the final rule, although policy-associated changes may have begun after issuance of the proposed rule in 2018. The post-PCRP era was examined separately to capture temporal changes occurring after the initial ACA expansion period and within the broader policy environment following the 2018 proposed and 2019 finalized federal Public Charge Rule.¹¹ Because this study assessed temporal trends, the post-PCRP period was interpreted in a broader policy and structural context.

Participants were categorized by nativity as US-born or immigrants using the MEPS nativity variable. To assess economic heterogeneity, the sample was stratified by family income relative to the federal poverty level (FPL); we selected 138% FPL a priori because it approximates the Medicaid expansion eligibility cutoff.⁸ Covariates selected a priori included age, sex, race and ethnicity, marital status, education, and family size (as a continuous variable). Sex was obtained from the MEPS survey instrument and categorized as male or female. Race and ethnicity were obtained from the MEPS survey instrument and were based on respondent-reported information. These variables were assessed because the study evaluated differences in health care coverage, use, and expenditures across sociodemographic groups. Race was categorized as American Indian or Alaska Native, Asian, Black, Hawaiian or Pacific Islander, White, and multiracial. Ethnicity was categorized as Hispanic or non-Hispanic. Marital status was categorized as married, divorced, widowed, separated, or living with partner; education was categorized as less than high school, high school or General Educational Development (GED), some college, or college or higher. The primary outcomes were person-year measures of insurance coverage, health care use, and health care expenditures. Health care use outcomes included office-based visits, outpatient visits, emergency department visits, hospital discharges, and prescription fills. Expenditure outcomes included total, self-paid, office-based, inpatient, and prescription expenditures.⁸ All expenditure outcomes were converted to constant 2023 US dollars using the annual gross domestic product implicit price deflator. Expenditures for each survey year were multiplied by the ratio of the 2023 deflator to the deflator for that year.

Statistical Analysis

Baseline characteristics were summarized among observations with known nativity and positive annual person-level survey weights. Categorical variables were reported as unweighted frequencies with survey-weighted percentages, approximately normally distributed continuous variables as weighted means with SDs, and non-normally distributed variables as weighted median values with IQRs. Covariate data were handled using complete-case analysis, and the extent of missingness was quantified overall and compared between US-born and immigrant respondents. Within each income stratum, US-born and immigrant adults were compared using design-based Pearson χ² tests for categorical variables and survey-weighted regression-based tests for continuous variables; multinomial logistic regression with joint Wald testing was used when sparse cells limited direct comparison. Era-specific univariable outcomes were examined by nativity and income stratum. Insurance coverage was reported as frequency with weighted percentage and use and expenditure outcomes as weighted median with IQR. Across-era comparisons used design-based χ² tests for insurance coverage and survey-weighted regression-based tests for health care use and expenditures. Multivariable models were estimated separately for adults with or without low income, with the pre-ACA era as the reference. Separate nativity-stratified models estimated within-group changes across eras, whereas models including nativity, policy era, and nativity by policy era interaction terms assessed whether temporal changes differed between US-born and immigrant adults. Survey-weighted logistic, Poisson, and linear regression models were used for insurance coverage, health care use, and expenditures, respectively. Results were reported as adjusted odds ratios (AORs), adjusted incidence rate ratios (AIRRs), or β coefficients with 95% CI. Models were adjusted for age, sex, race and ethnicity, marital status, education, and family size. Overall, interaction P values were obtained using survey-adjusted joint Wald tests.

Because expenditure outcomes included zeros and were substantially right-skewed among respondents with positive expenditures, survey-weighted 2-part models were performed as sensitivity analyses. In one sensitivity analysis, the post-PCRP period was subdivided into post-PCRP and prepandemic (2019), pandemic (2020-2021), and recovery and postpandemic (2022-2023) periods, while retaining the pre-ACA and post-ACA periods. Within each income stratum, models included nativity, temporal period, and nativity by period interaction terms and otherwise used the same covariate adjustment and outcome-specific specifications as the primary analyses. In a second supplemental analysis, the non–low-income stratum was divided into adults with family income of 138% to 400% FPL and 400% FPL or higher, and the same survey-weighted nativity by policy era models were repeated within each subgroup.

All analyses incorporated annual MEPS person-level survey weights, strata, and primary sampling units across the pooled 2010 to 2023 data. Annual weights were used as provided and not rescaled by the number of pooled years; therefore, weighted totals represent cumulative population person-years rather than a single-year US population or unique-person denominator. Accordingly, unweighted person-year counts were reported for analytic sample size, whereas weighted percentages, means, and regression estimates were used for population-representative inference. Analyses were conducted at the person-year level; because MEPS panels span multiple calendar years, some respondents contributed observations in more than 1 year, and within-person correlation was not modeled separately. The Benjamini-Hochberg procedure was applied to overall nativity by policy era interaction P values across the 11 outcomes within each prespecified income stratum. Nominal tests were 2 sided, with P < .05 considered statistically significant and false discovery rate–adjusted q <.05 considered significant. Survey-weighted adjusted annual expenditure trends and era-level estimates were also presented graphically by income group and nativity. Analyses were conducted using Stata/SE, version 18.0 (StataCorp LLC). Data were analyzed from April 1, 2026, to April 15, 2026.

Baseline Characteristics

Among 238 927 positive-weight person-year observations, 1752 had missing model covariates (0.73% unweighted; 0.52% survey-weighted). Survey-weighted missingness was more frequent among observations for immigrants than US-born adults (1.37% vs 0.32%; P < .001). Complete-case adjusted analyses included 237 175 person-year observations from 126 724 unique adults, including 67 076 female adults (50.8%) and 59 648 male adults (49.2%), with a survey-weighted mean (SD) age of 40.8 (13.2) years; the analytic sample included 56 561 low-income and 180 614 non-low-income person-year observations (eFigure 1 in Supplement 1). Because MEPS is a serial cross-sectional survey and the analysis was conducted at the person-year level, participant loss to follow-up was not applicable.

The descriptive sample included 57 227 person-year observations for individuals with low income, including 38 983 observations for US-born and 18 244 immigrant adults. Compared with US-born adults with low income, immigrants were more frequently Hispanic and married, reported fewer functional limitations, and had lower insurance coverage and Medicaid enrollment (Table 1).

Among 181 700 adults without low income, 137 328 were US-born and 44 372 were immigrants. Immigrants were more frequently Hispanic and married, reported fewer functional limitations, and had lower overall and private insurance coverage (Table 2).

Coverage, Use, and Expenditures Across Policy Eras by Immigration Status

In unadjusted analyses, insurance coverage increased across eras in both nativity groups, while median total expenditures remained consistently lower among immigrants than US-born adults in both income strata (eTables 1 and 2 in Supplement 1).

Among adults with low income, adjusted insurance coverage increased in both groups, but gains were smaller among immigrants. Relative to the pre-ACA era, coverage increased among US-born adults in the post-ACA (AOR, 2.07; 95% CI, 1.90-2.26) and post-PCRP (AOR, 2.84; 95% CI, 2.55-3.17) periods and among immigrants in the post-ACA (AOR, 1.57; 95% CI, 1.38-1.79) and post-PCRP (AOR, 2.29; 95% CI, 1.98-2.65) periods. US-born adults experienced increases in office-based and outpatient visits, whereas immigrants had no significant increase in office-based visits and an increase in outpatient visits only during the post-PCRP period (AIRR, 2.03; 95% CI, 1.43-2.86). Prescription fills decreased among immigrants during the post-PCRP period (AIRR, 0.76; 95% CI, 0.65-0.89).

Total expenditures increased among US-born adults with low income during the post-ACA (β = $1249.70; 95% CI, $529.10-$1970.30) and post-PCRP (β = $1684.60; 95% CI, $883.10-$2486.20) periods. No significant change was observed among immigrants with low income in either period (post-ACA: β = −$359.80; 95% CI, −$977.90 to $258.40; post-PCRP: β = $220.20; 95% CI, −$680.10 to $1120.50) (Table 3).

Among adults without low income, coverage increased comparably in both nativity groups. Office-based and outpatient visits increased, emergency department visits remained unchanged, and hospital discharges and prescription fills decreased during the post-PCRP period. Total expenditures increased among US-born adults in the post-ACA (β = $530.30; 95% CI, $257.40-$803.10) and post-PCRP (β = $1788.60; 95% CI, $1487.70-$2089.50) periods. Among immigrants, total expenditures did not increase significantly in the post-ACA period but increased in the post-PCRP period (β = $974.10; 95% CI, $517.50-$1430.70) (Table 4).

Interaction and Sensitivity Analyses

Among adults with low income, the overall nativity by era interactions were significant for insurance coverage (P = .002; q = 0.008) and total expenditures (P < .001; q = 0.003) after false discovery rate adjustment. Prescription fills, office-based expenditures, and prescription expenditures had significant P for interactions <.05 but the interactions did not remain significant after adjustment. Among adults without low income, only total expenditures (P for interaction = .007; q = 0.04) and prescription expenditures (P for interaction = .005; q = 0.04) remained significant after adjustment (eTables 3 and 4 in Supplement 1).

Findings of survey-weighted 2-part models were generally consistent with the primary analyses. Total expenditures increased in post-ACA and post-PCRP periods among US-born adults with low income but not among immigrants with low income. Post-PCRP total expenditures increased among both US-born and immigrant adults without low income (eTable 5 in Supplement 1).

In the 5-period sensitivity analysis, differences observed in the combined 2019 to 2023 period were more evident during the pandemic or recovery periods than in 2019 alone. Among adults with low income, the differential change among immigrants was significant during 2020 to 2021 for office-based visits (AIRR, 0.72; 95% CI, 0.57-0.92) and total expenditures (β = −$1905.00; 95% CI, −$3354.00 to −$456.00), but neither was significant in 2019. Among adults without low income, differential total expenditures were not significant in 2019 but were lower among immigrants during 2020 to 2021 (β = −$908.00; 95% CI, −$1675.00 to −$141.00) and 2022 to 2023 (β = −$1141.00; 95% CI, −$1798.00 to −$484.00) (eTables 6 and 7 in Supplement 1).

Further stratification of adults without low income demonstrated income heterogeneity. Among immigrants with income of 138% to 400% FPL, total and office-based expenditures did not increase significantly in either postpolicy period, whereas inpatient expenditures decreased post-PCRP (β = −$522.00; 95% CI, −$888.00 to −$156.00). Among immigrants with income at 400% FPL or higher, total, office-based, and prescription expenditures increased post-ACA (β = $690.33, $243.21, and $220.48, respectively) and post-PCRP (β = $2300.32, $1051.96, and $464.11, respectively), and office-based and outpatient visits also increased post-PCRP (AIRR, 1.42 and 3.12, respectively) (eTables 8 and 9 in Supplement 1). Adjusted annual and era-level estimates consistently showed lower expenditures among immigrants, with the largest disparities among adults without low income (Figure; eFigure 2 and eFigures 3 and 4 in Supplement 1).

Limitations

This study has several limitations. MEPS data are observational and cross-sectional, limiting causal inference. Immigrant status was defined by nativity and therefore did not capture legal or documentation status, visa category, citizenship pathway, duration of US residence, or eligibility-related classifications that may directly affect insurance coverage and health care access. As a result, the immigrant group likely included individuals with substantially different structural barriers and coverage options, and residual heterogeneity may have attenuated or obscured subgroup-specific associations. Although inflation-adjusted expenditures increased during some periods, spending should not be interpreted as a direct measure of health care engagement because it also reflects service intensity, treatment mix, clinical need, and price changes not fully captured by general inflation adjustment. The post-PCRP interval also overlapped with other policy changes and the COVID-19 pandemic, complicating attribution. Use and expenditure measures were partly self-reported and may reflect recall bias; because clinical need was not directly assessed, lower health care use rates could not be distinguished from lower need or care-seeking preferences. Although the MEPS data survey design variables were incorporated, respondent-level correlation across repeated person-year observations was not modeled separately, which may have modestly affected variance estimates. Finally, covariate missingness was low but more frequent among immigrants, creating potential for limited selection bias.