Importance There is increased interest in public reporting of, and linking financial incentives to, the performance of organizations on health equity metrics, but variation across organizations could reflect differences in performance or selection bias.
Objective To assess whether differences across health plans in sex- and age-adjusted racial disparities are associated with performance or selection bias.
Design, Setting, and Participants This cross-sectional study leveraged a natural experiment, wherein a southern US state randomly assigned much of its Medicaid population to 1 of 5 plans after shifting to managed care in 2012. Enrollee-level administrative claims and enrollment data from 2011 to 2015 were obtained for self-identified Black and White enrollees. The analyses were limited to Black and White Medicaid enrollees because they accounted for the largest percentages of the population and could be compared with greater statistical power than other groups. Data were analyzed from June 2021 to September 2024.
Exposures Plan enrollment via self-selection (observational population) vs random assignment (randomized population).
Main Outcomes and Measures Annual counts of primary care visits, low-acuity emergency department visits, prescription drug fills, and total spending. For observational and randomized populations, models of each outcome were fit as a function of plan indicators, indicators for race, interactions between plan indicators and race, and age and sex. Models estimated the magnitude of racial differences within each plan and tested whether this magnitude varied across plans.
Results Of 118 101 enrollees (mean [SD] age, 9.3 [7.5] years; 53.0% female; 61.4% non-Hispanic Black; and 38.6% non-Hispanic White), 70.2% were included in the randomized population, and 29.8% were included in the observational population. Within-plan differences in primary care visits, low-acuity emergency department visits, prescription drug use, and total spending between Black and White enrollees were large but did not vary substantially and were not statistically significantly different across plans in the randomized population, suggesting minimal effects of plans on racial differences in these measures. In contrast, in the observational population, racial differences varied substantially across plans (standard deviations 2-3 times greater than in the randomized population); this variation was statistically significant after adjustment for multiple testing, except for emergency department visits. Greater between-plan variation in racial differences in the observational population was only partially explained by sampling error. Stratifying by race did not bring observational estimates of plan effects meaningfully closer to randomized estimates.
Conclusions and Relevance This cross-sectional study showed that selection bias may mischaracterize plans’ relative performance on measures of health care disparities. It is critical to address disparities in Medicaid, but adjusting plan payments based on disparity measures may have unintended consequences.
Importance Work requirements are a controversial feature of US safety-net programs, with some policymakers seeking to expand their use. Little is known about the demographic, clinical, and socioeconomic characteristics of individuals most likely to be negatively impacted by work requirements.
Objective To examine the association between work requirements and safety-net program enrollment.
Design, Setting, and Participants This cohort study included Medicaid and Supplemental Nutrition Assistance Program (SNAP) enrollees in Connecticut. The impact of SNAP work requirements for able-bodied adults without dependents—the target population—was estimated using a triple-differences research design comparing outcomes before and after the policy (first difference) in affected and exempted towns (second difference) between the targeted population and untargeted parents and caregivers (third difference). SNAP and Medicaid enrollment trends were assessed for a 24-month period, and the characteristics of individuals most likely to lose coverage were examined. Data were collected from August 2015 to April 2018, and data were analyzed from August 2022 to September 2024.
Exposures The reintroduction of SNAP work requirements in 2016.
Main Outcomes and Measures Proportion of enrollees disenrolled from SNAP and Medicaid.
Results Of 81 888 Medicaid enrollees in Connecticut, 46 872 (57.2%) were female, and the mean (SD) age was 36.6 (7.0) years. Of these, 38 344 were able-bodied adults without dependents, of which 19 172 were exposed to SNAP work requirements, and 43 544 were parents or caregivers exempted from SNAP work requirements. SNAP coverage declined 5.9 percentage points (95% CI, 5.1-6.7), or 25%, following work requirements. There were no statistically significant changes in Medicaid coverage (0.2 percentage points; 95% CI, −1.4 to 1.0). Work requirements disproportionately affected individuals with more chronic illnesses, targeted beneficiaries who were older, and beneficiaries with lower incomes. Individuals with diabetes were 5 percentage points (95% CI, 0.8-9.3), or 91%, likelier to lose SNAP coverage than those with no chronic conditions; older SNAP beneficiaries (aged 40 to 49 years) with multiple comorbidities were 7.3 percentage points (95% CI, 4.3-11.3), or 553%, likelier to disenroll than younger beneficiaries (aged 25 to 29 years) without chronic conditions; and households with the lowest incomes were 18.6 percentage points (95% CI, 11.8-25.4), or 204%, likelier to lose coverage than the highest income SNAP beneficiaries.
Conclusions and Relevance In this cohort study, SNAP work requirements led to substantial reductions in SNAP coverage, especially for the most clinically and socioeconomically vulnerable. Work requirements had little effect on Medicaid coverage, suggesting they did not lead to sufficient increases in employment to transition beneficiaries off the broader safety net.