{"subscriber":false,"subscribedOffers":{}} Effects Of The ACA’s Health Insurance Marketplaces On The Previously Uninsured: A Quasi-Experimental Analysis | Health Affairs

### Research Article

Affordable Care Act

# Effects Of The ACA’s Health Insurance Marketplaces On The Previously Uninsured: A Quasi-Experimental Analysis

Affiliations
1. Anna L. Goldman ([email protected]) is a General Internal Medicine Fellow at Harvard Medical School and Cambridge Health Alliance, in Massachusetts.
2. Danny McCormick is an associate professor of medicine at Harvard Medical School and director of the Division of Social and Community Medicine in the Department of Medicine, Cambridge Health Alliance.
3. Jennifer S. Haas is a professor of medicine at Brigham and Women’s Hospital, in Boston, Massachusetts.
4. Benjamin D. Sommers is an associate professor of health policy and economics in the Department of Health Policy and Management, Harvard T. H. Chan School of Public Health, and an associate professor of medicine at Brigham and Women’s Hospital, both in Boston, Massachusetts.
PUBLISHED:Free Accesshttps://doi.org/10.1377/hlthaff.2017.1390

## Abstract

Descriptive studies have suggested that the Affordable Care Act’s (ACA’s) health insurance Marketplaces improved access to care. However, no evidence from quasi-experimental studies is available to support these findings. We used longitudinal survey data to compare previously uninsured adults with incomes that made them eligible for subsidized Marketplace coverage (138–400 percent of the federal poverty level) to those who had employer-sponsored insurance before the ACA with incomes in the same range. Among the previously uninsured group, the ACA led to a significant decline in the uninsurance rate, decreased barriers to medical care, increased the use of outpatient services and prescription drugs, and increased diagnoses of hypertension, compared to a control group with stable employer-sponsored insurance. Changes were largest among previously uninsured people with incomes of 138–250 percent of poverty, who were eligible for the ACA’s cost-sharing reductions. Our quasi-experimental approach provides rigorous new evidence that the ACA’s Marketplaces led to improvements in several important health care outcomes, particularly among low-income adults.

Evidence on the effects of the Affordable Care Act (ACA) continues to grow, even as the elimination of the individual mandate penalty and ongoing political debate has made its future impact uncertain. Numerous studies have examined impacts of the expansion of eligibility for Medicaid,13 but less research has focused on the ACA’s health insurance Marketplaces.

The Marketplaces offer subsidized coverage to families with incomes of 138–400 percent of the federal poverty level in Medicaid expansion states and to those with incomes of 100–400 percent of poverty in nonexpansion states, as well as providing cost-sharing reductions for people with incomes below 250 percent of poverty. In the Marketplaces’ first open enrollment period, from October 2013 to March 2014, 6.7 million people purchased insurance there;4 by 2017, 10.3 million had done so.5 One study estimated that subsidized Marketplace coverage accounted for 40 percent of the ACA’s reduction in the uninsurance rate.6

The Marketplaces were designed to improve the affordability and accessibility of health care for low- and middle-income populations. Observational studies have shown substantial increases in coverage among those eligible for Marketplace subsidies,79 increased likelihood of having a usual source of care and access to medications,10 and increased rates of blood pressure screening.11,12 However, to our knowledge, there has been no quasi-experimental evidence on the Marketplaces’ impact. Rigorous evaluations of changes in coverage and access in the Marketplace-eligible income range are needed, particularly because higher cost-sharing requirements13 and narrow networks in Marketplace plans14 may limit the generalizability of evidence from the ACA’s Medicaid expansion. Finally, observational analyses of people gaining Marketplace coverage that do not include a control group run the risk of multiple forms of bias—including those related to secular trends in health outcomes, the economic recovery, and selective enrollment among a subset of eligible applicants.

Our study assessed the impact of newly available Marketplace coverage on previously uninsured nonelderly adults who became eligible for subsidized Marketplace plans in 2014. Using longitudinal survey data, we performed a difference-in-differences analysis comparing health and spending outcomes among those who were uninsured before the ACA to outcomes among a cohort of adults with stable employer-sponsored insurance in the income range eligible for subsidized Marketplace coverage.

## Study Data And Methods

### Data

We analyzed data from the nationally representative Medical Expenditure Panel Survey (MEPS), conducted by the Agency for Healthcare Research and Quality. Data are collected for each member of surveyed households over a two-year period and are available in two-year, longitudinal data sets (for simplicity, we refer below to the first and second years of data as year 1 and year 2 for each observation).15 We used three longitudinal data sets, for 2011–12, 2012–13, and 2013–14 (see the study design diagram in online appendix A).16 We began our study period in 2011 since earlier data sets could be subject to changes related to the Great Recession, as well as the immediate effects of the ACA’s dependent coverage provision, which took effect beginning in 2010. The 2013–14 data set captures the year before and the year after implementation of major ACA policies on January 1, 2014. The 2013–14 panel was the cohort in which the intervention took place, which we refer to as the ACA-affected cohort, while the 2011–12 and 2012–13 panels were combined to form a pre-ACA cohort.

### Study Design And Sample

Our study used a difference-in-differences analysis, measuring changes in health insurance coverage, access to care, use, and costs before and after January 1, 2014.

Our intervention group (members of the 2013–14 panel) consisted of adults ages 18–63 with year 1 (2013) family incomes of 138–400 percent of poverty who had been uninsured for at least six months in year 1. Family income was defined based on the concept of the health insurance unit (an adult, their spouse, and any dependent children). Most people in the intervention group (subject to the limitations discussed below) became eligible to purchase premium-subsidized insurance on state health insurance Marketplaces in 2014, and those with incomes below 250 percent of poverty also qualified for cost-sharing reductions. We chose six months of uninsurance as our cutoff to capture the longer-term uninsured population, as opposed to people who had had brief gaps in coverage because of job changes or temporary fluctuations in income.

The control group (members of the 2011–12 and 2012–13 panels combined) consisted of nonelderly adults with incomes of 138–400 percent of poverty who had employer-sponsored insurance for all twelve months of year 1 (2011 for the first panel and 2012 for the second panel). Although there is no perfect control group for the creation of the ACA Marketplaces, previous research demonstrates that the ACA had little detectable effect on employer-sponsored insurance. This indicates that the group with such coverage was largely insulated from changes affecting our intervention group,1719 thus making it a useful comparator to account for potential secular trends in the economy and the health care system.

### Outcome Variables

Our sample was defined based on year 1 characteristics (age, income, and insurance status). All outcome measures were drawn from year 2 (2014 for the intervention group and 2012 and 2013 for the control group) of the data sets. Therefore, each person contributed only one observation for each measure to the final data set.

Our primary coverage outcome was health insurance status as of the last month of year 2. Insurance type is self-reported in MEPS. Categories of insurance included private employer-sponsored insurance, private nongroup insurance (including Marketplace coverage), Medicaid, and uninsured. As a secondary outcome, we also measured the number of insured months in year 2.

Access-to-care outcomes (again based on year 2) were self-reported inability to obtain needed medical care, delays in filling medication prescriptions, and access to a usual source of care. Utilization outcomes were year 2 receipt of a preventive visit, number of outpatient visits, number of emergency department (ED) visits, number of prescription medications used (including refills), presence of any inpatient admission, receipt of a blood pressure check, and laboratory testing for cholesterol levels. We also assessed changes in the prevalence of year 2 diagnosis rates of three common chronic conditions: high cholesterol, diabetes, and hypertension.

Cost-related outcomes were year 2 total health spending and out-of-pocket health spending. Total health spending is the sum of expenditures by all payers, including out-of-pocket spending, for all health care services excluding over-the-counter medications.20 Out-of-pocket health spending includes all payments made by individuals for outpatient visits, inpatient admissions, ED visits, and prescription drugs.21 MEPS cost data are independently verified whenever possible with providers of medical services and pharmacies.15,22 To account for the skewed distribution of spending data, we analyzed these outcomes with a two-stage model. First, we modeled changes in the proportion of people with any health expenses, and second, we analyzed changes in the logarithm of spending among those with any expenditures. All spending figures were adjusted to 2014 dollars using the Consumer Price Index.23

### Statistical Analysis

We used linear regression models for both binary and nonbinary outcomes for ease of interpretation.24 Our models adjusted for key demographic characteristics and economic indicators (all from year 1) including age, sex, race, region, marital status, family size, family income, birthplace outside the US, and employment status. All models included two binary indicators: previously uninsured (intervention) versus previously insured via stable employer-sponsored insurance (control) group, and pre-ACA versus ACA-affected cohort. The interaction term between these two variables yielded our difference-in-differences estimates.

In a subgroup analysis, we divided our study population into a low-income group (people with family incomes of 138–250 percent of poverty), whose members were eligible for both premium subsidies and cost-sharing reductions through Marketplace plans, and a middle-income group (those with family incomes of 251–400 percent of poverty), whose members were eligible only for premium subsidies.

Our primary analysis defined income based on the health insurance unit, but we also conducted a sensitivity analysis using the less-restrictive MEPS definition for family income.21

We tested for diverging trends before 2014 for our treatment and control groups. This analysis used a placebo difference-in-differences model as if the ACA had been implemented in 2013,25 and it compared study outcomes between the two groups in the 2011–12 versus the 2012–13 panel.

All regressions were performed with SAS, version 9.4, using survey-based procedures that accounted for the complex sample design of and weights in MEPS.

### Limitations

Our study was subject to several limitations. First, it included only a single year of post-ACA data and therefore captured only the first-year effects of the Marketplaces. The ACA’s effects would be expected to increase after the first year, given that enrollment continued to rise26 and recipients would have had more time to take advantage of newly acquired benefits. Changes in some outcomes, especially self-reported health status, might not have been apparent in the first year.27 The use of only a single post-ACA year despite the availability of 2015 MEPS data was unavoidable given our study design, which relied on two-year longitudinal data from MEPS and specifically exploited the single cohort that overlapped the launch of the ACA’s major policies. We feel that this trade-off was worthwhile, given the strength of our study design compared to prior Marketplace-specific analyses.

Second, although MEPS is a relatively large data set, our inclusion criteria resulted in a smaller study sample. This could have limited our power to detect small changes that may be clinically or economically relevant, and it also precluded our analyzing outcomes that were relevant to only a small subset of our sample, such as mammography or Pap smear rates. Additionally, following the majority of studies in the coverage expansion literature,3,2831 we did not adjust for multiple comparisons, which might have increased the risk of false positives.

Third, the MEPS data do not include state identifiers in publicly available files. This prevented us from examining the effects of state-specific programs such as the use of insurance navigators and the impact of state- versus federally administered Marketplaces, although this may have been challenging given sample-size limitations. In at least one study, state-administered Marketplaces were shown to be more effective in expanding coverage.6 Lack of state identifiers also eliminated the possibility of controlling for the effect of Medicaid expansion in some states, which affects the income range in each state eligible for Marketplace coverage.

Fourth, not all uninsured adults might have been eligible for Marketplace subsidies, as a result of either immigration status or having an affordable offer of employer-sponsored insurance that they did not accept. Our data set did not allow us to evaluate these factors. If anything, this would have led us to underestimate the changes in coverage and access among the truly eligible population.

Finally, as stated above, no perfect control group exists for the ACA’s Marketplaces, given that the policy that created them was implemented on a national level at a single point in time. Many observed differences, as well as potentially unobservable differences, exist between our intervention and control groups. However, the validity of our difference-in-differences study design depends not on the balance of characteristics between the study groups, but rather on the assumption that trends would have been comparable for both study groups in the absence of the policy change. Results from the placebo tests of the pre-ACA period provide support for this assumption. The use of our control group strengthened the analysis by reducing bias from mean reversion and secular trends in our estimates of Marketplace effects on the previously uninsured.

## Study Results

Our study population included 9,653 adults, of whom 5,770 had continuous employer-sponsored insurance (4,047 in the pre-ACA period) and 3,883 were uninsured for six months or more in year 1 (2,688 in in the pre-ACA period) (exhibit 1). Compared to the group with stable employer-sponsored insurance (the control group), the uninsured (intervention) group was younger and more likely to be male, be Hispanic, and live in the South or West. Mean family income was lower in the intervention group, whose members were also somewhat less likely to be employed before the ACA—though the majority of adults in both groups were employed.

 Control groupa Intervention group Number 4,047 2,688 Mean age (years) 40.7 38.8 Sex Female 53.9% 42.8% Male 46.1 57.2 Race/ethnicity Non-Hispanic black 11.9% 11.8% Hispanic 13.7 31.2 Asian 7.4 6.9 Non-Hispanic white or other 67.0 50.1 Mean family size 2.6 2.1 Region Northeast 17.1% 11.9% Midwest 25.9 18.2 South 35.5 43.3 West 21.5 26.7 Foreign-born 15.4 31.3 Employed 86.9 82.7 Mean family income \$48,539.00 \$35,272.00

SOURCE Authors’ analysis of Medical Expenditure Panel Survey (MEPS) longitudinal data for the period 2011–13. NOTES Panels for 2011–12 and 2012–13 were combined to form a study cohort for the period before implementation of major Affordable Care Act policies. Members of the control group had employer-sponsored insurance for the whole first year of the study (2011 for the first panel and 2012 for the second panel). Members of the intervention group were uninsured for at least six months in the first year of the study. All comparisons are significant ($p < 0.001$).

We examined the changes in health insurance status before and after the ACA for each study group (exhibit 2). Compared to the control group, the intervention group had a large drop in the uninsurance rate by the end of year 2 associated with ACA implementation (adjusted difference-in-differences: −10.8 percentage points). Also compared to the control group, the intervention group had an increase in the likelihood of having any private insurance of 8.8 percentage points ($p < 0.001$) (data not shown). More specifically, rates of nongroup private coverage (including Marketplace plans) increased in the intervention group relative to the control group (adjusted difference-in-differences: 6.0 percentage points), while Medicaid participation increased to a lesser degree (adjusted difference-in-differences: 2.5 percent) and enrollment in employer-sponsored insurance did not increase significantly. In 2014, 6.8 percent of people in the intervention group and 1.1 percent of those in the control group had Marketplace coverage.

 Control group Intervention group Difference-in-differences (percentage points) Insurance status Pre-ACA ACA-affected Difference (percentage points) Pre-ACA ACA-affected Difference (percentage points) Unadjusted Adjusted Private, ESI 91.0% 93.8% 2.8 18.0% 23.4% 5.4 2.6 2.8 Private, non-ESI (includes Marketplace) 1.0 1.8 0.8 2.2 8.9 6.7 5.8**** 6.0**** Marketplace —a 1.1 —a —a 6.8 —a —a —a Medicaid 1.5 2.0 0.5 4.5 7.6 3.1 2.6* 2.5* Uninsured 6.7 2.3 −4.4 72.4 57.2 −15.2 −10.6**** −10.8****

SOURCE Authors’ analysis of Medical Expenditure Panel Survey (MEPS) longitudinal data for 2011–14. NOTES The control and intervention groups are explained in the notes to exhibit 1. The members of the MEPS panels for 2011–12 and 2012–13 are in the pre–Affordable Care Act (ACA) cohort. The members of the panel for 2013–14 are in the ACA-affected cohort. Insurance status is self-reported in December of year 2 of the panel. Adjusted analyses controlled for age, sex, race, region, marital status, family size, family income, birthplace outside the US, and employment status. Age was transformed to a categorical variable, with five ranges (ages 18–24, 25–34, 35–44, 45–54, and 55 and older). Race/ethnicity was coded as black, Hispanic, Asian, or white or other. Family size was converted to an ordinal categorical variable. ESI is employer-sponsored insurance.

aNot applicable, because Marketplace coverage did not exist before 2014.

*$p < 0.10$

****$p < 0.001$

Monthly rates of nongroup private coverage increased sharply for the intervention group in the first three months of 2014 remained relatively constant for the control group (exhibit 3).

Regression-based difference-in-differences estimates are presented in exhibit 4, and appendix B16 presents plots of several outcomes over time for the intervention versus the control group. The share of adults in the former group who reported that they were unable to get necessary care declined compared to the control group (adjusted difference-in-differences: −1.9 percentage points). There were no significant changes in the probability of delaying prescription medications or having a usual source of care. Among adults with any expenditures, the adjusted estimate of total medical spending increased for the intervention versus control group (adjusted difference-in-differences: 14.3 percentage points), and out-of-pocket spending also increased (adjusted difference-in-differences: 9.7 percentage points). However, neither of these estimates reached statistical significance.

SOURCE Authors’ analysis of longitudinal data for 2011–14 from the Medical Expenditure Panel Survey. NOTES The pre-ACA cohort is explained in the notes to exhibit 2. Members of the low-income subsample had family incomes of 138–250 percent of poverty, while members of the middle-income subsample had incomes of 251–400 percent. Difference-in-differences estimates refer to differences between the intervention and the control groups (explained in the notes to exhibit 1) over time. Information contained in the following variables are based solely on self-report: unable to get necessary medical care; delayed filling prescriptions; had a usual source of care; had any preventive visit; had blood pressure check; had cholesterol check; self-reported health; diagnosis of high cholesterol; diagnosis of diabetes; diagnosis of hypertension. All models used survey weighting and accounted for complex survey design. Adjusted analyses controlled for age, sex, race, region, marital status, family size, family income, birthplace outside the US, and employment status. Age was transformed to a categorical variable, with five ranges (ages 18–24, 25–34, 35–44, 45–54, and 55 and older). Race/ethnicity was coded as black, Hispanic, Asian, or white or other. Family size was converted to an ordinal categorical variable. FPL is federal poverty level. OOP is out of pocket. ED is emergency department.

aFor percentage values, column shows percentage points.

bAll payers.

cAmong people with nonzero spending.

*$p < 0.10$

**$p < 0.05$

***$p < 0.01$

****$p < 0.001$

The average number of outpatient visits increased among the intervention, compared to the control, group (adjusted difference-in-differences: 0.87 visits). The rate of inpatient admission in the intervention group increased by 2.5 percentage points. The number of prescriptions filled increased for intervention versus control group members (adjusted difference-in-differences: 1.82 medications). The proportion of those having an ED visit did not change significantly, and neither did the rates of having a preventive visit or poor or fair self-reported health. The likelihood of receiving a diagnosis of hypertension increased among intervention versus control groups (adjusted difference-in-differences: 7.02 percentage points).

Results of a sensitivity analysis using a broader definition of family to measure income yielded results similar to those of our primary analysis (appendix C).16 In comparison to members of the control group, the number of adults in the intervention group who reported that they were unable to get necessary care showed a larger decline (adjusted difference-in-differences: −2.4 percentage points; $p = 0.03$), and the number of outpatient visits increased more robustly (adjusted difference-in-differences: 1.31 additional visits; $p = 0.03$).

### Subgroup Analysis

Both the low-income group (people with family incomes of 138–250 percent of poverty) and the middle-income group (those with incomes of 251–400 percent of poverty) experienced large increases in coverage by the end of year 2 (adjusted difference-in-differences: 11.1 percent and 13.2 percent, respectively) compared to the control group (exhibit 4).

Changes in access and utilization were generally larger in the lower-income subgroup than the higher-income subgroup, though there was some variation across outcomes. For the lower-income subgroup, rates of being unable to access care declined significantly (adjusted difference-in-differences: −2.7 percentage points) and outpatient utilization increased (adjusted difference-in-differences: 1.80 visits), compared to the control group. The rate of inpatient admission was greater among intervention than control members (adjusted difference-in-differences: 3.5 percentage points). The lower-income subgroup also showed a large increase in the number of prescription fills (adjusted difference-in-differences: 3.91 prescriptions per year) and in the proportion having a hypertension diagnosis (adjusted difference-in-differences: 6.1 percentage points). The proportion with a diagnosis of high cholesterol increased in the lower-income subgroup (adjusted difference-in-differences: 5.67 percentage points), which was not evident in the full sample.

In the higher-income subgroup, the treatment group experienced a decreased likelihood of delaying medications (adjusted difference-in-differences: −3.3 percentage points) and an increased likelihood of having a hypertension diagnosis (adjusted difference-in-differences: 7.72 percentage points). Total medical spending among those with any expenditures also rose in this intervention subgroup relative to the control group (adjusted difference-in-differences: 38.3 percent).

### Placebo Tests

Our placebo analysis showed no significant differential changes in the pre-ACA cohort for nearly all study outcomes (appendix D).16 Only one outcome—cholesterol screening—showed a significant difference ($p < 0.10$). However, this indicated that the likelihood of screening was decreasing in the pre-ACA period for the intervention group, which would bias us against finding a positive result in this group. Overall, these findings demonstrate comparable pre-ACA trends in outcomes for the intervention and control groups, which supports our difference-in-differences approach.

## Discussion

Using rich data from a longitudinal national survey, our study offers new evidence on the effect of the ACA’s health insurance Marketplaces on subsidy-eligible adults. Insurance coverage increased significantly among previously uninsured adults, with the majority of growth due to Marketplace coverage. These coverage gains were accompanied by a reduction in barriers to needed care, more outpatient visits, increased inpatient care, additional use of prescription drugs, and an increase in the rates of diagnosis of hypertension. Other outcomes, including out-of-pocket and total spending, did not change significantly. Our findings from the first year after the ACA’s implementation suggest that the Marketplaces are contributing to a broad range of positive effects for previously uninsured adults. Our analysis of income subgroups suggests that improvements in access to care were concentrated among the lower-income adults, who were eligible for cost-sharing subsidies.

These results demonstrate that despite initial challenges in the rollout of the Marketplaces, they were successful in improving coverage among many long-term uninsured adults. However, even after the expansion of coverage availability, more than half of this group remained uninsured at the end of 2014, which indicates that significant outreach efforts are still needed. Of course, our use of a data set ending in 2014 means that longer-term gains in coverage were not captured in our results. Of note, only 1 percent of the control group was found to have purchased insurance through the Marketplace in 2014. This indicates that crowd-out of employer-sponsored insurance by the new Marketplace subsidies was rare—which is consistent with prior studies of the ACA.6,19

Our findings of coverage changes are supported by several observational studies that have documented similar-size changes in the uninsurance rate using data from the National Health Interview Survey,8 Health Reform Monitoring Survey,32 American Community Survey,6 and Gallup-Healthways Well-Being Index.33

We also identified several significant changes in access to care and utilization after coverage expansion, consistent with studies of Medicaid expansion under the ACA.1,2 This suggests that despite concerns about potentially high cost sharing and narrow networks in the ACA,14,34 subsidized Marketplace coverage has still improved care for many middle-income adults. Of note, increases in coverage with employer-sponsored insurance and Medicaid accounted for just under half of the coverage gains in our previously uninsured group, indicating that our findings are not solely attributable to new Marketplace coverage. Still, our results are consistent with those of several observational studies of people in the subsidized Marketplace income range or who report enrolling in Marketplace plans.12,30,35

To our knowledge, no previous work has used quasi-experimental methods to examine changes in health care outcomes among a population of previously uninsured adults eligible for Marketplace subsidies. Examining longer-term trends in coverage and other health outcomes among these people is an area for future research.

Despite finding some beneficial changes in access to care and chronic disease diagnosis, many of our study outcomes—including total spending, preventive health measures, and self-reported health—did not change significantly, in contrast to prior studies of the Medicaid expansion.13 This may reflect the relatively small sample size and the fact that the study included data on only the first year after ACA implementation. Many of the people who obtained new insurance in 2014 began coverage midway through the year and thus had limited time to take advantage of new opportunities to seek care. However, it is also possible that Marketplace coverage may be less effective at improving some of these measures than Medicaid, and at least some descriptive evidence suggests that concerns exist about these plans’ cost-sharing requirements. For instance, a Henry J. Kaiser Family Foundation survey found that 36 percent of Marketplace enrollees were somewhat or very dissatisfied with the burden of their insurance deductible, compared with 25 percent of those with employer-sponsored insurance coverage.36 While our study showed that Marketplace coverage has been beneficial, our post-ACA estimates show that many adults still experience barriers to care, and cost-related challenges may be more common in Marketplace plans with high cost sharing than in Medicaid—which traditionally has required only minimal out-of-pocket spending.37

Our subgroup analysis revealed that improvements in access to care were generally larger in the low-income subgroup (those whose family incomes were 138–250 percent of poverty). This has important implications for the ACA’s cost-sharing reductions, which are available to Marketplace enrollees with incomes below 250 percent of poverty. Insurers are required to reduce out-of-pocket spending for this group and receive payments from the federal government for these expenses. However, the administration of President Donald Trump has announced a decision to stop making these payments.38 Our findings suggest that these cost-sharing subsidies may be an important driver of the positive impacts of Marketplace coverage. However, cost-sharing subsidies are only one potential explanation for the larger effect found in the low-income subgroup. Another is that this group simply had more unmet medical needs before the ACA.

## Conclusion

Our study provided some of the first quasi-experimental evidence on the effects of the ACA’s Marketplace plans on middle- and low-income nonelderly adults who previously lacked insurance. We identified several notable improvements in health care access and use that previously uninsured adults experienced. These benefits were larger for low-income adults, who were eligible to receive cost-sharing reductions. The Trump administration’s decision to cease payments for cost-sharing subsidies to insurers, as well as the elimination of the individual mandate penalty, could further erode Marketplace coverage and undo some of these gains. However, despite notable improvements, we also found that many barriers to care remain for this population. Legislative changes to Marketplace plans that increase the generosity of subsidies and boost enrollment might produce greater progress on a range of outcomes related to health care.

## ACKNOWLEDGMENTS

This study had no direct funding. Anna Goldman’s salary is supported by an Institutional National Research Service Award (No. T32HP12706). Danny McCormick received salary support for his contributions from internal funds of the Department of Medicine at Cambridge Health Alliance. Jennifer Haas received salary support for her contributions from an Institutional National Research Service Award (No. T32HP12706) and NIH 1UL1TR001102-03. Benjamin Sommers reports grants from the Agency for Healthcare Research and Quality, Commonwealth Fund, Robert Wood Johnson Foundation, and REACH Healthcare Foundation in the past twelve months.