Data for this study were used from two main sources. First, we obtained 2020 market-level SVI data (excluding Puerto Rico) from the CDC public website.15 The SVI is a composite index score representing each market’s percentile ranking (range from 0 to 1). We then merged market-level SVI scores with MA plan-by-market-level information on contract/plan identifier, plan types, service areas, and star ratings obtained from CMS Medicare Advantage Landscape File for the contract year 2020.

This study used counties to define a local market. Private insurers contract with CMS to offer MA plans, and each contract can include multiple plans in each local market, which may include a county or region. Our dataset excluded Alaska since 2020 MA plan data were missing for that state. The final analytic sample consisted of 3113 county-level markets with non-missing SVI information.

We counted the total number of MA plans at the market level based on their CMS quality rating (number of stars assigned by CMS). We excluded MA plan types that were only available to subpopulations such as dual-eligible plans, Program of All-Inclusive Care for the Elderly plans, part B-only plans, employer-sponsored plans, or special needs plans. We also excluded MA plans that had missing star rating information.

Our primary outcome measure was defined by counting the raw number of 5-star plans, plans with 4.5 or higher stars, and plans with 4 or higher stars. We also counted the number of all MA plans at the market level as an outcome measure to explore private insurers’ market entry and participation decisions. If private insurers avoid offering any MA plans in highly disadvantaged areas, SVI scores will be negatively associated with the number of MA plans regardless of quality. On the other hand, if MA plans in highly vulnerable markets are less likely to have high star ratings, we will find a negative association of SVI with the number of high-quality MA plans, while the association of SVI scores and the total number of MA plans (regardless of star rating) may not be significant. Each of these hypotheses would yield different policy implications.

We used SVI overall composite scores at the market level as our main exposure variable with percentile rankings ranging from 0 (least vulnerability) to 1(highest vulnerability). Following previous studies, we categorized SVI scores into five quintile group categories for our main analysis (very low (SVI scores lower than 0.2), low (0.2 to <0.4), moderate (0.4 to <0.6), high (0.6 to <0.8), very high vulnerability (0.8 or higher)).16,19

We conducted a descriptive analysis of outcomes and independent variables across markets and by SVI quintile group. We reported unweighted statistics by market population to ensure that more populated markets were not overrepresented. Then we specified multivariate linear regression models to estimate the association of market-level SVI scores and availability of high-quality MA plans. We also performed a subgroup analysis by census region and by rurality status to see if local SVI associations with the availability of high-rated MA plans were stronger in certain regions of the USA. For our subgroup analysis, we grouped markets into SVI quintiles separately for each region.

To account for sociodemographic and healthcare resource factors, which are not included in the SVI measure and might be correlated with geographic variations in insurance markets, we used rurality indicators from the Rural-Urban Commuting Area Codes and health care resource information from the Area Health Resource File. In our analysis, we controlled for rurality status (i.e., rural, micropolitan, or metropolitan), total population, number of active physicians per 100,000 population, number of hospital beds per 1000 population, and the estimated share of households without internet access. We also controlled for state fixed effects to account for region-specific policies that can be associated with market vulnerability and our outcome measures. We used robust standard errors clustered at the state level.

We also conducted sensitivity analyses, first, by running the main regression model with a continuous SVI percentile ranking (0 to 1) measure instead of using quintile SVI measures. This step allowed us to show if the number of high-rated MA plans changes as social vulnerability percentile ranking moves from the least vulnerable (0) to the most vulnerable markets (1). Second, we performed analyses using four subcategories (socioeconomics, household composition, racial/ethnic minority, and housing/transportation factors) of the SVI measure instead of the overall composite score to explore which SVI subcategory has stronger associations with the availability of high-rated MA plans. The overall SVI score was measured with 15 variables from the four themes, and each subcategory has separate market-level vulnerability percentile rankings.

This study was deemed exempt from the Institutional Review Board at George Mason University since it used publicly available market-level datasets. We used a significance level of 5% for our analysis and presented mean values with 95% confidence intervals. We performed statistical analysis in Stata, version 18.