Abstract:
|
We describe an application of the MBISG 2.0 algorithm to impute race/ethnicity/language (R/E/L) probabilities for 17,517,852 Medicare beneficiaries using surname, address, and administrative information. We used multinomial logistic regression and estimate rates of 2015 voluntary disenrollment for each of 7 mutually exclusive R/E/L categories: English-preferring Hispanic, Spanish-preferring Hispanic, White, Black, Asian or Pacific Islander, American Indian or Alaska Native, and multiracial. We predicted person-level disenrollment indicators from predicted probabilities of R/E/L and indicators of gender, disability status, and low-income status in regression models with and without indicators for health plan, to estimate overall and within-plan differences. We found that R/E/L minorities disenroll from plans at higher rates than non-Hispanic Whites, with only a small portion of the difference attributable to higher enrollment of R/E/L minorities in high-disenrollment plans.
|