Abstract
OBJECTIVE: To assess the quality and completeness of death records in Medicaid administrative data by comparing them with the U.S. Census Bureau Numerical Identification (Numident) file, a comprehensive and up-to-date database with mortality information. STUDY SETTING AND DESIGN: We conduct a cross-comparison between Medicaid administrative records and the Census Numident file to identify the completeness of Medicaid death records and the misclassification of deaths. The analysis is stratified by state, year, and beneficiary characteristics. DATA SOURCES AND ANALYTIC SAMPLE: We use the Transformed Medicaid Statistical Information System Analytic Files (TAF) Demographic and Eligibility File linked with the Census Numident file. Our base sample contains individuals enrolled in Medicaid/Children's Health Insurance Program (CHIP) and recorded as deceased in either the TAF or the Numident file from 2016 through 2022. PRINCIPAL FINDINGS: We find that almost all deaths reported in the TAF are found in the Numident. On the other hand, 13.2% of deaths are reported in the Numident file but are missing in the TAF. This share varies considerably across states; indeed, many states have few missing death records and thus a high degree of completeness, while some are missing over half of deaths. Furthermore, this share varies by beneficiary characteristics, especially age. Conditional on a TAF-reported death, the TAF death dates match with the Numident death dates in almost all cases. CONCLUSIONS: Medicaid administrative data include death information, but these records underreport deaths. Deaths that are recorded in TAF data are largely accurate.