Keywords: Record Linkage, Causal Inference, Missing Data, Bayesian
Meals on Wheels (MOW) provides a critical home meal delivery service to vulnerable older adults across the nation. Though home-delivered meals have been funded and operational for over 40 years, little is known about the health status and the healthcare utilization of clients who benefit from these programs. Assessment of the impact of the program on the healthcare utilization of its clients first requires linkage of client information to Medicare Claims Data. We explore a Bayesian approach to record linkage, which incorporates information exclusive to one file into the linkage likelihood, to create multiple datasets where MOW individuals are linked to Medicare recipients. For each potential link configuration, MOW recipients are matched to Medicare enrollees who did not benefit from MOW. The potential outcomes for MOW recipients had they not been enrolled in the program are multiply imputed using the matched Medicare control group. The resulting treatment effects properly account for the error encountered in both the linking and matching process.