Abstract
Objective: To discuss and compare kernel-based matching with one-to-one propensity score matching applied to disease management. Data sources: Administrative claims data from a US Medicaid fee for service plan. Study design: Matched two group analyses using both kernel-based matching and one-to-one propensity score matching. This comparison is applied to the estimation of diabetes disease management treatment effects. Principle findings: Kernel-based matching is found to be better than one-to-one propensity score matching when there is no sufficient number of potential controls from which to draw a matched cohort but similar when there is a sufficient number of potential controls. Matching was applied in the context of a diabetes disease management program that showed an increase in management of each person's medical care through the disease management program. Conclusions: The approach provides a methodology for researchers to evaluate healthcare service innovations without a randomized trial design and delineates the requirements for a matched analysis. Matching was applied in the context of a disease management program showing better patient management through the disease management program.