Probability matching of medical events

Article type: Research Article

Authors: Winglee, Marianne^a | Valliant, Richard^a | Brick, J. Michael^a | Machlin, Steven^b

Affiliations: [a] Westat Inc., 1650 Research Boulevard, Rockville, MD 20850, USA | [b] Center for Cost and Financing Studies, Agency for Healthcare Research and Quality, 2101 E. Jefferson Street, Suite 500 Rockville, MD 20852, USA

Abstract: This paper discusses the methods of probability record linkage and error estimation for the Medical Expenditure Panel Survey (MEPS). MEPS collects medical expenditure data from both household respondents and their medical providers. The medical events reported by the two sources are subject to reporting differences and probability linkage methods are used to determine if pairs of medical events belong to the same entities. Match weights are assigned to pairs of events based on the likelihood of being a match, and a decision made to declare pairs as matches or nonmatches. The linkage errors include false matches (false positive) when pairs selected are not true matches, and false nonmatches (false negative) when true pairs are omitted. This study used three approaches: manual reviews, cumulative weight curves, and simulation approaches to provide estimates of linkage errors.

DOI: 10.3233/JEM-2000-0176

Journal: Journal of Economic and Social Measurement, vol. 26, no. 2, pp. 129-140, 2000