Exact Matching as an Alternative to Propensity Score Matching

The comparison of different medical treatments from observational studies or across different clinical studies is often biased by confounding factors such as systematic differences in patient demographics or in the inclusion criteria for the trials. Propensity score matching is a popular method to adjust for such confounding. It compares weighted averages of patient responses. The weights are calculated from logistic regression models with the intention to reduce differences between the confounders in the treatment groups. However, the groups are only “roughly matched” with no generally accepted principle to determine when a match is “good enough”. In this article, we propose an alternative approach to the matching problem by considering it as a constrained optimization problem. We investigate the conditions for exact matching in the sense that the average values of confounders are identical in the treatment groups after matching. Our approach is similar to the matching-adjusted indirect comparison approach by Signorovitch et al. but with two major differences: First, we do not impose any specific functional form on the matching weights; second, the proposed approach can be applied to individual patient data from several treatment groups as well as to a mix of individual patient and aggregated data.