Hybrid Matching in Observational Studies: Combining Simple Matching with Propensity Score-based Matching Methods

• Activity Number: 275
• Type: Contributed
• Date/Time: Tuesday, August 9, 2005 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Daniel Parks*+ and Kwan R. Lee and Bong Kim
• Companies: GlaxoSmithKline and GlaxoSmithKline and Newmann College
• Keywords: observational data ; simple matching ; propensity score matching ; hybrid matching

Abstract

When a case-control study is conducted using observational data, matching methods are required to eliminate bias between the case and control. While it has been shown that propensity score matching methods (PMMs) effectively eliminate bias and select a large number of matched pairs, many researchers still prefer the simple matching method (SMM). The reasons underlying such a decision revolve around the intuitive, ease of use associated with the SMM, the perfect elimination of bias, and the sometimes bothersome selection of undesirable matched pairs by the PPMs. While this phenomenon cannot be avoided, the SMM also has a large disadvantage: it is unable to properly treat continuous variables. Here, we combine the strengths of the SMM and PMMs to introduce the hybrid matching method (HMM). Through simulation studies, we determine how effective the HMM is in comparison to other methods by considering the number of observations and the degree of balancing between the case and control. In particular, we assess the effects on balancing upon using an artificial range for a continuous variable by the SMM. Finally, we use actual observational data to compare the three methods.