Matching by Propensity Score in Cohort Studies with Three Treatment Groups

• DOI: 10.1097/ede.0b013e318289dedf
• Publication Date: 2013-03-26T12:54:53Z
• Authors: Jeremy A. Rassen; Abhi A. Shelat; Jessica M. Franklin; Robert J. Glynn; Daniel H. Solomon; Sebastian Schneew...
• Keywords: Comparative effectiveness research

Nonrandomized pharmacoepidemiology generally compares one medication with another. For many conditions, clinicians can benefit from comparing the safety and effectiveness of three or more appropriate treatment options. We sought to compare groups simultaneously by creating 1:1:1 propensity score-matched cohorts.We developed a technique that estimates generalized scores then creates matched sets. compared this methodology two existing approaches-construction cohorts through common-referent group pairwise match for each possible contrast. In simulation, we varied unmeasured confounding, presence effect heterogeneity, prevalence treatments method's bias, variance, mean squared error (MSE) effect. applied these techniques cohort rheumatoid arthritis patients treated nonselective nonsteroidal anti-inflammatory drugs, COX-2 selective inhibitors, opioids.We performed 1000 simulation runs. base case, observed an average bias 0.4% (MSE × 100 = 0.2) in three-way matching approach 0.3% technique. The showed differing MSE increasing heterogeneity decreasing score overlap. With highly unequal exposure prevalences, strong low overlap, 6.5% 10.8) 12.5% 12.3) approach. empirical study displayed better covariate balance using Point were substantially similar.Our offers effective way recommend its use over approaches.