St. James Ballroom
A Practical Guide of Propensity Score Analysis for Longitudinal Observational Study (303862)
Konstantinos Spaniolas, Stony Brook University Medical Center
Song Wu, Stony Brook University
Jie Yang, Stony Brook University
*Chencan Zhu, Stony Brook University
Keywords: propensity score, longitudinal observational study, stratification, matching, regression adjustment, IPTW, time-varying covariates, paired data
Propensity score methods are being increasingly used to reduce treatment-selection bias in observational studies. When data are longitudinal, the estimation and use of propensity scores may become a challenge. The most common approach is to match at baseline, and then apply longitudinal data analysis. Other popular propensity score analysis methods include stratification, regression adjustment, and inverse probability of treatment weighting (IPTW). Moreover, when time-varying covariates exist, methods have to be readapted. Therefore, this study aims to provide a practical guide on how to utilize propensity score in a longitudinal setting. Different types of longitudinal outcomes like binary, count and continuous are considered. Issues such as whether the paired nature of data need to be accounted for in analysis after propensity score matching are also discussed. A real data example to compare yearly post-operative hospital use and cumulative total cost between sleeve and bypass patients is shown for illustration.