230 – Applied Causal Inference
Propensity Score Analysis of Data from Case-Control Studies
Irina Bondarenko
University of Michigan
Trivellore Raghunathan
University of Michigan
Propensity Scores(PS) methods are useful to adjust for many covariates to account for the lack of randomization in many prospective or cohort observational studies. However, retrospective or case-control study designs are also used in public health research. The case and control sampling strategies provide unique challenges in developing propensity score analysis techniques to adjust for large number of covariates. We develop and evaluate a method that uses counter factual probability of exposure under the control conditions for cases and the propensity score to account for imbalances between the cases and controls in the covariates due to sampling to create matched sets of cases and controls and then perform matched analysis. We show how counter factual probability of exposure can be used to graphically diagnose confounding. The method is illustrated on the Pneumococcal Vaccine study to estimate effectiveness of vaccine in a presence of confounding. We also evaluate the repeated sampling properties of the significant tests and the associated point and interval estimates of the odds ratio, a common measure used in retrospective studies.