SOME METHODS FOR CAUSAL INFERENCE, WITH APPLICATION TO AN OBSERVATIONAL EPILEPSY DATA SET

Abstract

Causal inference attempts to attribute a causal mechanism to a treatment in an observational study. Attributing cause is a major focus of research in bio-statistics and application to observational biomedical studies. There have been a number of different proposals as to how causal inference should be carried out. In this thesis, two methods - propensity score adjustment and graphical causal modeling - are explored through application to an observational data set. The data set concerns several hundred pediatric patients with epilepsy collected over thirty plus years at the IWK hospital. The primary goal of the thesis is to identify which factors are important in determining whether patients remain on anti-epileptic medication at the end of follow up. The basic non-causal model - logistic regression, with or without stepwise selection - identi fies a number of signi ficant predictors in addition to the nominal treatment variable, which is the indicator of normal neurological status.