Statistical analysis of observational study

Abstract

In any observational study where an exposure of interest, such as pharmaceutical treatment, is not randomly assigned to subjects, a bias is often introduced in estimating the true effect of treatment. This is because patients who receive treatment usually have more severe medical conditions than those without treatment. Failure to control such inherent bias in patient characteristics when assessing for the true effect of treatment across comparison groups, may lead to confounding. The presence of confounding makes it difficult to evaluate the true treatment effect. In this paper, we will introduce statistical strategies which aim to remove the effects of confounding in observational studies.