Abstract

Confounding is a major issue in observational studies, and it is a bias that needs to be corrected, controlled or adjusted for whatever your research question is. In this lecture we will review the definition of confounding and ways to tackle it. We will discuss two specific biases related to confounding in drug studies: confounding by indication and channelling bias. In addition, we will explain how to handle confounding in the different phases of research: 1) design phase, with an overview, among other issues, of what causal diagrams (directed acyclic graphs, DAG) are; 2) data collection and mining; 3) and finally analysis and reporting. In this latter part, we will explain how confounding is explored in datasets and the difference between confounding and modification. We will end with an overview of propensity scores, as a ways to tackle confusion in an efficient way, and a list of recommendations.