Causal Inference in Observational Epidemiology

About

Causal inference from observational data is a key task of epidemiology and of allied disciplines such as behavioural sciences and health services research. Commonly used statistical methods estimate association measures which cannot always be causally interpreted, even when all potential confounders are included in the analysis. In contrast, a causally explicit approach formally defines causal effects, identifies the conditions required to estimate causal effects without bias, and uses analytical methods that, under those conditions, provides estimates that can be endowed with a causal interpretation. This course presents such framework for causal inference from observational data and recent methodological developments, with a special emphasis on complex longitudinal data. The application of these methods will be illustrated using data from a synthetic HIV cohort study. The course is aimed at epidemiologists, statisticians, and other researchers who work with longitudinal observational data.

The course will run over three days and consist of lectures, group work and computer practical sessions. We start early in the morning with a review of the previous day. During the extended break in the afternoon participants review course materials, catch up on emails or go skiing. We reconvene at 4:30 pm for the computer sessions.

Course leader

Prof. Miguel Hernan; Harvard T.H. Chan School of Public Health, Boston, USA Prof. Marcel Zwahlen; Institute of Social and Preventive Medicine (ISPM), University of Bern, Switzerland

Target group

The course is aimed at epidemiologists, statisticians, and other researchers who work with longitudinal observational data. Stata® will be used for the computer practical sessions, and so familiarity with the package is desirable, although code and solutions will be provided.

Course aim

By the end of this short course participants will have
• An in-depth understanding of confounding and selection bias
• An understanding of the role and potential of different methodological approaches to overcome these problems, including inverse probability weighting, marginal structural models and nested structural models
• Done practical data analyses implementing these methods using Stata®software.

Fee info