Analysis, Validity and Inference of Observational Epidemiology, 2 op

This intensive course addresses valid analysis and inference on interesting causal effects of potential risk factors based on data from observational studies. The effects of interest are defined in terms of carefully specified causal estimands, such as causal risk difference and causal risk ratio, or other contrasts of pertinent counterfactual quantities. The strategy of analysis utilizes extensively the modern framework of causal diagrams, i.e. directed acyclic graphs (DAG), whose basic concepts and principles are covered. These diagrams provide insight into some design decisions, too, like matching in cohort and case-control studies. The major types of systematic error that commonly threaten the validity of causal analysis, i.e. confounding, selection bias and information bias, are also characterized using causal diagrams. A few practical analytical approaches for dealing with the biases are introduced. Primary attention is devoted to the principles of controlling confounding using the causal diagram and statistical approaches like g-computation and inverse probability weighting. Other specific topics will include the limitations of conventional null-hypothesis significance testing in observational epidemiology, target trial emulation, as well as modern methods to deal with selection and information biases. Examples using R environment will accompany some of the lectures and practical sessions, but deep familiarity with R programming is not assumed. Students are expected to have a solid foundation in epidemiologic study design and biostatistics, including the interpretation of linear, logistic, and proportional hazards regression models. They will also be asked to read two to three papers each evening in preparation for the next day’s class. Lecture content will be reinforced by afternoon practical sessions, in which students can apply the principles and methods learned in lecture to the analysis and interpretation of both real studies and simulated data sets.

Requirements: Participation in lectures, laboratories, and discussions. Course readings. Instruction will be in English.

Target group: PhD students and researchers in health sciences.

Attendance: This seminar is open for PhD students of University of Tampere as well as staff at the University, Tampere University Hospital and research institutions. Limited to max. of 20 participants.