Mette Nørgaard (University of Aarhus, Denmark): The role of clinical epidemiology – why medical practice cannot be based only on results from experimental studies
When an exposure is allocated randomly, as in randomised controlled trials (RCTs), any association between a given prognostic variable and the exposure will be random. Accordingly, if the trial is adequately powered and well designed, randomisation will, on average, control both known (measured and unmeasured) and unmeasured confounders. RCTs, however, often have narrow inclusion criteria and therefore tend to enrol a selection of patients with only one diagnosis, no concomitant therapies, neither very young, nor old, and with a reasonable prognosis. Accordingly, trials often focus on short‐term efficacy and safety in a controlled clinical environment among well‐educated affluent patients.
In contrast, large population-based health care databases reflect the entire daily clinical practice for large and representative populations. Studies conducted from existing databases thus offer an alternative to RCTs, making it possible to study rare exposures, diseases, and outcomes inexpensively and rapidly. In addition, many health effects first appear years after exposure. Existing databases are often the only feasible source with which to examine delayed health effects.
Database studies also have important limitations, which relate to data selection and quality. A constant challenge in observational designs is to rule out confounding. The value of large databases for a given study question thus depends on completeness and validity of the information on confounding factors.
Causal inference is a core task of science and although observational studies cannot definitely prove causation, Miguel Hernán and others currently argue, that we should be more explicit about the goal of our observational studies. Are we aiming for prediction or causal inference? The answer to this central question should guide our choice of study design, the choice of potentially confounding variables, and the sensitivity analyses that we should include to improve the quality of our observational research.