Causation, bias and confounding: a hitchhiker's guide to the epidemiological galaxy

摘要

The focus in this review is on exposed and non-exposed, and diseased and non-diseased, individuals ('categorical variables'). The statistical assessment of measurements made on a continuous scale ('continuous variables') (e.g. mean height or weight) is not considered.Confidence in causality is strengthened if an association is stable: that is, if the numbers (the numerators and denominators of the compared rates) are sufficient to rule out chance ('sampling error') with reasonable confidence. By convention, 'confidence' is estimated by setting a confidence interval (CI) ('confidence limits') around the relative risk (RR) estimate ('point estimate').