From Association to Causation
Observational studies document what naturally occurs in populations; interventional studies manipulate variables and measure consequences. Observational studies come in multiple types. Prospective cohort studies recruit disease-free subjects, characterise exposures, and follow them forward, documenting disease development. Cohort studies provide temporal clarity—exposure precedes disease, supporting causal direction. Yet subjects self-select exposures, and unmeasured variables may explain associations.
Retrospective case-control studies identify disease cases, identify matched controls, and retrospectively assess exposures. Case-control studies efficiently study rare diseases but rely on memory-based assessment vulnerable to recall bias.
Three major confounding mechanisms plague observational studies. Confounding occurs when unmeasured variables influence both exposure and disease. Reverse causation occurs when disease causes the measured exposure. Healthy user bias reflects that health-conscious individuals adopt multiple healthful behaviours simultaneously, confounding intervention effects with overall health-consciousness.
Bradford Hill criteria provide a framework evaluating whether associations are causal. These include: temporal sequence, dose-response, consistency, plausibility, and strength. Even meeting multiple criteria remains insufficient for causal proof—randomised trials provide stronger evidence.