Evidence-based medicine was defined by the McMaster group as the integration of the best current evidence with clinical expertise and patient values, practised in steps: ask an answerable question, search, appraise, apply and evaluate [1]. Appraisal starts with design. Well-designed and properly executed randomised trials are regarded as the most reliable evidence on the benefits of a treatment [2], because randomisation balances the known and unknown differences between groups that confound observational comparisons.
The history of medicine has examples. Observational studies suggested that vitamin E might prevent cardiovascular disease and cancer, yet a meta-analysis of randomised trials raised the possibility that high doses increase mortality [3]. Observational studies supported the benefits of a Mediterranean diet, but the randomised evidence is more limited [4]. Statin product labels list adverse effects drawn largely from non-randomised, unblinded studies, which a meta-analysis of double-blind trials set out to test [5]. Some reviews compare randomised and observational estimates for the same question directly [6].