Three reversals, one lesson: surrogate substitution, confounding, and spurious consistency in the evidence that misled medicine

Abstract

Background: Clinical practice is periodically built on evidence that proves not merely incomplete but directionally wrong. We ask whether such reversals share a common structure. Approach: We re-examine three landmark evidence reversals — the Cardiac Arrhythmia Suppression Trial (CAST, 1989), the Women's Health Initiative oestrogen-plus-progestin trial (WHI, 2002), and the albumin controversy (Cochrane meta-analysis 1998 versus the SAFE trial, 2004) — recomputing each headline effect from the published counts and reading them as three instances of one phenomenon: truth recovery, in which a definitive randomised trial overturns an established belief. Findings: In each case the prevailing belief was confidently held and wrong in direction. CAST: antiarrhythmics that suppressed the surrogate (ventricular ectopy) increased all-cause death or cardiac arrest, 56/730 versus 22/725, relative risk 2.5 (95% CI 1.6-4.5). WHI: hormone therapy, judged cardioprotective across four decades of observational data (pooled observational RR approximately 0.50), instead raised coronary heart disease (HR 1.29), stroke (1.41) and breast cancer (1.26). Albumin: a pooled relative risk of death of 1.68 (1.26-2.23) across 30 small trials collapsed to 0.99 (0.91-1.09) in the 6,997-patient SAFE trial. The three errors have distinct mechanisms: surrogate substitution, confounding by indication, and spurious consistency among underpowered trials sharing biases. Conclusion: The reversals are not anomalies but a taxonomy. They argue for weighting evidence by its capacity to be wrong — privileging endpoints that matter, designs that break confounding, and adequate power over mere agreement — and for treating consistency in synthesis as something shared bias can manufacture, not a guarantee of truth.