90% of all claims about the problems with medical studies are wrong

A careful corrective to the viral 'Ioannidis proved 90% of medical research is false' meme. Scott traces the 90% figure to an unsourced Atlantic paraphrase distorted into a Time headline, grants the real base-rate/p-value problem (a 5% significance level has <5% predictive value when the prior is low — the dart-at-the-metabolic-chart picture), but then pushes back on four over-readings: (1) the 90% number is made up; (2) the studies people actually hear about sit near the top of the 'medical evidence pyramid' (big RCTs/meta-analyses, ~85% PPV), not the snake-infested bottom; (3) re-analyzes Ioannidis's famous '41% of influential studies contradicted' down to ~5% of large RCTs outright contradicted (the rest negative-results-excluded, exaggeration-not-falsity, or middle-of-the-pyramid epidemiology); (4) replication across multiple studies drives the posterior up fast, and doctors mostly believe well-replicated consensus, not single darts. Upshot: be skeptical of individual studies, appropriately trust large RCTs/good meta-analyses/consensus.

Quality 72 · Strong. Strong, upper. A lucid, numerate, much-needed corrective that defuses an over-cynical meme without denying the real problem; the evidence-pyramid framing, the base-rate×replication math, and the careful re-derivation of the 41%→~5% figure are all genuinely useful epistemic hygiene. Held below Excellent because the underlying tools (base rates, PPV, the pyramid) are established and the math is admittedly loose ('I'm even worse at explaining math than I am at doing it'); the value is the careful public correction, not a new idea.

Claude’s paradigm shift 48 · Moderate. Moderate (48). The fresh moves — tracing how the 90% meme propagated and re-framing the 41% statistic — were a useful corrective at publication, but the substantive ideas (base rates, the evidence pyramid, replication) were well-established.

Real-world impact 2 · Minor. Minor (2). Sharpens how readers treat medical-study reporting and gets cited in replication-crisis discourse, but no material-world/practice change traceable to the post.