Loading...
Mira is a 34-year-old software engineer who just completed her annual health screening at MedTech Labs. Three days later, her doctor calls with unsettling news: she tested positive for Harmon's Syndrome, a rare autoimmune condition. The test, her doctor explains, is 99% accurate — it correctly identifies 99% of people who have the disease, and correctly clears 99% of people who don't. Mira spirals. She cancels a vacation, tells her sister, and starts researching treatment options. The number '99% accurate' echoes in her mind. How could she be in the unlucky 1%? But here's what Mira doesn't consider: Harmon's Syndrome affects just 1 in 10,000 people. Her doctor ordered the test as part of a routine panel, not because of symptoms. Imagine testing all 10,000 people in Mira's town. On avera...
Popular framing: A 99% accurate positive test means there is a 99% chance you have the disease — a positive result is near-conclusive evidence of diagnosis. The 'test accuracy' narrative is a structural lie; a test that produces more false positives than true positives for a specific person is 0% 'accurate' for that person.
Structural analysis: Test accuracy (sensitivity/specificity) and diagnostic probability (positive predictive value) are entirely different quantities linked by Bayes' theorem. When base rates are low, a high-accuracy test operating at scale will produce a false-positive pool that dwarfs true positives. The positive predictive value — the only number that answers 'do I have this disease?' — was never communicated to Mira because the system isn't designed to surface it. The 'base rate neglect' frame is excellent but misses the 'Anchoring' effect that makes the neglect so difficult to overcome even with explanation.
The gap matters because it causes systematic, predictable harm: unnecessary psychological distress, financial cost, and downstream medical interventions for patients who are overwhelmingly healthy. It also undermines trust in medicine when the 'error' is eventually revealed. Closing the gap requires not individual Bayesian literacy but redesigned result communication that embeds prior probability into every positive result disclosure.
