Hacker News Reader

> Even a test that is 99% accurate in a high-incidence population is going to be utterly ineffective when applied to the general population.

I read that sentence a few times. 99% accurate seems quite high. WSJ recently opined about the 1-in-5 cancer misdiagnosis rate.

https://www.wsj.com/articles/are-you-sure-you-have-cancer-di...

"Are You Sure You Have Cancer?"

View on HN

The point is that a test that has a guarantee like "out of 100 people who test positive, only one will turn out to be negative" (99% specificity) can have an arbitrarily high false-positive rate when applied to the whole population: if a condition affects 1 in 10k individuals, then out of 10k individuals that take that test, 100 will be positive, but only 1 will be a true positive. The rarer the condition, the more specificity you need in your test to have a low false positive rate.

The WSJ article is behind a paywall so I can't double check this, but you can have a 99% accurate test with a 20% misdiagnosis rate. The difference is that 99% is per-person, and 20% is per-positive-result. A rare condition will have more false-positives relative to true-positives than the test accuracy would indicate.

It's a bit easier to understand if you think of the extreme case: if the disease is completely eliminated than all positives are false-positives.

Reply on news.ycombinator.com