Positive predictive and negative predictive value are very important concepts in any screening test. I'll quickly break it down for everyone.

/1 https://twitter.com/choo_ek/status/1260603128396697600

Let's say that a test is 90% specific (10% false positives).

Now let's say that the problem exists in 1% of the population.

If we screen everyone, we will pick up the 1% that have it, but with 10% false positives.

In other words, a + test is 10x more likely false than true.

The actual math (adjusting for the 1% prevalence) is:

Chance of a positive test: 10.8%
Chance of a negative test: 89.2%

Positive predictive value (chance the positive test is actually positive): 8.3%

We always talk about the sensitivity (false negatives rate) and the specificity (false positive rate). And yet, in real world scenarios, these values are not helpful even at 90% for rare conditions/issues.

For example, in my area, suicidology, the rate in America is 0.014% per year. Even if we had a test that was 90% specific and sensitive (only 10% false positive/negative) for detecting "will die by suicide in a year), a positive screen would be right 0.12% of the time.

So this amazing, sensitive and specific test for suicide would be wrong 794 times for every time it was right.