[SpicyLemonZest]

Their infographics convince me that they understand the statistics. But one of the key issues here is that the statistics are radically counterintuitive in a way that most people don't understand - the patients, the testing companies, and even some medical staff all incorrectly believe that a positive test for a rare condition means you probably have the condition.

[halpert]

Their graphics say the tests are “84% wrong.” Do you really feel that’s an accurate description? That doesn’t feel like an accurate description to me, and their usage of “wrong” in this context highlights that they don’t understand the distinction and importance of true positives, false positives, true negatives, and false negatives when measuring accuracy.

[isoprophlex]

Going through something like this is very VERY stressful. When you get a negative you immediately forget about it. When you get a positive you die inside. Speaking from experience here.

84% wrong sounds, to me, as an accurate description. Experiencing this from the inside out, only the false/true positive ratio matters. (Given sufficiently low false negative rates, of course)

84% of people whose world is turned upside down are actually getting a wrong diagnosis.

[andreilys]

You’re talking about precision (true positive / true positive + false negative) but that’s only one part of the story.

There is a real human cost to having a child born with a rare genetic disease (and I would argue is immensely more stressful). You can easily adjust the sensitivity to the test but at the cost of detecting actual true positive cases. The correct response to receiving a positive is to do another test to ensure it’s not a false positive.

To say 84% wrong is clickbait and used to elicit a legislative response (FDA regulation), which will help the reporters career.

The actual ratio to tell if something is “wrong” is accuracy (True positive + true negative) / (true positive + true negative + false positive + false negative)

[mnw21cam]

No, precision is true positive / (true positive + false positive).

Your first equation is sensitivity.

[fshbbdssbbgdd]

If you get a negative result and then your child is born with the condition, you won’t forget quickly either.

[SpicyLemonZest]

I really feel it's an accurate description. If you get a positive result on the test, there's a 16% chance your fetus has a 1p36 deletion and an 84% chance they don't.

[halpert]

As you said “if you get a positive result”. It’s true, if you ignore the 99.9% of the time the test is correct (true negative result), then you can say the test is 84% wrong.

[SpicyLemonZest]

84% of people who got a positive test result will end up telling their family "it's OK, the first test was wrong, my baby doesn't have a 1p36 deletion after all". The 99.9% of other people who got true negatives are important from a test design perspective, because specificity is closer to the actual levers you can pull on, but it's not super relevant to the decisionmaking process of someone who gets a positive result.

[andreilys]

Ignoring all the true and false negatives which themselves are markers of how accurate the test is.

16% precision is the correct statement, saying the test is wrong 84% of the time implies that those getting negative results might actually have positive results.

[robbedpeter]

He framed his statement correctly, limiting his observation to the condition that the test returned a positive result. Saying that 84% of positive results are false is correct if only 16% are true. You'd need to know false negative rates and base occurrence rates (modified by whatever other factors are unique to your situation) to inform the nature of information you get by performing the test.