5 ms·
Understanding medical tests: sensitivity, specificity, predictive value (2018)
- avalys 7y agoThis is not just medical tests - you use the same criteria to evaluate any binary classifier. “Precision and recall” are positive predictive value and sensitivity, respectively.
- ageitgey 7y agoThis site blocks visitors from the UK.
- neonate 7y agohttps://archive.md/Glbh8 https://archive.md/Glbh8
- SubiculumCode 7y agoHits, misses, correct rejections, false alarms, in memory research lingo. Can I get a d-prime?
- kryogen1c 7y agosaying a test has 85% accuracy tells you nothing about the results of that test. not being hyperbolic - literally nothing. simple proof: i develop a test for disease X and give it to 1 billion walking around grocery stores in the whole world. it is 85% accurate. results are 200 million people are positive? how many of those people actually have disease X? do you have a guess? disease X i was testing for was death. every single positive test was wrong. how close was your guess? you must, at a minimum, know test accuracy _as well as_ disease prevalence to form a statistical guess. death is prevalent in 0% of alive people, so test accuracy is worthless.
- bugzz 7y agoAs long as we are fairly confident about the specificity and sensitivity of the test, then it DOES tell us something, even if those numbers are as low as 85%. Here's the flaw in your proof - If the test has 85% specificity, then the chance of 200 million positives is essentially 0% (you'd get ~150 million if no one really has the disease). Getting a result of 200 million positive means either you DO have a significant number of true positives (50 million), or your 85% number for specificity was wrong.
- raverbashing 7y agoThanks for showing why Bayesian reasoning is superior to frequentist thinking. (Yes, adding the disease prevalence % is essentially bayesian, but testing everybody isn't) Even better than knowing the disease prevalence is knowing possible proxies (again, Bayesian). Your Covid-19 test might have a low sensitivity but if the person has a temperature and other symptoms you might want to redo the test later even if it comes out negative.
- jnbiche 7y agoPositive predictive value, which is discussed in the article, is derived from Bayes theorem. So they do discuss that.
- lean_industries 7y agoThis is amazing +1
- mlthoughts2018 7y agoUsing my VPN, this site says access is not allowed. It seems very scammy / low-quality as a result.
- matheusmoreira 7y agoWikipedia article on the subject: https://en.wikipedia.org/wiki/Sensitivity_and_specificity https://en.wikipedia.org/wiki/Sensitivity_and_specificity