PPV=TP/TP+FP
This value takes into account everyone that tested positive (both true and false). Once you have a positive result, this value tells you the chances of the test actually being a true positive
Say you take a test with 95% SN, 90% SP and 10% prevalence
that means if you test 1000 people you will have
95 TP
5 FN
810 TN
90 FP
PPV=95/95+90=.51 or 51%
So even with a reasonably good test, there is 50% chance that a positive result is actually a false positive if the prevalence of the disease is not very high. Which is another important point, PPV takes into account the prevalence of the disease! SN and SP are more indicative of how good the test in itself really is.
NPV works on the same principle, except that it is the probability that once you get a negative test result, that it will actually be a TN. in this case it would be almost 100%.
If you think about it, you can actually realize why you need high SP to rule disease IN and high SN to rule disease OUT, instead of memorizing SPIN and SNOUT
Yeah, in the case that a higher value means a positive result. So if you call a hematocrit of >45% polycythemia vera, you're gonna catch almost all the polycythemia cases (high sensitivity). However, you also have tons of normal people in there, so having a "positive" on this test doesn't mean much (poor PPV). On the other hand, if you have a crit of less than 45%, you almost certainly don't have Pvera (high NPV).
If we now change the criteria to >70% means Pvera, you're not gonna catch all the cases, but the cases you catch are almost certainly actually Pvera (high specificity). If you come up positive on this test, it's very likely that you have Pvera (high PPV). If you come up negative (but get a 60%, for instance), you still might have Pvera that hasn't been caught (poor NPV).
Lastly, remember that population dynamics change the PPV and NPV, but not sensitivity/specificity. If a test is 80% sensitive, 20 percent of people with disease get missed. However, if everyone in the population has the disease, 20% of them will still get missed (the sensitivity/specificity hasn't changed), but getting a negative result doesn't mean squat (very poor NPV).
This site uses cookies to help personalize content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies and terms of service.