Sensitivity, Specificity, Positive Predictive Value, Negative Predictive Value
These terms provide the basis with which we evaluate all tests in medicine, however they are all often misinterpreted or confused for each other.
Last night at Journal Club Ben Hunter gave a great synopsis and I have recorded a recap of it here
All these values interpret the percentage of the time a test is correct in a certain population
Sens– % correct in population with the disease
Spec– % correct in population without disease
PPV — % correct in population with positive test
NPV — % correct in population with negative test
PPV sounds like a dream. Exactly how worthwhile is a positive test. But it is deceptive. PPV and NPV are highly dependent on prevalence. Here is an example.
Group A — Pts < 40 with chest pain. 1% prevalence of MI
Group B — Pts 65-75 with chest pain, SOB, nausea, diaphoresis, HTN, DM, ESRD. 50% prevalence of MI.
I am going to evaluate a test where I take blood from the patient, throw it in the trash, and say “Your test is Negative.” Lets see how it works in these patients
Everyone gets a negative test result. In 100 patients, 99 times this is correct so NPV is 99%
Everyone gets a negative test result. In 100 patients, 50 times this is correct so NPV is 50%
This shows how NPV is affected tremendously by prevalence.
The real power of a test is in LR, which is a combination of sensitivity and specificity. This is not affected by prevalence and is a whole other lecture.