Can you confirm it is acceptable to calculate PPV for each level of a multi-level test (as long as the sampling is such that it is valid to calculate PPV)?
I can't find an example in the text, but please point me to one if I have missed it.
Thanks!
As long as sampling is cross-sectional and the overall proportion with disease (D+) reflects the pre-test probability P(D+) in the clinical population in which the test will be performed, then the predictive value of a test result P(D+|r) is valid, regardless of whether the test result is one of only two possible test results or one of several possible test results. The terms "positive predictive value" and "negative predictive value" imply that the test has only two possible results, so avoid those terms for multilevel tests.
I do not see an example in Chapter 3 on multi-level tests. Of course, you could calculate the LR(r) for the result in question , convert pre-test probability to pre-test odds, multiply by LR(r), and convert back to post-test probability, but if the pre-test probability P(D+) is the same as the proportion with the disease in the study, then it is easier just to divide the number of D+ subjects with result r by the total number of subjects with result r. This is the multi-level equivalent of the 2x2 table method outlined in Chapter 2, page 8.