How can multiple binary diagnostic tests be best integrated to form an overall diagnosis? I have three diagnostics tests. Each test indicates classifies an individual as either 0=ill or 1=non-sick. I assume these test are not independent.
How can multiple binary diagnostic tests be best integrated to form an overall diagnosis?
 A: Here's a few quick thoughts, based on a general understanding of statistics. I'm not sure what's the convention in the diagnostics literature:
If you have a gold-standard indicator of whether a person is sick or not, you could perform a logistic regression predicting whether a person is sick or not from the three diagnostic tests. You could then do a sensitivity and specificity analysis to decide which combinations of the tests were sufficient for you to make a given diagnosis.
If the component tests were originally based on continuous variables, then I'd generally use those continuous variables as predictors in the logistic regression equation. This would generally be a more powerful approach. For example, someone who is just below threshold on all three indicators may be highly likely to be sick, whereas someone is very low on two but about threshold on one, may be less likely to be sick.
If you don't have validation data, then I imagine you'd have to fall back on expert judgement above the relative efficacy of different diagnostic tests. 
