Or only consider it a positive once the confidence is high enough that false positives are not a problem anymore.

Getting a test good enough to still make it useful (detect enough of the true positives) would of course be a challenge, but the more data is available, the more feasible that might be.