I think this article misses that there is a statistical conditioning issue. If the research literature says that patients who get scans showing lumps in their

I think this article misses that there is a statistical conditioning issue. If the research literature says that patients who get scans showing lumps in their lungs have X% chance of having lung cancer, there is an implicit conditioning there in the patient having complained to the doctor of some kind of symptom that prompted the scan. X% isn't the probability of having cancer given lump on scan, it's the probability of having cancer given a lump on a scan in patients have breathing trouble. If you test too much you run the risk of there not being data about what the baseline is! Studies for a lung cancer treatment usually take 1000 random people and give them all lung x-rays and then wait to see how many get cancer later and survive the experimental treatment, instead they usually find 1000 people already diagnosed with lung cancer that had it diagnosed by traditional means (e.g. scans prompted by breathing problems). The same reasoning applies to deciding whether to do a biopsy -- most biopsies are going to be prompted by scans from a relevant patient complaint, so statistics computed to determine the risk/benefit tradeoff are implicitly conditioned on that. You need a separate study to determine the risk/benefit for incidental discovery.