3 ms·

If we optimistically assume the claimed 0.5% false positive rate is accurate, and all of the US got tested annually, that's 1.64M false positives per year. Canc

If we optimistically assume the claimed 0.5% false positive rate is accurate, and all of the US got tested annually, that's 1.64M false positives per year. Cancer.gov is telling me that approximately 1.8M USians were expected to get cancer last year. That's a positive predictive value of 52%. That still seems highly informative, to me; much higher than the PPV for mammography according to this (admittedly old) study: [1]. Assuming the 0.5% FP rate holds (again, I know that's optimistic), would you still regard universal testing with this method to be harmful? [1] https://jamanetwork.com/journals/jama/article-abstract/409356 https://jamanetwork.com/journals/jama/article-abstract/40935...

TaupeRanger 5y ago
Good question. The proof is in the pudding. Run a well powered RCT and if the intervention helps patients live longer and/or better, then we should consider using it! It will be hard to design a study that includes all possible detected cancers, but there are ways around this - perhaps we begin by studying those cancers that are most common and lethal, and go from there.
amelius 5y ago
Isn't this exactly why the headline reads "blood test good enough to be rolled out"? I.e., higher FP rate, and it wouldn't be rolled out.
- AlexCoventry 5y ago
  IIRC, the actual article says they're going to do a large trial, with 150k people, and someone in the article expresses skepticism that the estimated FP rate of 0.5% is accurate, because so far it was tested in people where there's some evidence of cancer.
dan-robertson 5y ago
Is that actually good though? It feels to me that there’s a big difference between explaining the implications of a positive result to someone who understands bayes theorem and someone who doesn’t. Another aspect of cancer screening is that detecting cancer earlier can improve key statistics like 5 year survival without affecting the actual disease in any way. Which can make screening sound more effective than it is.

TaupeRanger 5y ago

Good question. The proof is in the pudding. Run a well powered RCT and if the intervention helps patients live longer and/or better, then we should consider using it! It will be hard to design a study that includes all possible detected cancers, but there are ways around this - perhaps we begin by studying those cancers that are most common and lethal, and go from there.

amelius 5y ago

Isn't this exactly why the headline reads "blood test good enough to be rolled out"? I.e., higher FP rate, and it wouldn't be rolled out.

AlexCoventry 5y ago
IIRC, the actual article says they're going to do a large trial, with 150k people, and someone in the article expresses skepticism that the estimated FP rate of 0.5% is accurate, because so far it was tested in people where there's some evidence of cancer.

dan-robertson 5y ago

Is that actually good though? It feels to me that there’s a big difference between explaining the implications of a positive result to someone who understands bayes theorem and someone who doesn’t. Another aspect of cancer screening is that detecting cancer earlier can improve key statistics like 5 year survival without affecting the actual disease in any way. Which can make screening sound more effective than it is.