> So that could artificially make it look like we've gotten better at treating cancer, when actually we've just gotten better at detecting non-life-threatening

> So that could artificially make it look like we've gotten better at treating cancer, when actually we've just gotten better at detecting non-life-threatening cancer. No, because the death rate is measured at the population level, not at the incidence level. Fewer people are dying today of cancer than 30 years ago. The only way that detection bias could impact that outcome is if you make the argument that a large number of terminal cancer cases were going undetected in 1991 at the time of death.