Its interesting to see this valid argument raised against this use of AI to identify breast cancer. The lack of control groups is one of the more common concerns raised related to vaccines as well, the argument lands like a lead balloon there.
Subsequent trials generally compare against the best known current treatment as the control instead.
This study has no such concerns. It's ethical to include images of non-cancerous breast tissue. The things are not comparable.
Vaccine studies today almost always use a previously approved vaccine as the "control" group. That isn't a true control and if you walk back the chain of approvals you'd be hard pressed to find a starting point that did use proper control groups.
Anyway, my point here wasn't to directly debate vaccines themselves, only to point out that its interest to me as someone without a career in health to see the same effective argument used in two different scenarios with drastically different common responses.
1) a double blind RCT with a placebo control is a very good way to understand the effectiveness of a treatment.
2) it's not always ethical to do that, because if you have an effective treatment, you must use it.
Even without a placebo control you can still estimate both FN and FPs through careful study design, it's just harder and has more potential sources of error. A retrospective study is the usual approach. Here, the problem is they only included true positives in the retrospective study, so they missed the opportunity to measure false positives.
And the problem with -that- is that it's very easy to have zero false negatives if you always say " it's positive". Almost every diagnostic instrument has something we call a receiver operating curve that trades off false positives for false negatives by changing the sensitivity for where you decide something is a positive. By omitting the false negatives, they present a very incomplete picture of the diagnostic capabilities.
(In medicine you will often see the terms "sensitivity" and "selectivity" for how many TPs you detect and how many TNs you call negative. It's all part of the same type of characterization.)
We may broadly agree that submitting a control group to a placebo treatment for a particular disease is immoral, but that doesn't mean such a study isn't necessary to prove out the efficacy or safety of the treatment. As for modelling, for example trying to estimate FN and FP, it can only ever indicate correlation at best and will never indicate likely causation.
If you have a new vaccine for a disease for which there is no existing vaccine you do a standard placebo controlled RCT which gives you a direct, high quality measurement of efficacy and side effects.
Not just vaccines, in each study on the effectiveness of a drug, especially when dealing with potentially life-threatening conditions, the same question is posed. From[0]:
. . . ethical guidance permit the use of placebo controls in randomized trials when scientifically indicated in four cases: (1) when there is no proven effective treatment for the condition under study; (2) when withholding treatment poses negligible risks to participants; (3) when there are compelling methodological reasons for using placebo, and withholding treatment does not pose a risk of serious harm to participants; and, more controversially, (4) when there are compelling methodological reasons for using placebo, and the research is intended to develop interventions that can be implemented in the population from which trial participants are drawn, and the trial does not require participants to forgo treatment they would otherwise receive.