The real problem seems to be the knee-jerk biopsy/surgery.
If a QA engineer points out a potential bug, it's foolish immediately/blindly delete chunk of code containing the bug. It's just as foolish to say we're randomly deleting too much good code, so let's fire the QA engineers.
Should it? Running around with a known cancer that your doctors refuse to treat (because they expect it to go away), is a major stress factor and may, in the end, damage your own immune response.
Actually, "more information may lead to worse performance" has been shown in another psychologically heavy field: the stock market. [0]
Humans are complicated.
[0] https://behaviouralinvestment.com/2019/01/09/can-more-inform...
From a Bayesian/logical POV, more information should still allow for a better decision.
The problem was the decision to undergo treatment when the complications outweighed the benefit. What if we did more screening but also less treatment?
Even if we perform zero treatment, this data argument should be sufficient reason to increase rather than decrease the number of MRIs we do. CT is different since there is some radiation exposure risk that must be considered.
If you do disclose the results and proceed with a normal patient relationship, you are in the same dilemma as above. You don't know if the tumor is dangerous or not, so the safe course is to do a biopsy, which is itself often invasive. Depending on the results that may lead larger surgeries to exicse large parts of organs and/or the starting of chemotherapy.
Also MRIs are expensive, and in the U.S. a significant amount of that cost (usually hundreds of dollars) is borne by the patient, even if they are insured.
But I don't think we lack for MRI or CT data, there are lots ordered every year. The biggest problem with gathering your data set would be medical privacy regulations. I would guess that medical researchers have already tried longitudinal studies on this to see if different imaging characteristics lead to different outcomes. It's a hard problem, the only data you really get from scans is tissue density, which can probably never tell you what you need to know.
That's a lot of wasted time. Time better spent having the QA engineer only look at bugs reported by users and not looking for them on their own.
In the case of the mammogram, it's even less clear cut. The mammogram only shows spots. A biopsy is necessary to determine if it's cancer. A biopsy is surgery and could have complications. And could in fact make the problem worse.
So the idea is that you only do the test if you actually have symptoms or a family history of breast cancer, until the risk level gets high enough that it warrants going to look for cancer without symptoms.
In either case you can choose to do nothing. Knowing that the bug exists should put you in a better place to make the right call.
If you have next to no risk of a disease, don’t get tested for it. You might get treated for it!
If this is true, then clinical decision makers don't understand or are not correctly applying Bayesian priors.
And I don't understand your comment about knee-jerk biopsy--if you're not going to do the biopsy why did you even bother with the mammogram??? My wife has been through that: Routine mammogram--it got read while she was still there. The radiologist didn't like a spot. Ultrasound. They still didn't like the spot. An hour of sitting around while they get authorization, then they do a biopsy there in the radiology facility.
That came back abnormal but not cancer. At that point the urgency was over, she was referred to a surgeon who removed what pathology said was a precancerous lump.
It's not, though. There's a cost to collecting the information, it's not always actionable, and it can be wrong. On the human side, there's a reason they say "ignorance is bliss."
Now if I could find its source...
Edit: a search revealed that it was about knee surgeries that do not always make things better.