Or it's because the problem is genuinely hard. See above ('imperfect sensitivity')... the same problem exists in e.g. blood tests.

I'm no radiologist but I've seen enough of this done that I'd lean that way. Especially if you consider realistic bounds on time and additional information.

yes, of course the problem is genuinely hard, but having less information about your body makes it harder, not easier

In the abstract more information is always better. In practical healthcare with real constraints, sometimes it makes things worse globally for reason above.

Think about say a diagnostic blood test with a poor sensitivity (but better than chance) on a very rare disease. As an oversimplification: for a bound take your population * workup rate * fatality rate due to workup and you have a baseline needless death due to the screening (NB there is basically always a fatality rate). The number you are comparing to is lives saved due to earlier detection.

Better than chance means it's strictly more information always. But as policy, you shouldn't do it unless you can be extremely confident that lives saves is much larger than lives lost due to the process.

This is a bit counter intuitive, but it is a fundamental issue in a lot of healthcare policy. The real calculations are more complex but in the end it is unavoidably actuarial in nature.

i explained this in https://news.ycombinator.com/item?id=37267690 except that i suggested using a better measure than crude fatality rate, such as daly

this is not a global optimization problem; it's a local optimization problem. one person is deciding whether or not to spend their own money on an mri, or a blood test or whatever, on themself. they are the ones in the best position to do that, and for very-low-risk tests like mris it is very rare that the optimal decision is for a rich person not to get them

Well I did say the actual calculations would be (and are) more complicated. And as I said, short of availability constraints, the MRI itself isn't an issue, it's the follow up.

Ignoring the fact that in the real world due to constraints this isn't really a local optimization, if we assume infinite availability you still have a problem that low sensitivity tests lead to excess unnecessary work up risks. Your assertion that it is always easy to get good information leading to correct choices here is just incorrect and probably naive.

The trade offs you are talking about are very well known and have been studied by professionals for decades; none of them will claim it easy - for good reasons.

Note I'm not appealing to authority here. Many people can educate themselves to the degree that in a narrow area of their own healthcare they are as or more knowledgable than the average professional. However, this is not at all easy, and will require a ton of work. The cost benefit for a lot of screening etc. just isn't there; for a particularly symptomatic disease it may well be.