I understand it is not a black-and-white situation, and that data is obviously not perfect. And I certainly am not advocating that doctors should order a gamut of tests unnecessarily. But if there is any doubt, how likely is the added data to harm more than help? And I firmly believe the doctor should not fight the patient on a test unless there is a strong reason not to (apart from an obvious medical disorder such as Munchausen syndrome it is hard for me to think of a reason).
If the added data is noise, it can easily be ignored. If it's not noise, then you're lucky to have it. And figuring out whether or not it is noise is important, but you can't do that without the data.
But again, I am not a doctor, and how much bandwidth they have and where their priorities lie is not something I intimately understand. However something that has become blindingly obvious to me is that most doctors do not have a firm grasp of statistics, and will advocate for new drugs when the actual test results are borderline statistically insignificant and easily explained away by confounding factors (most obviously the placebo effect). Sadly not all trials are double-blind, something else I do not understand.
You can’t ignore the noise because we don’t know it’s noise. I know I’m failing to get that idea across, but I honestly don’t know how to articulate it better than I have been. I can tell you’re honestly trying to understand, and I feel the blame is likely on me as a communicator.
Possibly what I’ve failed to communicate is this:
MRI, CXR, etc are not images of the body. It’s not like getting a photograph of a liver and saying, at least we know this is or isn’t going on in the liver. They’re indirect measures of certain attributes of the body, such as tissue density, which we use - coupled with their medical information, and the mechanisms of likely diseases - to infer what’s happening. That’s why reading radiology is a medical specialty, and not something anyone with an anatomy background can do. (There’s a radiologist currently browsing the thread - he’s very welcome to correct me if I’m wrong about what radiology “is”.)
Because of this, every such finding has to be interpreted in a context, and studies tell us how.
Completely out of context findings aren’t a big problem if they’re completely unambiguous: hey, that bone is in two pieces and it should be in one.
What about the finding that isn’t, though? This is equivalent to not having any information on a test’s false positive / false negative rate, only now it’s open-ended to “every condition that could look like that thing” because the defining characteristic of an incidental finding is that it’s -incidental-. It’s not related to any symptoms. So what do I do with “every disease or non-disease process that could potentially look like a spot on the lung, without any accompanying symptoms of that disease”?
What I believe is the responsible answer is: “if I think the pre-test probability isn’t borderline zero, AND the post-test probability would change my course of treatment or diagnosis, order the test. If the pre-test probability is so low that any positive test result would be very likely to be a false positive and thus force me to act in a manner harmful to the patient, don’t order it - it shouldn’t be allowed to change the course of treatment. If the pre-test probability is already so high that any negative result is likely a false negative, don’t order it - it shouldn’t be allowed to change the course of treatment. Only order tests whose results should impact the course of diagnostics or treatment.” What do I do with findings that haven’t been studied in a given context, so I have no clue what their impact on the post-test probability of a diagnosis is? I don’t know. But “test just in case” isn’t the zero-risk option. There aren’t any zero consequence options.
Not every test is an RCT because grant funding agencies don’t provide the budget, plus or minus, many RCTs we’d like to do are unethical (if you have good reason to believe one course of therapy is superior to another, you don’t have the clinical uncertainty to ethically allow randomizing people into an inferior therapy), plus or minus many sub-populations are just too uncommon to build an RCT on without a gigantic budget that facilitates long collection periods across multiple institutions.
If the added data is noise, it can easily be ignored. If it's not noise, then you're lucky to have it. And figuring out whether or not it is noise is important, but you can't do that without the data.
But again, I am not a doctor, and how much bandwidth they have and where their priorities lie is not something I intimately understand. However something that has become blindingly obvious to me is that most doctors do not have a firm grasp of statistics, and will advocate for new drugs when the actual test results are borderline statistically insignificant and easily explained away by confounding factors (most obviously the placebo effect). Sadly not all trials are double-blind, something else I do not understand.