4 ms·

So work the numbers. Put together the chance of false positives, false negatives, rate of negative effects from treatment, and rate of death with and without tr

So work the numbers. Put together the chance of false positives, false negatives, rate of negative effects from treatment, and rate of death with and without treatment. Tell the patient the risk of each outcome. The math is easy. If we don't have the numbers for it, then get them. Plenty of people get prostate cancer and some of them choose to just monitor. We should have plenty of information to make a rational decision. This seems preferable to blinding ourselves out of fear that we'll do something stupid with the information we might get.

idopmstuff 3y ago
> Tell the patient the risk of each outcome. This isn't a simple problem of calculating EV. Telling somebody that there's a 40% chance that the positive test is actually wrong and in the 60% case that it's right, 40% of the time it's going to be benign, but if it's not benign it might kill them but if it is benign and they do surgery they might be left wearing diapers is not a simple thing for a person to evaluate. Add to that the fact that people have a bias towards action, so doctors tend to overindex on treatment vs. just ignoring something, and you have an incredibly complex problem. > The math is easy. No, it's not. It's a series of probabilities combined with extremely subjective outcomes (getting erectile dysfunction may have a very different impact on your life if you're 40 vs. 80). > If we don't have the numbers for it, then get them. You're just trivializing medicine and medical research here. Why don't you just go ahead and build some AI that'll solve this whole problem by diagnosing cancers based on a blood sample? That seems easy enough. > This seems preferable to blinding ourselves out of fear that we'll do something stupid with the information we might get. Ironically what you're describing here is the opposite of everything you've just talked about. If we understand the numbers well, and from those we can conclude that tests are highly prone to false positives and thus that treatment based on positive results is more likely to be harmful than helpful, then we shouldn't take those tests. That's not blinding ourselves, it's acting appropriately based on understanding the math.
- DennisP 3y ago
  No, I mean you take the test and if it's positive then present options like this: Risk of death: X% with treatment, Y% without treatment. Risk of side effect A: X% with treatment, Y% without treatment. Those numbers take into account the rate of false positives and false negatives. They are clear and understandable. There are definitely situations in which you shouldn't test: where the rate of the cancer is low, the false positive rate is high, and the risk of treatment is high. In that case, the numbers can show that risk of death is higher with treatment than without, so while (noninvasive) testing doesn't make things worse if we're giving clear numbers, it doesn't help either; we might as well not test at all. But that's not true for everything. As for side effects, we should give patients clear numbers like this so they can make informed decisions. Adding up the number of false positives and negatives, and the number of patients with various outcomes, is not comparable to using AI.
  - throwaway290 3y ago
    Simply having the information that they maybe have cancer, people are not great at just living unstressed because it may or may not be a false positive. And stress is known to cause a lot of problems including weakening immune system.
    - DennisP 3y ago
      Which is why you only test in situations where, if the test is positive, you have better odds if you treat.
      - throwaway290 3y ago
        Better odds to live longer or have good life? Looks like this study says it's definitely not the former?
        DennisP 3y ago
        Untreated cancer is not without its effects on quality of life. As I said above, with a positive test, work out the numbers for death and other unpleasant outcomes, with or without treatment, and let the patient decide. If those numbers would say not to treat regardless of test results, then don't test. The study just says that most, not all, cancer screenings don't extend life. And it doesn't say whether doctors are giving patients the numbers I described. I suspect they aren't, which likely means that sometimes they treat even when these numbers would say they shouldn't.

idopmstuff 3y ago

> Tell the patient the risk of each outcome. This isn't a simple problem of calculating EV. Telling somebody that there's a 40% chance that the positive test is actually wrong and in the 60% case that it's right, 40% of the time it's going to be benign, but if it's not benign it might kill them but if it is benign and they do surgery they might be left wearing diapers is not a simple thing for a person to evaluate. Add to that the fact that people have a bias towards action, so doctors tend to overindex on treatment vs. just ignoring something, and you have an incredibly complex problem. > The math is easy. No, it's not. It's a series of probabilities combined with extremely subjective outcomes (getting erectile dysfunction may have a very different impact on your life if you're 40 vs. 80). > If we don't have the numbers for it, then get them. You're just trivializing medicine and medical research here. Why don't you just go ahead and build some AI that'll solve this whole problem by diagnosing cancers based on a blood sample? That seems easy enough. > This seems preferable to blinding ourselves out of fear that we'll do something stupid with the information we might get. Ironically what you're describing here is the opposite of everything you've just talked about. If we understand the numbers well, and from those we can conclude that tests are highly prone to false positives and thus that treatment based on positive results is more likely to be harmful than helpful, then we shouldn't take those tests. That's not blinding ourselves, it's acting appropriately based on understanding the math.

DennisP 3y ago
No, I mean you take the test and if it's positive then present options like this: Risk of death: X% with treatment, Y% without treatment. Risk of side effect A: X% with treatment, Y% without treatment. Those numbers take into account the rate of false positives and false negatives. They are clear and understandable. There are definitely situations in which you shouldn't test: where the rate of the cancer is low, the false positive rate is high, and the risk of treatment is high. In that case, the numbers can show that risk of death is higher with treatment than without, so while (noninvasive) testing doesn't make things worse if we're giving clear numbers, it doesn't help either; we might as well not test at all. But that's not true for everything. As for side effects, we should give patients clear numbers like this so they can make informed decisions. Adding up the number of false positives and negatives, and the number of patients with various outcomes, is not comparable to using AI.
- throwaway290 3y ago
  Simply having the information that they maybe have cancer, people are not great at just living unstressed because it may or may not be a false positive. And stress is known to cause a lot of problems including weakening immune system.
  - DennisP 3y ago
    Which is why you only test in situations where, if the test is positive, you have better odds if you treat.
    - throwaway290 3y ago
      Better odds to live longer or have good life? Looks like this study says it's definitely not the former?
      - DennisP 3y ago
        Untreated cancer is not without its effects on quality of life. As I said above, with a positive test, work out the numbers for death and other unpleasant outcomes, with or without treatment, and let the patient decide. If those numbers would say not to treat regardless of test results, then don't test. The study just says that most, not all, cancer screenings don't extend life. And it doesn't say whether doctors are giving patients the numbers I described. I suspect they aren't, which likely means that sometimes they treat even when these numbers would say they shouldn't.