You're 100% wrong here, this is not a given. Prostate cancer screenings, which are very common, can have both false positives as well as findings of cancer that is and would remain completely benign. These can both lead to unnecessary treatments that cause serious negative health effects, including incontinence and erectile dysfunction.
https://www.cdc.gov/cancer/prostate/basic_info/benefits-harm...
An ultrasound of the thyroid often leads to finding a nodule. Which leads to a biopsy. Which comes out indeterminate. Which leads to a thyroidectomy and life long dependence on thyroid hormones.
Another one is ductal carcinoma in situ. Read somewhere there is a 1% chance that will evolve into cancer. And yet you have women having double mastectomies and chemo for it.
This is by far the most inaccurate medical claim I've ever seen on HN. Where on earth did you get this from?
The whole point of diagnosing DCIS on screening mammography is that it avoids systemic therapy and mastectomy. It also wouldn't be bilateral.
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.
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.
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.
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.
I like your idea of working these things out statistically but by your phrasing it's not necessarily what's being done (as in there are no reliable enough numbers for odds that matter).
But even if those odds were available, no matter what patient chooses, once you tell then there is a positive result, the fact that false positive is possible and odds are not in favor of treatment so they make a choice to do nothing does not mean they get to live a normal life from now on. Patients are not pure bayesian choice machines. The choice you make will affect you in big unknown ways and the existence of such choice already affects you until your EOL and has repercussions. That cancer was detected in your body cannot be "unheard" or "unread" back. Maybe you manage to deal with it, maybe you will live in constant stress. and of course chronic stress is connected to tissue inflammation, sleep disruption and other issues.
Therefore your argument that screening is automatically good does not seem to be convincing to me
As I said above, I don't think that screening is automatically good because in some cases, working out the above numbers will tell you not to treat even if you had a positive test. Those cases should not be tested since they're not actionable anyway. That would likely be the case for a rare cancer.
I agree that a false positive is not great. But if you have the data, then the impact of false positive tests is already included in your data. You're comparing the total rates of death and other unpleasant outcomes for a tested population vs. an untested population.