Hacker News Reader

Comment by polalavik | Hacker News Reader

You would think doctors would be able to collect data and then come up with statistical tables of “hey you may have xyz. The risk to look into this further has these possible complications with these success/failure rates. The risk of no intervention has these other set of outcomes with these likelihoods”

All you get at a doctor is someone typically saying “eh you’re young don’t worry about it” until it’s too late. It would be nice if the medical world was more data driven but it’s more handy wavy “ehhh I went to med school and I think you are ok”

View on HN

I think the issue is that it becomes hard to estimate risk factors on an individual level with enough resolution to come up with such a table. Its easy to conclude that on average, walking a lot every day leads to better health outcomes, but probably hard to say whether you walking a lot would lead to better outcomes for you specifically. All this stuff has variance. On average, overweight people die sooner, but there are still overweight people who are long lived, and its hard to say whether you share those same latent variables that are actually influencing this result.

Your expectations are unreasonable. Most doctors are practitioners, not researchers. They are applying existing care protocols rather than collecting data and coming up with statical tables. Good doctors will follow evidence-based medicine practices where applicable, but individual cases often diverge from the standards.

There is a huge amount of medical research going on but the field is so complex that progress is necessarily slow. Carrying out long-term studies in humans is extraordinarily expensive because researchers need large study populations to extract a useful signal from the noise caused by confounding factors. You could make a case that doctors should intervene earlier to prevent chronic conditions while patients are still young rather than waiting for more serious signs and symptoms to develop. But on the other hand, the available drugs often have significant side effects that impact quality of life or cause other harmful side effects. So it's not an easy call and there are still many unknowns.

With all due respect, I think it’s unreasonable to call my expectations unreasonable.

That mindset is the mindset of an old company - “we just do things around here like that because that’s the way they’ve always been done.” The field should move forward with technology we shouldn’t settle for something because that’s the way it is.

Your expectations are unreasonable. This is not a problem that technology can solve. Large scale, long term human trials will always be expensive no matter how much technology we throw at the problem.

If you have a suggestion for improvement then be specific. Vague complaints aren't helpful.

High-fidelity simulations to reduce the number of failed trials.

OK? Where can I buy a high-fidelity simulation? You might as well say that the solution to high energy costs is just to build fusion reactors. Great idea in principle but we don't actually know how to do it yet.

Simulations are already used in the early-stage drug development process and they're useful to reduce the number of substances that move on to animal trials. If you build a better one then you can make a fortune. But we are at least decades away from being able to accurately simulate the complex interactions in a human body. And I doubt that simulations will ever be useful for writing something like a clinical practice guideline for prescribing statins; that type of knowledge can only be gained by conducting human trials to see what works in the real world. Engineers accustomed to working with machines and electronics generally don't understand how messy biological research can get.

> You would think doctors would be able to collect data and then come up with statistical tables of “hey you may have xyz.

There's a project that's attempting to collect such data on a large scale and long time span:

  The purpose of Project 10K is to develop methods that will predict
  diseases years before they break out.

<https://www.project10k.org.il/en>

AI is going to very early optimize this use case. I fear for the Dr's job because I can't see what many of them will be needed for.

Reply on news.ycombinator.com