Hacker News Reader

Comment by murican22 | Hacker News Reader

Reading this article makes one appreciate just how much expertise doctors have. As soon as the doctor was presented with a few symptoms, he was able to begin running through a checklist of problems related to this specific issue and arrived relatively quickly at the solution.

Many doctors are able to routinely do this for dozens of ailments under intense pressure, making them even more impressive.

View on HN

And unfortunately some doctors cannot make a proper diagnostic even after visiting them several times, and having a rather uncommon, but not rare, condition. First hand experience.

Well that's life.

I'm not surprised by that. There is an exhausting amount of information that they have to digest & recall. It makes sense that they would assume that the most common conditions are likely to be the cause of your ailment. Maybe AI or automated diagnosis could help in the future, by suggesting rare conditions that fit the symptoms as well as the common ones.

How can I rapidly ascertain (within the first minutes of meeting for the first time) if I'm talking to such a doctor, vs someone who's going to complacently sit around and pontificate about what's possible without actually actioning anything?

I’ve not found a way yet, other than being around when they handle some kind of crisis, seen that a few times while working in a hospital. Wandering though on my way to/from lunch break was very occasionally surprising. It’s quite a surprise the first time you see a hospital bed roll past as a doctor on top is doing cpr compressions then literally hops off, while a second doctor hops, mid-run, to resume the compressions... like trick horse riding or something. Only ever saw it once but damn it was something.

It makes me appreciate how much expertise these doctors have. Generalizing it to all doctors is going one step too far.

The pattern matching could easily be automated. It's basically a game of 21 questions. Doctors sometimes fail here by getting too fixated on prior probabilities ("this condition is very uncommon so you don't have it").

The expertise, and IME this is rare, is in turning vague descriptions into matchable patterns, prompting for more detail when needed, and explaining things at the appropriate level (which for some patients might be quite high, others quite low).

In theory sure, in practice nobody had built a general system that works well yet.

I think the point was that the difficulty for machines is not the differential diagnosis, but in coaxing out the right details of patient history to feed into that algorithm. I don’t know if differential diagnosis is automatable but machines definitely suck at the latter issue.

"While in theory there is no difference between theory and practice, in practice there is."

Maybe the patterns change and that is what makes humans good at it and automated things not.

Reply on news.ycombinator.com