| The doctor's doing the equivalent of trying to diagnose and fix a bug where they're only given 10 minutes once a week to try and solve it (losing all their context in between) and can't directly check anything but the most superficial things. If they need more information, they have to refer the patient to somebody that can run check for one specific kind of issue in one specific place and will then provide a two sentence report back. Then they'll get another 10 minutes with that extra data point to try and come up with next steps. They can refer them to a specialist, and that person is going to only look at one small part of the context to try and come up with recommendations. In this sort of setup it's _very_ easy to miss relevant information. Having somebody involved who is focused on the big picture and can surface relevant information and advocate based on it is really essential to getting the best outcomes. And generally the only person involved to take on that role is... the patient. I don't think that makes the entire system broken, but it could definitely be improved. The best solution I can think of is essentially to have someone with a medical background designated to take that role on your behalf. And apparently that's already a thing--a "case manager". (Though apparently in the US that term's been overloaded to mean "the person that works with your insurance company to do the wallet biopsy".) Though I haven't ever really run across anyone taking that role, and I'm certain wherever they are they're completely overworked and can't fulfill the role as effectively as we like. I'm very much _not_ on the AI bandwagon and am really looking forward to all of this falling on its face so I can stop hearing about it and we can go back to finding solutions to problems instead of problems for solutions. However this does seem like one place where a LLM and adjacent developments could provide some benefit. A large part of advocating for yourself effectively is just knowing the right questions to ask and what information might be important to surface in the current situation. And all of this is being handed off to human experts, not being directly relied on or acted on by a laymen. As a brief example: My wife developed a heart arrhythmia. The ER doctor was following the standard framework for dealing with this and was prescribing beta blockers and a blood thinner. While the ER doctor had a _lot_ of information in front of her face, what she didn't have was time to sit and dig through my wife's entire medical history. If she did, she would have found that the out-of-wack blood pressure readings she'd written off as "your heart's kinda fucked right now" were actually an on-going and previously diagnosed issue of low blood pressure. In that moment I didn't need to understand everything or make any concrete recommendations, just make the connection "slower heart = less output?" + "already low output" = "problem?". Surfacing that information and asking the doctor whether we should be concerned didn't set the direction of her treatment, but it did change it. The doctor went to consult with a cardiologist before prescribing anything and came back with a different treatment plan. It definitely seems like an area playing to an LLMs strengths (digesting a bunch of information and trying to relate it to the situation at hand) while avoiding the pitfalls. Ultimately everything is still in the doctor's hands, and the information it's giving is purely _additive_ and already being treated with suspicion. Even if it misses or hallucinates something we're not much worse off than we were without it. |
It's full of people, and things can happen in an unavoidable rush with no prior context. It's not possible to have enough context and act quickly enough in many cases, regardless of intentions or money. So people do what people do: learn from past experiences where the effort/success line lies, and sometimes over-fit for the next case.
Personal attention is needed to adjust that line. Better tooling for personal attention does a lot to make that more possible. I do wish legitimate attention were easier to distinguish from hypochondria, and more doctors were willing to listen, but it's full of people - any pattern you can come up with will be wrong in both directions.
I dunno if LLMs are the current best option or not (they do have a huge benefit of understanding vague phrasing though), but I am definitely in favor of more tools. It has almost always led to better outcomes.