Hacker News Reader

Comment by version_five | Hacker News Reader

I'm actually less concerned with the privacy implications (which are concerning) than with the idea of more "AI" in healthcare. AI is a shiny object that does not solve any of the most urgent problems in healthcare, and almost certainly distracts from them, while all the "quick wins" are going to be the usual automation of what should be personal interactions, making it worse for everyone that is not the "average" customer. AI investment in healthcare is for the benefit of the AI providers only. It is not going to benefit patients, certainly not vs other ways the money could be spent (more staff, more beds)

View on HN

Right now rich people in developed countries can afford the best doctors. Most people in the world get shitty healthcare. An AI assisted medical staff can significantly raise the bar.

In many low income settings, even access to a medical professional means a long wait time. I see AI serving a valid purpose of guiding busy humans who have little free time. An AI can take current symptoms, past medical history, clustering the data based on travel history, real-time outbreaks, etc, etc and come up with candidate diagnoses.

If applied in the US, where a huge amount of people dont have access to decent healthcare who is responsible when a patient is misdiagnosed by an AI and treated for the wrong condition?

If a doctor is responsbile for vetting all the answers then it is questionable how much time will be saved.

All expected relevant information may be collected and presented in a handy report. That is good but doesn't require AI.

In order to pick a diagnosis in all but trivial cases the MD will have to invest real time into it and at that point making trivial diagnosis is also simpler to do manually.

I suppose the medical treatment can come with an EULA that informs the patient that the diagnosis might not be accurate. (Though that can happen from flesh and blood MDs as well).

Watson did not work out really well https://www.nytimes.com/2021/07/16/technology/what-happened-...

>who is responsible when a patient is misdiagnosed by an AI and treated for the wrong condition?

Good question, and it has been discussed plenty on the internet:

https://aublr.org/2019/10/what-happens-when-ai-unintentional...

https://www.scientificamerican.com/article/who-is-liable-whe...

https://www.sciencedirect.com/science/article/abs/pii/S02673...

https://www.griffithinjurylaw.com/blog/automated-cars-may-af...

There isn't any other way except governments implementing a legal framework of liabilities around AI tech.

>In order to pick a diagnosis in all but trivial cases the MD will have to invest real time into it and at that point making trivial diagnosis is also simpler to do manually.

People around the world still die from trivial, preventable illnesses simply for lack of access to a doctor/diagnosis. Its wrong to look at AI as magically solving everything. I see a use for it where it can improve the existing situation.

>Watson did not work out really well https://www.nytimes.com/2021/07/16/technology/what-happened-...

Our entire scientific journey is filled with failed experiments. Its not possible to stop AI from progressing, but we can and must use an ethical approach.

Please explain how an AI can take current systems and past medical history. No AI system has ever successfully demonstrated such a capability. Especially not with patients who may have limited literacy.

There isn't much to explain. Existing medical data is processed by the AI. New data is collected by means of electronic questionnaires. Questions are chosen in succession based on past responses.

People who have learning disabilities, or disabled in other areas - sight, hearing, or generally unable to interact with the electronic system will require human assistance. The global literacy rate is 87%, and there is room for more than one system so we can cater to everyone's needs.

Your explanation is inadequate. The necessary information can't be effectively gathered with electric questionnaires. This has been tried and it doesn't work. Your comment is typical for software developers who are ignorant about the basics of healthcare delivery.

>The necessary information can't be effectively gathered with electric questionnaires.

Your explanation is inadequate. Which systems are you referring to and how do you know it can't be done?

> AI is a shiny object that does not solve any of the most urgent problems in healthcare, and almost certainly distracts from them [...] It is not going to benefit patients, certainly not vs other ways the money could be spent (more staff, more beds)

I am baffled. AI clearly accomplished amazing things very recently, at the very least in language and image generation, self driving and problem solving as it pertains to games and upset a lot of expectations in the process. AI also, clearly, is used as a buzzword to confuse a lot of people (often times including the ones who are using it).

What grand insight could anyone have into what is going to on right now everywhere (and, clearly, there is a lot of stuff going on, right now, everywhere) to justify such a sweeping statement?

Maybe I don't understand your point, but as I read it you're saying "AI clearly accomplished amazing things very recently" as evidence that it will be the best (or a comparatively effective) use of healthcare dollars?

IBM Watson used this argument in marketing: "it can play jeopardy, now we'll focus it's powers on healthcare", and all their projects ended in failure.

AI doesn't work on edge cases (and typically doesn't know when it gets an edge case), distracts from resource scarcity by optimizing averages of shitty metrics (same as most "data driven" consulting projects), and diverts money from more helpful things. In a world without resource constraints, it would be worth implementing, but for now, it's a grift to make money for AI consultants, and will be worse for patients

My take on this, although it is not entirely clear from the article, is that Epic wants to move most of their computing to the cloud and Google felt left out and now they have a hospital chain who wants to partner which gives Epic a reason to add Google cloud compatibility (i.e., a customer will pay for the translation). There are a variety of things that might happen to the data once it is in the cloud. AI's can be trained on all the patient data and look for correlations (such as "all doctor X's patients do poorly" or "Unit 12 is responsible for most sepsis problems".) AI's are NOT going to be making a diagnosis any time soon, the doctors won't stand for it, and there is no AI capable of it yet. Even AI's that inspect radiology can only report to a real radiologist who then has to make their own decision (and is legally responsible for that decision.)

Another use of the cloud data is to aggregate data across many institutions, mostly to allow medical studies. The NHS in Britain has all their patients in one database which allowed them to do incredible studies related to the Sars-Cov-2 pandemic. Currently the US has no such capability, all that data is in silo's in each health institution, each with their own data formats and their own workflow. Get all the data into clouds, all in the same format, and it can be used for medical research across a much larger population. There are privacy concerns, but there are very desirable benefits to everyone. It would be preferable if government got involved rather than having industry do this for profit. The forced change to electronic health records would have been the perfect time to do this, but the lobbyists won instead so the US health datasphere is still highly fragmented. Migration to the cloud might be another opportunity to fix the problem.

I can tell you that the current focuses of AI implementations are around real, impactful issues: sepsis risk, readmission risk, deterioration index, etc.

The problems with AI in healthcare are:

1) People don’t want it to be a black box - that means quantifying the factors that go into a recommendation

2) Operationalizing AI recommendations is hard. AI tends to give gradiated information on binary decisions (e.g. there’s a 68% chance this patient is septic. Should someone go check on them? What if they were 49%?). The challenge becomes deciding how that information should be shown to people and what the acceptable false positive and false negative rate are.

3) The same problems of AI everywhere. Things like garbage in garbage out, unrealistic user expectations, feeling like it basically tells you what you already know, the challenge of getting insight from a pile of data.

No, the problem with AI in healthcare is that like much of healthtech is that it further reduces the ability of providers (especially in hospital settings) to respond to fluid and evolving situations that may fall outside the dotted lines that the AI understands or scenarios the system allows you to work within. Specifically, it creates further red tape that providers need to worry about, more checkboxes on an iPad to be clicked, more time required per patient on administrivia.

It could be done well but it will be done poorly, will increase the burden of front-line workers while making administrators feel like they can say they accomplished a big project this year. At the end of the day rather than making healthcare more auditable, practitioners will learn to just quickly fill in bogus data on the new system so they can go deal with the patient that's coding and when the AI gives a recommendation a provider doesn't like they'll just ignore it anyway.

In a good system that wasn't falling apart at the seams, AI in healthcare would be a boon, but in a broken system that's falling apart and failing its front-line workers, it will just serve as a distraction and another burden.

I think what you described falls under #2 in my reply. “Doing it well” is not a trivial option that people are ignoring; “doing it well” is the thing people are trying to solve. AI is not a magic bullet that always makes everything better.

Honestly that's worse than I thought. I work in the field, particularly in relation to accountable AI, and it's not OK to have models that tell you whether to check on people to make sure they're not dying unless there is also a human checking every case, which I hope is what's going on. How would you like to be different than the training data and deemed "no risk, 100% confidence" when you actually have a life threatening problem?

In a hospital setting, nurses and doctors round regularly. No one is talking about using AI as a replacement for that, because no one has anything approaching that much trust in predictive models.

Predictive models are most often used as either an alerting mechanism or an additional data point on a dashboard. You need to careful of alert fatigue, where too many false positives cause humans to disregard all alerts from the model. And if you don’t get people ignoring alerts, you can waste a lot of people’s time and energy by having constantly having them run to check on someone who is actually fine.

On the other hand, basic statistical techniques with access to a large amount of patient outcomes may find drug side effects faster. It's kind of crazy that this stuff is not better tracked already. Of course AI is the shiny new thing in the press release.

Perhaps, but previous attempts by researchers to generate clinical insights by merging clinical records from multiple provider organizations and then doing retrospective outcomes analysis have produced disappointing results. Data quality is generally bad, leading to a fundamental garbage in / garbage out problem. There is little consistency between documentation practices across providers, and many findings aren't recorded at all. So in order to get a useful data set, researchers have to do a huge amount of manual data cleansing. Current AI/ML algorithms are incapable of automating much of this work.

It could be greatly beneficial to the patient. Time will tell, but I also expect the financial optimization issues will take precedence over the rest.

Additionally, this is the typical example of the medical industry tackling problems from the wrong angle. We should be (MUCH) improving data retrieval and measurement reliability in medicine before we can hope to make anything approaching "stable diffusion for medicine"...

It's okay, it'll "optimize for outcomes"

Meaning... those with critical conditions will be left in a room (acceptable losses), those with mild conditions will receive a high amount of treatment (ensuring no one moves from mild to critical, overall reducing fatalities).

Reply on news.ycombinator.com