I built an open source AI tool to find my autoimmune disease
old.reddit.com
old.reddit.com
It won’t always be correct, and needs all the right disclaimers.
A friend of mine whose father is sadly very ill, made use of ChatGPT to interpret the oncology reports written about his dad. It helped him have deep technical conversations with multiple specialists and advocate for his dad, culminating in his advocacy for a treatment - that none of the specialists had thought of - which proved critical in extending his dad’s life. Now my buddy can chat to specialists in their own language and hands the info to ChatGPT to turn into layman’s terms for his dad - all of which my friend can check over for accuracy so he knows his dad is getting correct non-technical explanations.
That is massive empowerment.
In my local area’s subreddit, people are entirely dismissive of AI and will keep moving the goalpost about why you cant use it. It hallucinates! (But you can fact check it and I wrote that disclaimer) It warms the planet more than other research! (…. you’re not distinguishing between self hosted and data centers….)
But its Los Angeles and writers and actors and VFX are all getting replaced, so they have to keep saying ignorant stuff to sound relevant
Given the published and (and currently expressing) philosophies of the primary investors and shareholders of those companies, it is no longer any mystery why!
As a part time artist, AI has been great for me. I’m creating more, and spending less time on scaffolding (I am a middling-at-best programmer). I’m feeling empowered, not crushed.
How can you say something like this as Meta steals 82gb of text from the worlds writers and producers of knowledge without compensation, on the sly, to build their system?
Its the thieving and the disrespect!
This does not bode well for the future. If you think it stops at the artists you have so much contempt for, think again.
This is exactly about tools, tools and the fact that people with no artistic ability have relied on them heavily, it's not my problem or the problem of OpenAI that people are just bad at art, but people are indeed bad at art, and it's always people who blabber on about the tools.
Oh yeah, and while I'm at it, I don't believe in copyright, so there is that too.
Are you over-generalizing, something like “meta did something wrong to build AI, therefore AI is bad, therefore nobody can ever do anything good with AI”?
The reason for both things is that the best models perform, at best, on the level of a recent graduate.
When would you hire a recent graduate in either role, if you could afford better?
These models are essentially the same models for both science and art, and it was a surprise to everyone that GPT-3 was able to turn into ChatGPT, or that Stable Diffusion was able to generalise so well with relatively few issues (even despite the occasional Cronenberg anatomy study). The flaws with the LLMs that prevent accurate science are the same flaws that cause object impermanence in written stories; the flaws that prevent image and video models from being physically plausible are the same flaws — incorrect world model — that cause them to be wrong about weather forecasts, chemistry, etc.
In both cases, increasing quality of AI raises the metaphorical water level, and in this example rising tides don't lift all boats, but instead drown (the careers of) people who can't swim. I don't have a fix for that, and I'm deeply skeptical that any of the suggestions from the AI firms will work — they're not economists, and even if they were (or even if they hired loads), if the AI companies are right, this change will be at least as big as the industrial revolution, which upended old economic models.
Healthcare data does not. What data does exist is much more private and extracting data from research papers is much harder than extracting data from images. When something goes wrong, the impact is much more negative if it is being used in any serious capacity.
This is also why AI is just beginning to crack into short video and does nothing for long form videos yet, and even less on games, despite these falling in arts and crafts.
And so many tech bros misunderstand. Its not a problem with the tools and its not the money. Artists have extensive experience going hungry. So have I. I doubt many tech bros and technocrats know what starving feels like. Real hunger. But I digress.
Its how the tech is being wielded. Its not just: can the robot imitate a nice picture. Its: artists are deprecated.
And it mistakes artifice and artifact for art as a practice, and so conveniently cannot process any information about the subject beyond that final conclusion.
None of this is hidden or arcane stuff, but effort has to be made to learn new perspectives… kind of like how the world has readily adopted new technology.
What should be a reciprocal exchange is rapidly turning one sided and tyrannical.
Woe be the future where any such single-minded ignorance takes absolute control (as is the current missive emerging out of SV at present)
See Yarvin’s missive as adopted by so many VC’s in the current political plays. The way the tech is being wielded is all about power. Forced power over others. https://graymirror.substack.com/p/the-butterfly-revolution
This cannot be overstated.
The UK "not-for-profit" system has been systematically underfunded for decades [0], in the yacht-class' hopes that it can be privatized like the American system. Yes, it was once different.
Not that these things don't happen even in well funded systems, but the NHS isn't a great example of "not-for-profit" healthcare.
0 - https://www.ft.com/content/b593116d-f948-4757-b2fa-c74adadc8...
- no funding
- no ads
- etc... just an idea I had while locked up in SF county.
Its nothing that will fix healthcare or anything like that, but it's been fun learning about things like "Digital Twins", "FHIR" etc... and it actually works; meaning the Agent can help you achieve a desired health outcome (sometimes)
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.
U.S. health care spending grew 7.5 percent in 2023, reaching $4.9 trillion or $14,570 per person.
That's about 5% of healthcare spending going towards actual physicians. The logistic exigencies of dealing with insurance companies, seeking permission to treat, justifying, arguing, interacting with medical secretaries, is something on the order of an additional half of the typical doctor's time. They are strongly disincentivized from seeking out low probability explanations using tests which they need to seek permission to apply. They barely have the time to re-familiarize themselves with your chart.
We locked down the supply of physicians in 1997 via Congressional limits in how we fund medical residency slots. We further restrict physician time with problematically low rate setting in Medicare/Medicaid compensation, relative to market rates.
Everybody dies eventually. Everybody has healthcare needs.
A very large fraction of us will at some point suffer from some uncommon or rare condition, something that a physician doesn't see every day - a "zebra" rather than a "horse". There are enough of these conditions - tens of thousands now recognized - that the very low odds of having any particular problem adds up to high odds of suffering from at least one.
To recognize some of them, the doctor would need to synthesize an idea from half a dozen disparate pieces of information, two of which you've mentioned casually in a visit three years ago, two of which you've never told anyone, and one of which is in your chart, and one of which you mentioned casually to a specialist but who did not think it relevant enough to note down... And then they would need to do a differential against hundreds of megabytes of medical research. In five minutes allocated to your visit.
Doctors are trained with a certain degree of professional image to maintain, a sort of 'Wizard' status that must be kept to a tone of mysterious respect. Nobody wants uncertainty from a doctor, despite the fact that all of statistical reasoning and diagnostics is uncertainty, quantified. If they lose this image, it is felt, people start injecting bleach or suing them for malpractice or killing themselves over an innocuous granuloma that their doctor assures them is nothing. That is not a healthy environment to drill-down low evidence of probability outcomes.
Particularly if every test and treatment impoverishes you, and the impact of that poverty on your stress level and lifespan is dramatic.
While tripling the number of doctors would certainly help, AI and a great deal more population-scale diagnostics is just better suited to some of these limitations.
Is that what it will be used for? No. Initially, at least, it will actually be used to make everything worse, to automatically deny treatments, to argue with a doctor with the persistence of a chatbot.
And that's just for the stuff that's been medically recognized. Medical science has put comparatively little effort into researching ailments that cause chronic, non-contagious, non-fatal problems in your life, and less into resolving them. There is no agreement on how dandruff works, lower back pain was dismissed as psychogenic for decades, most headaches are undiagnosable, skincare is witchcraft.
A good doctor who sees you for a brief period every year is staring at your chart and furiously googling some of the things listed there to re-familiarize themselves for five or ten minutes before the visit, just to maintain an image of professional competency by not forgetting the words. Actually applying recent research... how?
>We locked down the supply of physicians in 1997 via Congressional limits
This part though is wrong. We temporarily limited the number of resident slots, but total number of residents has been growing for a while now. Not as fast as many people would like obviously.
Just looking at the numbers for recent years the total number of residents jumped from 134k in 2019 to almost 163k in 2024.
If you look at total number of physicians per 10k people, that number has been growing a steady rate since the 60s.
There are now 2x as many physicians per capita as there were in 1975.
We can improve healthcare systems, sure. But everyone needs to take personal responsibility for their own health. What is daily life-altering pain for a patient is just a JIRA ticket to a health care provider. No one is more incentivized than you to dig in and fix it.
Absolutely. I'm a caregiver to my father. He has many health issues. Some caused and/or exacerbated by poor lifestyle choices - smoking, poor diet, too much alcohol, etc. Granted his generation knew less than we do about these things today.
There's a popular myth today that doctor's just want to push pharmaceuticals on patients. Maybe this is true in some places or with some doctors, but I've been taking my father to doctor appointments for the better part of a decade. Every one of his doctors - primary care provider, nephrologist, cardiologist, and pulmonologist - emphasized the importance of eating better, quitting smoking, exercising, and drinking more water. It always landed on deaf ears.
After living one way for most of your life is very difficult. We also found out a few years ago that he brain damage from surgery to fix a leaking aneurysm in 1985. Both of these complicated his ability to make changes. Regardless, the doctors pushed pharmaceuticals as a last option.
I'm 50 now. I row almost every day, I workout with kettlebells, have a regular mobility routine, and walk outside often. My diet is pretty clean, I don't smoke anything, and do not drink. I could die tomorrow from any number of things, but it won't be because I didn't try to take care of my health.
All of that aside, I think it's important to not blame people for their health situation. It's easy to critique from the outside, but who knows what choices we'd make if we'd walked in their shoes and lived their life.
Worldwide, doctors aren’t given the time to work through the issues properly.
Doctors have many patients and limited time. Patients have one concern (themselves) with many more available hours to think about that one case.
It is no different than lawyers, teachers, etc etc. you have to advocate for yourself because you have much more time to dedicate to your situation than they do. Think of professionals as conduits of expertise that you can rely on to a degree but are best used to augment your passion for your situation.
In other words, I don’t think this particular issue warrants calling the healthcare system broken (though there are many other reasons to do so).
I'm not sure how your friend could know that it was critical in extending his dad's life.
All treatment decisions in oncology are basically "look at the available data, make an educated guess which one is best for the patient".
When it comes to some treatments, it's more an art than a science. Plenty of the available data is anecdotal at best - small trials (or retrospective data) on half a dozen patients with no control.
The role of the oncologist is often being able to interpret that data to make the best recommendation they can. This isn't easy, and for most oncologists who aren't doing research (and arguing with other oncologists about the latest data), they are going to rely on guidelines which attempt to look at all the literature and make recommendations based on the most robust data.
But whether your friend's dad has gotten therapy X or Y, nobody knows what the course of the disease would have looked like using the alternative.
I did a large amount of tests, including RSI and carpal tunnel, chasing what could be the source of my issue. I was only able to find out that I have Ankylosing spondylitis by talking to other people online, and asking my primary care doctor to test me specifically for the HLA-B27 antigen. I was just lucky to receive help from a random person on the internet: "why don't you check for X".
I imagine all the other people out there that were not lucky as me, or did not have access to that information. If AI can help with this, then it is great.
So I welcome the use of LLMs and other technology for this! Doctors are just humans, in every country.
It took 8 months from my first acute back pain problems to the diagnostic.
For years I thought it was due to a lack of exercise but after spending a year doing yoga, Pilates and physiotherapy things only worsened to the point where I could barely walk and could not bend down at all.
I had to strongly insist to my GP that this was not normal. After a referral to an orthopedic surgeon he ordered an MRI. The results indicated possible AS so he referred me to a rheumatologist. Tested positive for HLA-B27. Another MRI and an X-Ray and the diagnosis was confirmed.
AI was very useful in explaining the test results and doctors letters in plain English.
Now, I spend my evenings trying to learn as much as I can about basic medicine and delving into endocrinology. I don’t blame the doctors themselves—they’re overworked. But I do blame the system that pushes people toward becoming YouTubers instead of doctors.
I remember telling this to my dad 8 years ago and he still thought of me as just being weak or a hypochondriac. "It's all in your head son!" he would say.
2 years ago he started losing function of his left leg. A very snarky "It's all in your head dad!" was my first response when he started complaining. That was the beginning of his journey into understanding what I've been going through all these years. Today his left leg no longer works and he is also losing function of his right leg. He has abnormalities on his MRIs and EMG/Nerve conduction tests. Even with this, doctors have not been able to diagnose him with anything and all he gets is shrugs. He's extremely frustrated with the US health system and has decided to just give up and wither away instead.
It's nice to finally be validated by my dad, but I wish he didn't have to go through the same experience just to show a little empathy and understanding. I think most of the people in this thread will similarly have an impossible time understanding our predicaments until it affects them the same way.
If I feed in my symptoms today into AI, the same list I gave to doctors, I do get hEDS and specifically I get TNXB as a candidate gene. I don't include my WGS DNA data which came after I stoped bothering with doctors. AFAIK only specialists will diagnose hEDS and AFAIK absolutely none would predict TNXB. They call it clEDS instead and consider it 1 in a million, only if you have a specific TNXB SNP and it’s homozygous. Where I think most of hEDS is caused by presumed benign TNXB SNPs. Even the studies that specifically search for DNA candidates for hEDS skip over TNXB due to the belief that hEDS is rare and TNXB SNPs are common.
So I don’t know where AI it’s getting its info from but the TNXB theory is mostly a patient led theory held by some of us who have it. There are some fringe giftedness researchers who focus on the RCCX gene cluster - which incidentally cause a ton of autoimmune conditions.
So it’s definitely a situation where I think the answers from AI are far better than from doctors - I would have found out decades sooner. hEDS is such a perfect match for my symptoms that it becomes obviously true once I researched it so even if the AI guessed wrong it wouldn’t be worse than all of the doctors who guessed wrong. One even told me I should have thyroid surgery, thankfully I had already lost faith in them by that stage.
I have a suspicion that some non-trivial amount of the non-genetic EDS folks may suffer from fluoroquinolone antibiotic toxicity.
Among the many effects of FQs is how it can cause a net-loss of collagen in the body. It increases MMPs that breakdown collagen in the body. Part of the reason FQ antibiotics have a black box warning for tendon tearing.
I have a family member that was sure they had hEDS, but it turned out to be FQ toxicity. 20-30 courses over years that would degrade their health, one small bit at a time, unbeknownst to them. While safe for many, for some amount of the population, it can cause what feels like every bodily system slowly failing.
The frustrating bit of it all is how resilient the body is, and most common blood work shows up "normal."
Another possibility is that FQ toxicity is caused by certain genes relating to detoxifying (GSH, COMT, etc) - where the body incorrectly metabolises it, leading to toxicity against cells.
It's unfortunate that there's an insanely high number of folks in the population that are taking FQs and running into toxicity without any hope of figuring out what is going on with their bodies.
It seems that it would behoove folks to be able to take the test and know whether they are safe candidates to take FQ antibiotics.
By far the most common (and massively increased over the past few years) is the crowd that attracts all the eye rolls. These are almost exclusively women, usually white, age 16-35, presenting with a constellation of nonspecific chronic complaints including myalgias/arthralgias, fatigue, GI symptoms, etc. None of them have stretchy skin or abnormally hypermobile joints. Very high rate of fibromyalgia/POTS/CFS in this group. They're usually coming to the ED for diffuse pains, dizziness, or something else I really can't fix. Maybe there is some underlying organic disorder we haven't sorted out yet, maybe these are somatic manifestations of untreated anxiety/depression.
One thing about the hEDS diagnosis is the explosion of self-dx from social media (TikTok) which makes physicians cringe.
There's no good genetic test, and even if there was there is not much to be done other than supportive measures.
Hear me out, what if hEDS was 1/50 and not 1/50K or 1/15K or 1/5K or 1/500 or whatever the current literature has it at - which as a data science person is some rather large order of magnitude error bars for a figure. The 1/50 would include those diagnosed with generalized joint hypermobility, this would allow for the more common TNXB SNPs to be causal for hEDS.
And yeah, I've been told by many doctors that it's not worth even testing for because there is no treatment. Also consider the second order effects of not testing - how would you validate the original % numbers if people keep being talked out of getting tested, would subsequent surveys of population include those being talked out of the test? It's a good thing I don't rely on people who can't detect a condition to treat the condition. It is indeed very treatable - the dysautonomia aspect especially so.
What is the treatment for the condition itself? As far as I can tell, the recommended options treat the symptoms, but not the conditions.
It was suggested that a family member of mine has EDS, but if they do, it’s not acute and there would really be nothing to do. The doctor even suggested getting tested for it, but it would not be covered so they didn’t do it.
It's generally not the hypermobility that is the problem, it's all the autoimmune conditions that come with it. Here is a list of the most common ones, https://ohtwist.com/about-eds/comorbidities. The comorbidities often have specific treatments and it helps to know the name of the things when looking for the treatments. The one that I think limits people the most if they have it is the ME/CFS that is very common in people with hEDS. I'm of the opinion that ME/CFS is downstream from dysautonomia. Low Dose Naltrexone should be step one for dysautonomia, after that weak ligands such as modafinil in the morning and amitriptyline at night - these medications seem compliment each other rather nicely. Modafinil can cause gut issues though so if that's already a problem modafinil can make things worse. A very strict zero sugar diet as well, dopamine dysregulation can cause strong sugar cravings which does make this rather difficult. I think semaglutide will likely be shown to be effective for auto-immune conditions and is probably the best medication I've taken for mine - but I am extremely sensitive to it and had to start at 1/20th the normal starting dose and still had pretty bad side effects. I'm basically 100% except that I have to still be careful with PEM - I'm in the process of building up a tolerance to PEM with the hope to eventually fix it. For the actual hypermobility I think weightlifting is the most effective. I was getting Craniocervical instability for a while and did a year long course of hGH peptides and Test Cypionate and that seemed to resolve it - but I have a severe form of hEDS and was worried about it putting me in a wheelchair. I was willing to take some risks to avoid that.
Medical malpractice requires a deviation from evidence-based practice that results in harm.. hEDS is diagnosed based on guidelines, not just "obvious statistical patterns.".. If diagnostic criteria exclude certain patients, that reflects the current medical consensus, not physician incompetence. Med schools teach Bayesian reasoning and differential diagnoses, not just pattern matching.
If I incorrectly diagnosed a patient with hEDS without ruling out vascular EDS (vEDS) and the patient suffered an undiagnosed arterial rupture, that would be actual malpractice.
If a subgroup of less hypermobile hEDS patients exists, the solution is research, not accusing physicians of malpractice for following evidence-based guidelines.
If you consider the role of doctors to be finite state automata that execute instructions handed to them from standards bodies then we wouldn't need anything nearly as sophisticated as a LLM to replace them, a collection of decision trees would be sufficient.
Your input on the topic has not dissuaded me from my belief that doctors in general are bad at stats and therefore bad at their jobs.
In my opinion, hEDS is a real condition, but sick-tok is causing patients to demand from physicians a dx - regardless if they meet the clinical criteria for it or not.
Stick to your lane.
I would hazard a guess that I've read far more books and scientific literature on hEDS than you have - and I have the statistical background to understand it. It's a condition that is important to me. I think you're the one who should get out of the lane - you clearly don't know what you're talking about. My reference to the error bars on published figures of propensity of hEDS over time should have been an obvious clue.
Set up a "Health & Fitness" project in Claude (or whatever). Feed it:
* Basic data: height, weight, age, sex
* Basic metric snapshots from Apple Health or whatever: HRV range, RHR, typical sleep structure - go through everything and summarize it
* Typical diet (do you track it in MFP or Cronometer? Great, upload a nutrition report)
* Any supplements and medications you take
* Typical exercise habits
* Any health records you have - bloodwork results, interpreted imaging results, etc.
* Family history like you would describe it to a doctor
* Summary of any health complaints
* Anything else that seems relevant.
Then go through a few conversation loops asking it if there's any more information you could provide that would help it be more useful.
Then ask it things like "Given <health complaint>, what should I be doing more of? Less of?"—or "Please speculate about potential causes of <thing>".
Or, even if you don't have any particular health complaints you're working with, just being able to ask it questions like "What's one supplement I should consider starting or stopping today?" (and then obviously do some follow-up research...)
This is life-changing. Anyone skeptical of this has not tried it.
It's suggested a few supplements that have helped a lot, helped me figure out dosing and timing, pointed me towards taking gut inflammation more seriously as a part of what's going on (and suggesting various tests and experiments to help prove/disprove that), explained correlations in various bloodwork results, the list goes on.
It's—of course—not perfectly trustworthy but a lot of things are either trivially verifiable or are low-risk experiments.
Also, perhaps more often useful, is just "hey, tell me about ___ in the context of whatever I've told you about me" kind of thing".
Sometimes, yeah, the suggestions are very generically applicable and it's just a tool for thought.
>> It wasn't until I visited a rheumatologist who looked at the combination of my symptoms and genetic test results that I learned I likely had an autoimmune condition
On the other hand, interpreting domain specific data takes a lot of training and experience. Somewhat similar to looking at logs, knowing what to ignore is often more important. LLMs can dump a bunch of possible causes, some of which are perhaps hallucinated, however knowing what next steps to take with the results isn't really something most people will be capable of.
Perhaps the most likely immediate benefit for tools like this might be for the individuals who are motivated and determined enough to put in the time and gain the knowledge necessary to be effective with such tools(like with most things in life).
One thing is for sure though. As others have pointed out in this thread, you need to be your own advocate in the healthcare system.
Also, guess what rheumatologists specialize in. I'd give you a hint, it starts with 'a' and ends with your body's immune system mistaking healthy tissue for a disease.
Technically, the rheumatologist didn't do anything any other doctor didn't already do. They saw the problem in terms of their specialty.
Which isn't the worst way for them to approach things. I will go to a foot doctor if I believe it's my foot. A joint doctor for my joints. Going to specialists is pointless if you don't have any idea of what's going on. That's why GPs exist, to try and steer people to the appropriate specialist.
It was pretty good! Unsurprisingly, it did hallucinate sometimes or miss some nuance. But, I knew I was talking to something that did that. It was helpful for me, even with errors. I didn't, however try to just say "hey given all this what's going on with my baby"
On another topic:
If I were you, I'd make this thing speak FHIR or some other common format. That'd make using this thing in an ecosystem of other tools a lot easier. There are even FHIR graphql apis you could just polp on top maybe.
Why didn't the medical procession find my autoimmune disease after $100k and 30 hospital visits?
Old broken systems tend to get replaced, not improved.
> The most frustrating part wasn't just the lack of answers - it was how fragmented everything was. Each doctor only saw their piece of the puzzle: the orthopedist looked at joint pain, the endocrinologist checked hormones, the rheumatologist ran their own tests. No one was looking at the whole picture.
The more you go beyond the common diseases that they see a hundred times a year the more your chances of getting the right treatment approach zero. Occam's razor is usually right when diagnosing people, but it kills the uncommon cases.
"it wasn't until I visited a rheumatologist who looked at the combination of my symptoms and genetic test results that I learned I likely had an autoimmune condition."
It should prbly be phased as
"why ppl fall for any BS on the internet"
I've shared this story before on here a few times: I have ankylosing spondylitis, and it took me a few years to get diagnosed, but even after I got diagnosed treatment didn't seem to help. Humira, steroid shots, none of it. All I could do was eat NSAIDs like candy, which helped in the short term but kinda fucked me over in the long term.
Then I saw a gastroenterologist to ask about a hemorrhoid, and he took one look at me and my file—walking with a cane at a very young age, HLA B27+, offhand complaints about "IBS" (note: it's not officially IBS unless you've ruled out other problems!)—and recommended me for a colonoscopy. My health insurance is decent, so I agreed, and he saw a bunch of inflammation in there. Turns out I have Crohn's, and the AS is likely downstream of that; when I started taking oral medication that isn't supposed to have any effect outside the GI tract, my arthritis vanished.
But your average provider won't see a patient with joint pain and skin problems and think "well obviously we need to stick a camera up their butt."
The post is vague, but the doctors did find the “likely” autoimmune condition around the same time that the LLM suggested it.
Which is not surprising, because this tool was simply taking the medical records and doctors reports and putting them into the LLM.
The author has also been dodging questions about what the LLM actually said. Reading between the lines, it sounds like it gave a list of possible explanations including “maybe an autoimmune condition”, not an actual diagnosis.
You see a lot of this in AI hoopla and even in some of the industry papers where AI is relying on data that only becomes available because of a doctor's orders, so you are not "replacing" it per se.
If anything this demonstrates that value based care would provide better experience and outcomes for people with hard to diagnose conditions.
I have done 23andMe where can I look for those genes?
There are plenty of studies showing that caloric restriction calms inflammatory responses which are responsible for the majority of common diseases. Without balancing the energy intake, I'm always slightly suspicious that any improvements with changing what I eat are actually caused by the change in how much I eat.
Example:me. Another example is https://reddit.com/user/MythicalStrength who's a strongman competitor.
> often different people report completely different dietary changes helped them
Nobody fixed T2D, (some types of) epilepsy, (some cases of) mental illness like bipolar/psychosis on high carb or high protein.
As a result, some people may find it more pleasant and easier to balance energy intake with different dietary styles.
One of the subject genes is called MTHFR. It reduces the methylation of Folic acid by ~30-40%, depending on which paper you read. Tons of people have this. It was this AND a gluten sensitivity which resulted in me having actually low levels of usable folate.
To find out if you have this go to this link: https://you.23andme.com/tools/data/?query=MTHFR. You can enter various other SNPs there if you find any of interest.
You can download your DNA there and enter it in another site that can tell you the other genes that could affect this sort of thing.
There are other things that could affect other people in unique ways so of course this isn't the only one.
What I did is ask Claude "why would eating meat make me feel dramatically better" and it told me about these genes. And then I went to 23andme where you can see individual SNPs and then I searched for other related SNPs.
I also seem to have non-Celiac gluten sensitivity which affects my absorption of Folate (and other things if I have recently eaten gluten which was almost always until recently) which is why this was a problem for me. And zero doctors or companies have ever developed a test for this or even take it seriously, including allergists and immunologists that I saw.
> Despite lots of research – and lots of buzz – the existing scientific data doesn’t support the vast majority of claims that common MTHFR variants impact human health.
https://blog.23andme.com/articles/our-take-on-the-mthfr-gene
Is their take on MTHFR out of date?
MTHFR as told by the internet is an alternative medicine topic. It has been abused by influencers and supplement sellers because it’s extremely common to have one of these variants. The influencers use it as an “aha!” moment because most people who go looking for it will have it.
Most of the internet advice about MTHFR is completely without basis in science. If you have one of the rarer homozygous mutations then it’s worth looking at, but it’s not the explanation for all ills that you see from influencers.
Chances are good that if you’re reading this, healthy or not, you have a MTHFR variant. It does not mean anything by itself.
Methylfolate is also used as an adjunct to antidepressants because it has some efficacy in depression in some cases. That fact is probably not coincidental to all of these reports about people feeling “better” when they start taking it. MTHFR may not even be involved at all.
as an adult I'd get it every ~5 years or so. I got it last year and I decided to talk to ChatGPT about it and within a few minutes it had correctly diagnosed it as dyshidrotic eczema. 30+ years of wondering what it was and finally I found out, thanks to AI.
what a time to be alive!
Dyshidrotic eczema is pretty common. What doctors were you seeing that could not at least suggest that as a potential cause?
Beyond the obvious issues with LLMs, confirmation bias seems like an especially big problem for a tool like this. It is seeing only the tests doctors ordered, only the notes doctors bothered to take, only the symptoms the patient thought were relevant, and so on.
Would it come to the same conclusions if plausible synthetic results for other tests were included? Could it reliably predict results of a removed test given its previous conclusion? Could it predict the next action (test, referral, prescription) given results of a test?
Is anyone working on this kind of adversarial approach to LLM outputs? Obviously LLMs are far from ideal for an application like this but could this type of probing for internal consistency improve the error rate and allow for some measure of confidence?
For something like diagnosis where you can type things out with specific tests I think this isn't bad.
Those symptoms could be caused by almost anything. In the end the author says they "likely have an autoimmune disorder". To be honest just from that list of symptoms, my first guess would be aging. I really don't buy that this person's AI tool did anything here. I think they probably got lucky even assuming the "autoimmune disorder" is the actual cause of these symptoms.
> I spent over $100k visiting more than 30 hospitals and specialists, trying everything from standard treatments to experimental protocols at longevity clinics
I get it that this sucks. But when you have a list of vague symptoms that aren't very serious, could be caused by anything, and your records show 30 visits to hospitals and specialists, and you start talking about experimental life extension treatments, the vast majority of doctors are going to write you off almost immediately as a hypochondriac.
It's the same in any profession. I worked at Geek Squad during college. Someone comes in 30 times because "my computers feels a little slow", we've ran every test we can think of, and nothing is wrong. In that case I'm going to place the blame on either user perception--this specific customer has unreasonable expectations about what is slow and fast. Or there's a minor non-serious issue with his computer that no one is likely to be able to diagnose without doing something seriously invasive that is likely to cause more harm than the issue.
AI didn't actually give the initial possibility of having an autoimmune disease.
So it’s functioning as a second opinion
“Likely” had?
I read your post and comments last week when you copied it across a lot of subreddits, but I feel like the specific details of your story are still changing.
I understand you want to promote your side project, but the way you’re vaguely promoting it as having solved a medical mystery is playing a dangerous game with people who are desperate to hear answers from your tool.
I don't really know what the answer is. If AI systems are going to be used to deny people coverage (and they will be), why shouldn't patients have AI of their own? On the other hand, there's plenty of examples of AI giving out completely wrong advice, and it may be difficult for doctors to get them to accept that. Not to mention entire politicised subcultures are already built around medical misinformation.
To an external observer, there is no difference between someone who has a very low pain threshold and experiences normal body wear as extremely painful, and someone who has some mechanical condition causing pain.
> but endometriosis seems to be particularly bad.
I've known people who had endometriosis successfully treated after a long struggle to get someone to believe them.
And I've known someone who after a long struggle to convince someone to believe them, finally found someone. Then underwent extremely invasive and expensive surgery with no improvements.
It's an extremely difficult issue and it sucks for everyone involved.
I would hope a medical professional has some baselines to compare against unlike your methodology at Geek Squad.
You could have ran benchmarks, found a machine with identical hardware, etc.
IMO, in both of these cases (medical and geek squad) there are issues with diagnostic methodology.
1. In similar cases I couldn't find a machine with identical hardware to a 5 year old computer without spending way more money than this was worth.
Even if I did the only way to get an accurate comparison would be to wipe both machine and install an OS from scratch so that they are running the same software--the customer doesn't want us to do such an invasive procedure.
Extrapolate both of those problems to actual humans and you'll see the issues here.
2. Almost all of his complaints are subjective and impossible to objectively measure. We have no way to objectively measure things like feeling fatigue and pain.
> Even if I did the only way to get an accurate comparison would be to wipe both machine and install an OS from scratch so that they are running the same software--the customer doesn't want us to do such an invasive procedure.
You don't need to do any of that if you run benchmarks against averages. For example, PassMark or Geekbench are decent tools for assessing expected performance.
The problem with extrapolating this to humans is that benchmarks for people are wildly variable.
Given the specific nature of the authors complaints, I can almost guarantee that he underwent some standard benchmark tests.
Some people are fine playing games at 20 fps, some people swear they can notice a 10fps difference at near 120fps.
Severe joint pain and fatigue to one person might be something not even worth mentioning to another.
> In similar cases I couldn't find a machine with identical hardware to a 5 year old computer without spending way more money than this was worth.
> Even if I did the only way to get an accurate comparison would be to wipe both machine and install an OS from scratch so that they are running the same software--the customer doesn't want us to do such an invasive procedure.
Then you said:
> Benchmarks were part of every standard diagnosis, not just the weird cases.
So, you had a base objective measure, so why did you say that you could only measure it if you had a clean computer? You could install a fresh hard drive, boot into a PE and demonstrate to the customer its not slow, etc, so many things.
And then:
> Of course, but the entire point of that story is that people can have different subjective experiences of the same objective truth.
I am not trying to be a dick, but it seems more like the story was contrived to give your analogy some armor.
Because I didn't realize when you said "You could have ran benchmarks, found a machine with identical hardware, etc."
That you were talking about 2 separate things. I thought you were saying I could have ran benchmarks on this computer and compared to the benchmarks on an identical computer.
>I am not trying to be a dick, but it seems more like the story was contrived to give your analogy some armor.
There was no one single story. I saw plenty of cases like that. The contrived part was that they came in 30 times--that was exaggerated to match the number of cases in the original post. The kind of people who bring their computers into Geek Squad general don't care about objective measurement. They care about their subjective experience of whether this computer feels fast or not.
Sometimes it feels slower than it did when it was new for a real reason that isn't fixable--the OS was upgraded and some of the interactions feel slower even though the synthetic benchmarks are the same. The software they are running now has been bloated over the last 5 years by new features etc...
Sometimes it's some weird combination of stuff that you're never going to see unless you sit and watch them for a whole day. Sometimes it's stuff they'd never do while you're watching like opening 10 different porn videos at a time.
But the point is it's an analogy. It doesn't matter if it actually happened or not. It's designed to make a comparison to the original problem but in a domain that more people here will have had experience with--diagnosing computer problems vs being a doctor.
And 2. If you have chronic fatigue it’s nothing like fatigue a healthy person would experience - it’s next level. There were some thoughts to change the name to mild paralysis because sometimes you can be so fatigued that you literally cannot move.
There are objective ways to measure it. Post Exertional Malaise has a number of tests, two day max exertion test for one. There are also immuno assays which are pretty definitive. You can take plasma from someone with ME/CFS introduce healthy white blood cells, expose it to a salt water stress test and measure how quick the cells run out of energy and the results are definitive here too. Hard to fake that.
As for the subjectivity, when looking at post viral fatigue there seems to be 3 categories of people; 1 people who appear to be incapable of getting post viral fatigue, they'd die before reaching some threshold which suggests that there is a genetic aspect to susceptibility. 2 people who grew up with a progressive ME/CFS who are highly susceptive to post viral fatigue but consider the fatigue within their normal experience. 3 people who don't have ME/CFS but are at least partially susceptible, specially with Covid, and anecdotal accounts of their experience tends to have a common thread of never had experienced that level of fatigue before in their entire lives.
As for Post Exertional Malaise, again maybe this was a complaint of theirs, but they only mentioned slow recovery times.
If their symptoms are severe, they're really underselling them with their descriptions.
>That changes the equation such that spending more than the computer hardware is worth is now worthwhile.
The worthwhile point of that particular analogy is comparing invasive procedures. Doing invasive procedures like wiping the hard drive and reinstalling the OS vs non invasive testing.
You get to a point where invasive testing is only worthwhile if the symptoms are severe enough because of the risks involved in invasive procedures.
I would not call a PEM test invasive but I would consider it likely damaging and wouldn't recommend in general. A blood draw is hardly invasive and there is absolutely no reason why the tests cannot be cheap. Like $5 cheap - far cheaper than the $100K this person already spent.
Sure that's not an example of an invasive test.
Just did it last Saturday in fact.
I fed into ChatGPT my similar symptoms and test results and it came up with a variety of diseases that I, after further testing, don't have.
When my first symptoms appeared, I initially thought it was just aging, despite being a healthy 28-year-old man who enjoyed exercise and maintained a proper diet. I noticed health issues arising even though I took better care of my exercise, diet, and sleep compared to my friends.
Hospitals provide fragmented information across departments, which can extend the diagnosis period for rare conditions.
I believe medical services can be delivered through personal medical data + medical domain knowledge + intelligence (excluding physical intervention). AI can process more data than doctors have time to review, contains more knowledge than one person can study, and its intelligence will continue to improve.
This gives me hope, and I want to help others who could benefit from this hope.
The key part of that statement is "continue to improve" we have no way of knowing where we are on the sigmoid curve of improvement.
But right now. Today. Tools like this that spit out "you might have an autoimmune disease" aren't helping. Until they improve dramatically, the most likely outcome is that doctors get even more overwhelmed with people who think they have an auto immune disorder, but don't.
I'm much more medically knowledgeable than the average person. I'm reasonably smart, and interested in the subject. And then I married a doctor.
But even given all of those things, I'm very bad a diagnosing myself. I once swore that I had Giardia. I read dozens of papers and case studies. A close family member of mine was also having some similar GI symptoms at the time, and I even calculated the probability that our symptoms were independent.
After that was ruled out from testing and I kept having stomach pain, I did more research. I was pretty sure I had stomach cancer.
Eventually I got scoped--gastritis. Took a PPI for a little bit and it cleared right up. Gastritis wasn't even something that I had considered. It showed up in my search results, but because it wasn't a serious problem, I didn't focus on it.
Multiple times she was turned down for breast cancer screenings, she was told that she was too young (33 at the time). It was not until she had requested that the doctor write a note of their denial, along with reasoning, that they finally gave her some screenings.
It was stage 2(3?), it had affected several lymph-nodes, but she was lucky that she persisted and did not just say “oh ya it’s fatigue”. When I ask her about how she knew, it’s always the same response, that it was just a feeling, that she knew something was not normal.
There are costs to testing, financial and otherwise. For example you take someones family history, and other risk factors into account. There is a 1/100,000 chance that they have cancer. Then you add in the symptoms, fatigue, and minor aches and pains. That brings it to a 1/50,00 chance.
We don't have the money, technicians, or machines to run an MRI on everyone who shows up to the doctor with a problem that has a 1/50,000 chance of being cancer.
Then for many tests, with such a low prior probability, a positive test has a higher chance of being a false positive. So now you patients with massive anxiety out over a positive that has high chance of being a false positive. And you need to send them or further more invasive anymore expensive tests, that have real risks of harm.
A guy launched a car into space; I think we have all this stuff but we — as a society — just don’t find the business case for “poor people being alive” very compelling.
You are incorrectly calculating the “value of the appointment” from precisely the wrong viewpoint. Remember that healthcare is about the patient’s health, not the bill.
It's also just so wildly irrelevant to the topic at hand
Of course, to the doctor's defense, patients are not an easy population to work with, "Why can't I eat candy and cake 3 meals day and my diabetes is getting worse?"
"I HAVE cancer. I HAVE this autoimmune condition. I GOT a cold."
So people are in search of a label and the "thing" to get rid or fight, but imagine if all technology problems were labeled similarly. Rendering-itis. Memory-itis. PCI-Bus-itis. Imagine someone coming to a person going, "I think I have keyboard-itis, because when I type a key into the keyboard, it takes a couple seconds for the character to show up on the screen." When in reality, as the engineer, your intuition is that there's something amiss in the auto-complete microservice. Perhaps that service needs to be scaled up or something else is DOS'ing the service, Perhaps the autocomplete service is running across the country instead of the same datacenter, so you know which tests to run to isolate and solve the source of the problem. Medicine is basically that now.
Medicine has developed to the point where they have several hammers where if bloodwork == X || symptoms == Y, then A or B or C drug/procedure. So doctors are mostly the level 1 engineer after the level 1 tech support rep on a script.
----
Most medicine is merely a collections of latin labels, and unless the conditions map to well-known treatments, most doctors are in the dark. You were lucky that you responded to PPIs.
Gastritis is "inflammation of the GI lining" and the use of a PPI, is similar to a common second-line option like "reset the software settings" after the first suggestion of "reboot your computer." They scoped you, because they had no clue what was up. Scoping is basically like lighting up the debugger. They go in with a camera, and basically found nothing but inflammation, thus came the PPIs. They were mostly ruling out cancer or growths or other weird things like outside items. (ie ate a non-digestible item).
Basically, doctors are relying on pharmaceuticals to help out for a couple days/weeks, while the body addresses things on their own such that they don't become necessary anymore. That's probably most medicine. As most pharmaceuticals are symptom management, doctors really only have tools to help someone manage symptoms or alleviate the pain. They are just practiced in that troubleshooting process, the way tech folks get deep into the stack of the architecture of computing, networks, etc and can troubleshoot issues in the tech stack.
----
The way that there's only a handful of the most brilliant engineers who deeply understand the stack top to bottom in such a way that they can thread their way to "that's a problem in this specific subsystem and the latest change now makes the datastructures exceed the L3 cache size of that node and that's causing an additional 10 millis degradation." There's really only a handful of doctors who are intimate enough with the all the subsystems of the body to be able to drill into what's going on with a person, otherwise most operate on rules and mentally null pointer when they run into something that doesn't fit into their flow chart.
Outside of very specific genetic conditions, incredible amounts of sickness out there could be alleviated by supporting the biochemistry (limiting reagents) of the body, so its processes can keep it in tip-top shape.
I've met doctors that are passionate, but are hamstrung by "the process". The hospital has policies, the insurance company has policies, local health departmen thas policies, whatever it is. I hate that "some process" needs to be in place to keep things in check, lest we end up with patients abusing things, or doctors abusing things (SURGERY ALL THE THINGS, for example).
I think it's "the process" we have lost trust in. Individual people really do want to do good things for each other. But "the process" sadly has an awful lot of inertia, and doesn't make it easy to change.
There is a legitimate issue with concentrated ownership of provider organizations in some regions. More antitrust enforcement is needed to protect consumers but unfortunately that seems unlikely.
https://www.theatlantic.com/ideas/archive/2023/05/private-eq...
https://www.ineteconomics.org/perspectives/blog/private-equi...
"What if you didn't generate the heat in the first place?"
Personally, I have a hypothesis that there are systemic areas we need to address in our food supply to lower healthcare demand. This will sound kooky, but if we can get more sulfur and trace minerals into the food supply, we could improve the general well-being of massive amounts of the population. We really need to look at doing more research into additionally fortifying our foods.
Here's a nature article on the reduction of sulfur from the food supply: https://www.nature.com/articles/s43247-021-00172-0
NPK farming has ensured that the food supply doesn't have a rich set of additional minerals outside of (N)itrogen, (P)hosphorous and (K)potassium.
Then when looking at sulfur-containing molecules, the research on them tends to herald them as miracle-health molecules: MSM Taurine, B1, NAC. All bring sulfur into the body.
Then there's companies like Buoy drops that offer 87 different trace minerals and folks are swearing by the effects of them. Asian countries consume things like seaweed and get trace minerals like this from the the sea water, but a the standard american diet is really lacking in a lot of additional nutrition that would be productive to health.
https://peterattiamd.com/outlive/
As for the food supply, I have no doubt that many people are impacted by subclinical micronutrient deficiencies, whether it's sulphur or zinc or D3 or something else. But the evidence indicates that's less important than excess macronutrients for most people.
I wonder though if with the advent of GLP1 drugs, if obesity may already be a solved problem, but merely a matter of time?
I imagine, as with anything, perfect doesn't need to be the enemy of good enough. A huge multifaceted issue like obesity could be addressed from many angles.
Japan has an obesity rate of 4%. By law they require every school to employ a professional nutritionist, which requires 3 years of additional training on top of their education degrees. In a particular example of 1000 kids, not a single child is overweight. Every nutritionist is required to design the school's food curriculum such that all meals are made from fresh foods and to be balanced according to all the food groups, and they forbid even the use of processed pastes. All meals are made from fresh foods on-site.
Then the kids are then taught how to eat well and to fully cover their nutritional needs from 6-18. This would be the non-ozempic way to influence a nation to eat well.
If you're paying out of pocket, you have the right to demand as many tests and as many treatments as you can afford. If you're relying on an insurer to foot the bill, then that insurer has two choices - refuse to pay for some treatments, or watch costs (and premiums) trend towards infinity.
a) The knowledge is democratized on the internet (albeit often without proper context, and with somewhat limited accuracy)
b) You just care a lot more about your own, or a loved one's health/life and are willing to put a lot more effort than a doctor can or an insurance company is willing to pay for.
You may not have a medical degree, but if you're willing to put in 40 hours doing research and your condition is even somewhat unusual, you can probably exceed the knowledge of your doctor about your specific condition.
>> you can probably exceed the knowledge of your doctor about your specific condition.
It depends on the doctor. That's why in the US patients first see a general practitioner, then are referred to a specialist. I've met some less knowledgeable doctors for my condition, but I don't discredit them for it. It's a difficult profession.
Plus "doing research on your own" is not a bad idea, but if one deep dives and starts believing ivermectin cures everything, well, I'd say they did poor research, and poor research can be worse than listening to a doctor.
When I need my car fixed, I don't go to the local psychic or consult Fox News. I go to a mechanic.
The medical shills are working hard on this thread.
Such a tired take. What we do know is we are not on the flattening part and literally every company working on these technologies is saying there is no end in sight.
Literally every company making money selling Foobars is saying there's no end in sight selling Foobars?
WebMD has been like this well before current iterations of LLMs. Pretty much any symptom you looked up would very meme-like say "it could be cancer". Only, it wasn't trying to be a meme. I always saw it more of a CYA on webMD's part though, without trying to sound authoritative like LLMs do.
https://gutenberg.org/files/308/308-h/308-h.htm
Published 1889
Food, bitter beer, exercise, early bedtimes and "stop stuffing your head with things you don't understand".
> I followed the directions, with the happy result—speaking for myself—that my life was preserved, and is still going on.
I read the book as part of my English - II in school, but no part stuck to me as much as the Liver cure for children.
The most dramatic was in 2009. My wife was admitted to emergency at 22 weeks pregnant with severe cramping, nausea, and severe IBS like symptoms. She couldn't keep food down, leading to daily stomach pumping and eventually IV nutrition.
The episodes started every few days but grew more intense and frequent. After 8 days, the first doctor seemed lost. When I asked about a second opinion, he lectured me that it was "his call to make, not mine." That killed any trust - I asked him not to return.
The next doctor, an eccentric but brilliant GI specialist, ordered new tests. I obsessively googled every result (partly to cope, partly to help). When he suggested a motility drug, my research found a case of spontaneous abortion linked to it. We declined.
Three weeks in, I noticed something odd in her lab work - different blood counts from different locations. More late-night googling led me to placental percreta, where the placenta grows through the uterus and attaches to other organs. Several ultrasounds proved inconclusive.
With daily episodes now and my wife deteriorating, we insisted on an emergency C-section despite warnings about our baby's lung development. During the procedure, even with an epidural, she went into extreme pain. Her eyes rolled back, the surgical team's demeanor changed instantly, and I was told to leave the OR.
Waiting outside for who knows how long. I knew the risks. Some noise down the hallway, a few nurses called me over. They were wheeling a humidicrib down the hall to the ICU. Little champs whole leg was no longer than my pinky finger. They took him into the ICU and sent me to the hospital room to wait.
The OB arrived an hour or so later and explained she was alive and they had to remove something. He took out a photo and and showed how the placenta had indeed broken through her uterus and attached to the intestine. He then pointed to a particular point and explained how it would have ruptured at any moment. The decision to push for the C-section likely saved both lives. Our son, never developed lung issues. Lots of luck.
This experience showed me how even amazing doctors, overwhelmed with cases and data, can miss crucial connections. In 2009, they didn't have the tools to correlate data the way I could with google search-fu.
AI diagnosis tools will save lives - they can make these connections faster and more reliably than either doctors or desperate husbands with search engines. And it will just improve over time.
Medical training and experience can reduce the impact, but not overcome it. It is a reason why someone can't entirely outsource their heath to the medical system, and something that I think people miss when they set medical expectations.
From the perspective of the medical system, it is a numbers game, where the optimal number of deaths is non-zero. More time and effort spent in one area means less in another. Saving one life might cost two others.
Because of this, I think it is most useful to discuss ways to increase medical efficiency. Better data transfer, better collation, better processes. AI/big data review has a place in this, but we should be aware of it's limitations, particularly on the data it doesn't capture.
I don't have the ins and outs of your case. I'm a student, not an obsterician, or neonatologist. That said, I do know delivering a 25 weeker is insanely risky even if perfect care is provided, for both his immediate outcomes and avoiding long-term disability. (80% of babies born before 26w have some form of disability e.g low IQ, blindness, etc [1]).
No sane obstetrician is going to deliver based on a family opinion like that - it is a true last resort. If you were wrong, or less lucky, as horrible as it is to say, your son would not be with us and the surgeon would be out of a job with the "what if" on their conscious (which trust me, all obstericians have anyway).
When your son needed to be delivered, it sounded like he was delivered. Frankly no neonatal respiratory distress at 22 weeks sounds insanely lucky to me; normally before 35 weeks it's a concern.
I can't say if this was a failure of the medical professionals in charge of your care or not, but from the information you've given us it doesn't sound like it. Tens, maybe hundreds of patients will have presented identically to your partner with a far less severe diagnosis. Getting their babies out early could have left them permanently disabled or dead, on the offchance it was percreta/accreta/increta: especially in the absence of definitive imaging, which they did attempt to perform. Medicine is not perfect by a long shot. Managing uncertainty is one of the many skills of senior doctors.
Medicine isn't as simple as a lot of people in this thread think it to be.
DOI: Medical Student
[1]: https://www.tommys.org/pregnancy-information/premature-birth...
In our cases, her condition was deteriorating at a frightening rate. She could barely speak. Under such stress one must also consider the possibility her body giving up and aborting the pregnancy. Doctors had put cameras up and down and ultra sound etc. They were trying to keep her going as long as possible to give baby the best chance. I definitely would not advocate for anyone to make rash decisions in such delicate situations, and of course, my post omitted myriad details for brevity.
The cause was later determined to be the result of a D&C for retained placenta that was performed following the birth of our first child a few years earlier.
Do you know how many times I've got prescribed a drug that would harm me more than do good? Do you know how many times I've got prescribed an antibiotic I'm allergic to, even after telling them I'm allergic to it, even after three different people wrote that down in many different places? Most modern doctors are criminals and I will never be ashamed to say that, because it's true.
Please keep doing what you're doing and ignore the naysayers, they're usually on some pharma's company payroll, sad.
But it could be something I eat, definitely, what I have feels like a sudden allergy to something that I haven't been able to single out from the environment.
You probably won't believe it, but that's not a thing. My wife is a doctor and I can tell you exactly how she gets paid. The only real misaligned incentive I've been able to find is that she could pad her paycheck by spending more ordering tests.
But she her pet peeve is doctors ordering tests that aren't going to change the course of treatment, so she's definitely not going to do that.
Anecdotally, I have heard that hypochondria is a problem in this space. So I would want to know the ratio of incorrect self-diagnosis to confirmed diagnosis, and what percentage of the system's totally human and technical resources are spent recognizing false positives.
I think that this sort of effort is probably a positive one. Patients should have access to their records are are more likely to be able to centralize data for analysis by a 3rd party - with all the attendant risks - more effectively than any other party. Secure systems without conflicts of interest - ie, insurance companies - to aid multidisciplinary diagnoses over longer time frames is a good thing.
But the risk of AI hallucinations and doctor/diagnosis shopping are real. This post was light on details on the diagnosis, how many rheumatologists they saw to get it, what tests confirmed it, and if the treatment has been effective in a way that cannot be skewed (https://pubmed.ncbi.nlm.nih.gov/30139289/)
It's like how the tech world did not notice or address "the website is slow" in the 90s until folks started writing about how the 95th, 98th, 99th, 99.9th percentile times affected user experience. Once we could link 95th percentile page latency to revenue/profit, all of a sudden, investing in performance became a priority.
I'm not sure that we have managed to link and make a case of linking GDP (or some other metric) to overall health of the population at the fringes.
She had five knee surgeries, starting at age 20, to address instability and knee pain that were not caused by any injury (nor has she ever been overweight), just standing and walking like an average person. She has no foot arch, bruises very easily, has velvety soft skin, has an unusually high arch in the roof of her mouth, and has POTS (postural orthostatic tachycardia syndrome). These are all common hallmarks of EDS, but no doctor connected the dots until the 5th surgery. BTW, none of the surgeries helped and in fact made things worse.
A lot of men won’t have that same experience, they’ll have the experience of waiting too long to ask for help. Which in a dark ironic manner is one way of keeping things equal, one won’t be taken seriously for 5 years, the other won’t ask for help for 5 years!
I’ve concluded that most doctors are highly credentialed decision trees, they all start at the top of the tree no matter your history, and if they hit a dead end you’re going to told you are the problem, directly or indirectly.
It's weird how people who complain about inequality and lack of empathy towards one group, are completely incapable or unwilling to offer any of that to other people.
Not sure why you find that weird. It's a complaint about inequality towards women, and also the inequality that men face due to misguided masculinity encouraging them not to seek medical advice (and instead rely on DIY nonsense).
https://pubmed.ncbi.nlm.nih.gov/9934383/ https://info.primarycare.hms.harvard.edu/perspectives/articl... https://www.aamc.org/news/why-we-know-so-little-about-women-... https://pmc.ncbi.nlm.nih.gov/articles/PMC9657442/
It's a generalization that paints men in a negative light, which I took personally.
Anyhow, I don't usually pick fights online, I'm just particularly miserable today because my stupid health keeps hurting me for no reason.
To say something on topic - I've been talking to Claude about my health problems, and, while I still can't figure out what's wrong with me, I find it extremely helpful. It teaches me all sorts of things about health and medicine that no doctor would have the time or patience to explain, and helps out a lot, both with helpful information, and emotional support. I realize that it's kinda bleak to be receiving those things from an LLM instead of a person, but here we are.
I've been to dozens of doctors, spent a lot of money, they all were extremely unhelpful. Claude does offer a lot of advice and information that no doctor ever shared with me.
> It's a generalization that paints men in a negative light, which I took personally.
With compassion, I seriously suggest that you adopt a radically different perspective. OP did not come after you personally. Yet you say that you took it personally. How did you come to that reaction?
It helps to not look at all facts as binary. This leads to a brittle world view as one tries to reduce reality’s complexity into a bits. Social group membership is not like sets and propositional logic. If you’re applying that system of logic to the real world, you’ll derive false things. I look at it like the axioms are verifiably false by observation, thus poisoning the whole system.
I don’t usually wade into deep comment threads, but I want to help here. I see something you said that I used to struggle with myself. And I struggled until someone else helped me.
Good luck with your health.
Yeah, no good comes from taking things personally, or getting offended at things, or arguing about who's getting treated more unfairly. I'm just not my best self today.
No one wins here, but some suffer particularly egregious losses.
Doesn't this explain why doctors could take men complaints more "seriously"? If doctors are aware that men wait longer before asking for help, they'd know that there's a smaller window for a positive outcome in the event of a serious underlying medical condition.
I was talking to a doctor who specializes in a hard to diagnose condition. Somewhere close to 80% of his referrals are women. If you search online there are countless stories about how it takes longer for women to get diagnosed despite higher prevalence in women. The Facebook groups are full of complaints that women are at a disadvantage with this condition and it’s the doctor’s fault.
But when you look at research where the condition is discovered incidentally in people’s records, the gender difference disappears. When you randomly check for it, there is no gender disparity. It’s just random.
The difference, he said, is that many women are more persistent about getting diagnosed while many men avoid going to the doctor as long as possible. The fact that women are diagnosed at a higher rate and with longer time to diagnosis doesn’t, by itself, explain that a condition is more common in women or that doctors ignore them longer.
If anything, it's a bit sad that this person felt so let down by their healthcare system that they decided to build software that glues their EMR and LLMs together (maybe that was therapeutic in and of itself)
The author fed their medical records to the tool and then, as far as I can tell, the tool included autoimmune disease as a possible explanation.
Every comment that I saw asking for clarification received a vague response. The closest I can see is that the LLM and rheumatologist both proposed “an autoimmune condition” around a similar time, though the answer about which came first is also unclear.
The catch is that for the author to even have a rheumatologist referral to begin with, some previous combination of directors would have had to have suspected an autoimmune possibility and created some documentation to that effect. Those documents were supposedly fed into the tool, which obviously would have surfaced some of the data and possibilities that led to a rheumatologist referral.
So I agree. The way this tool is being presented as diagnosing a mystery illness by itself feels very misleading. I saw this spammed across multiple subreddits last week when the author was trying to promote it. I think maybe they set out trying to get attention on their AI side project on their GitHub profile, but got a little carried away with the angle that it was diagnosing medical mysteries.
This is a good point. Even when a diagnosis is obvious, a doctor may not say it outright and instead make a referral to a specialist because it’s not their specialty to diagnose that condition. This is a good thing because they’re not that specialist so they could be wrong about it, and if they are right, then they should make that referral anyway - and end result is the same i.e. a referral.
I think the issue is people might assume it's going to spit out your exact diagnosis 100% of the time right off the bat. That's obviously not true, but that's not what I got from reading the description of the project. It sounds more like a potential useful tool for folks to have as an additional option, and then to bring that information to a doctor to see if it pans out.
Regardless, one extremely helpful thing I got out of reading the reddit thread was the mention of an unrelated project called Fasten, which will apparently allow you to request all your medical records from many sources and then self host them:
Fasten is an open-source, self-hosted, personal/family electronic medical record aggregator, designed to integrate with 100,000's of insurances/hospitals/clinics
That's a massive "if" that the rest of your argument hinges on. You can't train your own LLM on your own data. You can fine tune it's responses to be more in line with the format or style you want. But this is usually used as a cost saving exercise when you are feeding s big system or inital prompt for lots of queries.
> there's a decent probability it could give some perspective you might not be able to get elsewhere
There's also a decent probability that it will give a wildly incorrect perspective that influences how you approach your possible disease in an unproductive way.
The author has no idea how accurate the tool is. They list no effectiveness measures. They just talk about the cool things it does (again, without accuracy) and their subjective experience with it.
It's the same pattern with LLM based tools we see so often: someone gets lucky with a useful result that no one else has done yet, they assume it will work just as well for all inputs, they build an app around it. You _have_ to measure effectiveness across a wide variety and quantity of inputs because of how inconsistent LLMs still are.
Incorrect self diagnosis is a big problem for doctors. Patients will show up convinced they have one condition and reject their doctors’ attempts to explore other explanations.
It’s a big problem right now with TikTok self diagnoses. People view a lot of TikToks and think they know better than their doctor for themselves or, worse, their kids.
The easiest example is when people decide they have a bacterial infection when they really have a common cold. They will shop around until they find a doctor who is a pushover and will prescribe antibiotics on demand. This is a big problem right now with antibiotic overprescription.
The problem with this specific post, though, is that it’s misleading. The author tries to present this tool as if it did a better job than $100K of medical testing and 5 years of doctors, but when you read the details it’s clear that the results from all of those doctors and medical testing were put into the tool.
Many of the comments here and on Reddit are missing this key point: The tool was mirroring the diagnosis from the doctors because it was given the medical records from those doctors, not because it was diagnosing the symptoms from a clean slate.
This doesn't mean I'm dismissing AI for diagnostics outright, but multiple good doctors have confirmed to me that many auto-immune diseases are fiendishly hard to diagnose as diseases in and of themselves vs. inflammation downstream from some other condition. The main problem isn't medical incompetence or apathy.
Autoimmune diagnosis is broken, and it IS a systems problem. The hospital system itself incentivizes compartmentalization of your different body systems to different specialists, because it allows the hospital to do less central planning and smoothes out the financial side of things (e.g. which departments do we invest in next year).
Even if the cause is found, for many diseases the treatment is just "eat healthy and exercise".
People think medicine is like fixing bugs in a code. Just a lot of tests, eventually a single cause is found, a pill is taken and the person is fixed. Reality seems to be closer to this: as you age, you accumulate damage everywhere. Eventually this damage is going to kill you, but before that it just makes you a bit slower and more tired. Sometimes, if you're lucky, some of this damage can be reversed, but mostly not.
Tell me you don't know anyone with an autoimmune disease, without telling me you don't know anyone with an autoimmune disease!
This is exactly the experience of a lot of people, my wife included. It took years to finally get a diagnosis, partly because it took her a long time to convince herself that it wasn't all in her head (much of which was due to gaslighting and general incompetence of doctors), and then once she was convinced she had a problem, finding a doctor who would actually take her seriously was a total mission. There is an enormous class of poorly-understood conditions which result in weird vague symptoms that come and go, and could be explained by many different factors. These symptoms can really destroy your quality of life, but it can be almost impossible to get them diagnosed, and there are often poor treatment options even if you do.
I really try not to take my health for granted now. We are well off financially, but it doesn't matter how much money we have, there's nothing anyone can do about the sort of condition my wife has, at least currently.
My dad has rheumatoid arthritis, my grandmother had rheumatoid arthritis and died from leukemia likely caused by treatment for it. The same grandmother's brother died of an autoimmune lung disease. My favorite uncle has been living with MS for 20 years. I have a family friend with Lupus. And a friend with Type-1 diabetes.
I guess looking at your list they're mostly the better-known autoimmune conditions (except for Lupus and perhaps the lung one), and diagnosis might have been easier.
My mom has a non-autoimmune related health issue (Costochondritis) that has been difficult to diagnose and she's been dealing with that for the last 7 years or so. It sucks having a health problem that no one can diagnose. Even in the case where you don't have a mechanical issue, the pain is still real.
My concern is that for every person with a rare disease causing vague symptoms, there are more people with similar vague symptoms that don't have an underlying rare disease. When you tell people things like "keep going to new doctors until you find one that believes you", or "do your own research", or "use my AI tool that I diagnosed myself with", those people who don't have an underlying disease will do those things too.
And if they do enough tests, or see enough doctors, they'll eventually get a diagnosis. If they see enough doctors, and push hard enough, they'll eventually find a quack who will do all kinds of procedures on them.
One of my mom's friends was convinced she had endometriosis, she doctor shopped for years until she found someone willing to do surgery in a city several hours a way. She still suffers from complications from the surgery, and her other symptoms are no better off.
I know of a kid who came very close to dying from an infection because his parents were convinced that he had chronic Lyme and they doctor shopped until they found someone willing (but not qualified) to install a central line to administer antibiotics.
Even if you have an underlying condition, doctor shopping and doing your own research is no guarantee that your eventual diagnosis is the right one.
I'm not saying you shouldn't advocate for yourself. Nearly everything is a continuum, but when you cross the line into "Doctors are idiots, only I can diagnose myself territory", you're headed into dangerous waters. Educating yourself is fine, diagnosing yourself is the problem.
This is a problem, for sure, and additionally the people with vague (perhaps even imagined) symptoms probably vastly outnumber the ones with legit but weird conditions. But for the people with those conditions who can't find a doctor who will take them seriously, what are they supposed to do? I know a few of them, and I'm unwilling to basically throw them under the bus for the greater good. IMO (going back to OP), if LLMs give them a few options that they can go back to a doctor and ask "Could it be this? How can we rule that out?" could be really life-changing.
Is the average person in such bad health that they believe you're heading downhill in your late twenties??? In most sports this is peak athleticism.
I'm in my mid forties, and obviously I'm slowing down, but if I felt those symptoms I'd be concerned, too.
I was curious so I looked up the author on LinkedIn and he appears to be in his 40s. Where did you get late 20s?
>but if I felt those symptoms I'd be concerned, to
The problem is "felt those symptoms" is subjective. One person might complain because they're a little more tired after a workout than they were in their 20s, but another person might not even notice the exact same thing.
One person might be hyper aware of every little muscle spasm and twinge of pain, another person might just ignore it.
There's actually a fair bit of evidence that depression can make people experience normal muscle contractions used to move food through your GI tract as pain. Doesn't mean it's not real pain to the person involved, but it does make non specific belly pain very unhelpful for doctors trying to diagnose underlying medical issues.
I've worked in tech support too, and at the beginning some people were just doing things wrong. I tried to give them advice, but they weren't always willing to listen, or it led to lots more work to fix the root cause. I started to learn to help them with the current problem only, and not get philosophical because the time saved could help other people.
But this gets turned around if you are the person with the problem. It turns into people helping you by treating a symptom instead of doing systemic / holistic diagnosis and treatment.
I've found lots of people who have struggled with medical conditions, only to finally get a diagnosis (and treatment!) that looking backwards was blindingly obvious.
(You were racking up the visits and bills, and each little practitioner was going through all the proper motions & procedures for his or her own little pigeonhole - so everything was going perfectly, from the system's PoV.)
Now we have 10 minutes online consultations with random doctors we'll never see again. Last time I went to mine they refused a preventative blood check, the reason? "it's too expensive and you're healthy anyways". I went to a private lab and paid for it, it ended up costing less than 10 euros, meanwhile I pay 500+ a month for health insurance.
We have far more doctors per capita than we did in 1960.
Only partly. Where I'm living (SE Michigan), one of the two huge local hospitals is owned & operated by a major research university. The other one by a major Catholic non-profit.
My (older) family & friends have been on the receiving end of lots of "the system cares only about itself, and many of the individual doctors obviously feel the same" horrors at both of those hospitals.
And it's been that way for decades, minimum.
But I guarantee you PE is still involved in that health system to a major extent. Many if not most of the doctors are going to be employees of private groups that the University hospital contracts out to.
30 years ago, when there was vastly less PE in the medical field, those hospitals had the same problems. If Dr. Fubar was socially adroit with his fellow MD's, then it really didn't matter that he was a disaster for his patients.
I never said private equity was the cost of any bungled cases despite the fact that they surely are.
What I said you can blame private equity on is this
> A few generation ago you had family doctors who knew pretty much everything about you from your birth to their retirement (or death even). Even I had that as a kid, and I'm not 40 yet.
But to be fair. It's not just PE, it's consolidation in general, which can be lead by large non-profit hospital groups as well.
.. many of whom are NHS employees, since we don't train enough locally.
https://www.nhsconfed.org/articles/immigration-harming-nhs
"Recent NHS data shows that across all NHS staff in England (1.4 million), more than 17 per cent (264,815) are from overseas. Percentages for nurses are much higher – nearly 27 per cent of NHS nurses are from outside the UK (99,856 of the 372,605 nurses and health visitors recorded).
Further, the workforce plan is supported by the government, for example, having set a target to recruit between 51,000 and 57,000 international nurses by 2023/24"
There is no shortage of doctors. There is a shortage of consultants/GPs.
Blood tests are not for curiosity or entertainment.
Better catch diabetes or cholesterol when it's not a problem yet, when the symptoms are bad enough for you to notice by yourself you're already in deep shit.
Autoimmune diseases have non-local effects, both in time and space, and thus are very hard to investigate or diagnose. They have population dynamics (e.g., as you mount a response to an infection), which means feedback effects. They're also sensitive to and productive of systemic stress and anxiety, so the feedback runs all the way to neurology, psychiatry, and environment.
In a sense we have multiple overlapping immune systems, with triggers and effectors that interplay differently in the gut, brain, lungs, and other organs. Each organ has evolved locally-appropriate systems, and are also subject to the blood/lymph-borne systems, or might be more encapsulated/resistant like brain or distal, poorly-perfused tissues.
Then it's complicated by personal history and genetics. E.g., some forms of EBV with some HLA genetics produce B-cells that happen to love a protein on the foot process of myelin; this truly does come down to a specific molecular antibody shape, now prolifically produced...
So aside from the few diseases that have single point of effect, autoimmune diseases are even more personalized than cancer (which at least is tissue-local), making research into causes even more difficult.
I don't doubt that AI might arm patients with enough possibilities to be a lot more persistent in seeking help, or that sometimes a possibility might be right. I do think that a lower threshold for analytical testing would help, but that would require a higher threshold for acting on results.
AI has been shown to be better at image recognition than radiologists, and to have more extensive differentials for primary care doctors, but both are applying implicit likelihood filters based on what can be investigated.
I'm fairly confident investing in patient-driven LLM-based diagnosis for autoimmune disease using existing tests and records is a waste of time and money.
I'm looking forward to a world where we gather more data more systematically, and use analysis to find connections that can knit together disease dynamics we can validate and block.
But the most impact would be a world where anyone can access all their health information as data, and where doctors don't spend their evenings with EPIC.
I imagine any potential user of a tool in this state would know what be wary of, but I do worry a little that LLMs have been trained on bad health advice or misinfo, and there will be very limited training data from real patient diagnoses. Even if dangerous data is minuscule in training terms, can good prompting de-risk dangerously bad advice?
I have a habit of moving around (thus changing doctors) and a lack of a history / being able to easily consolidate them has been a reoccurring pain. I recently had a full physical while in Thailand and there was some potential concern around my labs & some imaging --- of course, without history it was a "check again in 6 months!" which prompted me to capitalize on Gemini's PDF parsing abilities...
I still have work to do, but it's amazing what you can do in just a few hours now: https://health.martinamps.com
If you included some genetic information like race or national origin, this could help people learn about the genetics behind the disorders.
I have a connective tissue disorder that presents as Marfan syndrome, but I went to a geneticist at UCSF back in the 10's. He looked at Marfan's, Loeys-Dietz, Lujan-Fryns, Ehlers-Danlos, and a bunch of other disorders that would explain any part of my physical symptoms, my cardiovascular problems, and my skeletal issues - he found nothing.
If you are still looking for an answer, try again. Genetic tests (in some areas) have improved significantly in the past decade and you may get an answer now.
Unless you know how to act dying you'll just get the usual:
* Just take some paracetamol/ibrufen and come back in a few weeks * That's anxiety * Do you drink coffee? * You just getting old
And even if you get referred for something the waiting lines in the UK it can be months or years
But there is no doubt a lot of people has started to complaint about strange health issues since 2020 and nothing seems being done about it
Then media/fact checkers/health authorities are surprised if people are diagnosing themselves with Dr. Google. And buying the latest drug online recommended by twitter
I can export all of the records in to a single pdf - I airdrop that PDF to my Mac.
Is that single pdf importable? And does the system know enough that this pdf represents multiple tests/labs/blood pressure readings/letters/etc?
Otherwise to download each individually, I have to go through each record, one by one, and hit download.
There are way too many. I hope this supports uploading a single, all inclusive pdf, of all visits/tests/etc?
That said, self-diagnosis using a AI tool makes me cringe.
As an engineer, I know I could use this properly the same way OP did. As in, I would look at what the AI is telling me and triple check with all the information I can get from actual medical reference books. Also, I know I am fairy good at not making up symptoms not to match some diagnosis.
But I would never put this in the hands of people who visit webMD twice a week only to conclude that they probably have toe cancer.
And because UI matters. AI chat seems more tailored to your own problem since your giving it your inputs. And AI chat is more authoritative, cause it mimics the way doctors talk.
This sounds like a really good use of AI in its current form, and is a great case study for how one-offs can be made.
But until it fully runs locally, it shouldn't used by anyone to upload their PHI into (unless you anonymize beforehand). And definitely shouldn't allow use with commercial AI systems (ChatGPT, Claude, and certainly not Deepseek). I would actually be surprised if any of them are HIPAA-compliant.
I’ve performed a test with the simple questions, "what are the genetic candidates for hEDS?" followed by "where did you read about TNXB as a cause for hEDS?"
My hypothesis is that heterozygous TNXB SNPs beyond the known clEDS variants cause a substantial subset of hEDS which occur on an autosomal dominant basis. Most research considers clEDS far too rare (1 in 1 million) to be a significant factor in hEDS diagnosis, and heterozygous TNXB SNPs is considered far too common to be a cause for hEDS. I think it is the hEDS classification that is wrong and hEDS is indeed common, just rarely diagnosed. Enough to have causal heterozygous TNXB SNPs.
In testing OpenAI, Gemini, and Deepseek all give TNXB as the first answer for hEDS candidate genes. When pressing for references most give normal clEDS references which suggest that the LLMs are forgetting that clEDS is not equal to hEDS and clEDS as defined is extremely rare. When able to give specific references for heterozygous TNXB and hEDS OpenAI and Deepseek only gave Zewers, M. C., et al. (2003) https://pmc.ncbi.nlm.nih.gov/articles/PMC1180584/. Gemeni did not give such references.
DeepSeek did use the HEDGE study as a reference in support of heterozygous TNXB for hEDS but the actual text from the source is “Earlier research trying to answer this question suggested that a gene known as TNXB may be responsible for both some cases of hEDS and a different type of EDS known as classic-like EDS (clEDS). It is likely that this is true, but this has proven to be a difficult gene to understand, and at best it represents a very small percentage of hEDS patients.” (https://www.ehlers-danlos.com/heds-hsd-research/#16940818260...) Possibly again in reference to the original Zewers paper.
It could be that the Zewers paper is referenced by many other papers as a plausible hypothesis without providing contradictory evidence, perhaps the LLMs see this and give the paper much more weight because of that. Is the LLM incorrectly including it due to confusion with clEDS? Are doctors incorrectly excluding it for the same reason? I would love to know.
The doctor didnt give you a diagnosis, they told you a likely cause and thats because your symptoms are all over the place.
Its direcly dishonest pretending that this could help diagnose people who may or may not have real issues.
You don’t know that, the doctors don’t know that and AI certainly wont inform you it has no clue.
It wasn't until I visited a rheumatologist who looked at the combination of my symptoms and genetic test results that I learned I likely had an autoimmune condition.
I call BS on this. Sick of these AI hustlers making whatever shit up.
direct Money is the only angle in the world?
and occasionally self-gratification.
Umm, the importance of privacy was decided to prevent insurance companies from abusing your data & history to jack up your rates or refuse coverage.
>>us as individuals allowing our doctors to use existing LLMs to help them diagnose us
What prevents you from giving your human doctor permission to upload your data/history into an existing LLM to assist your diagnosis? Just fill in the HIPPA form and hand it to their administrator...
>>study shows human doctors don't effectively use LLMs even when they do have access
YUP. That is the problem; they are not yet taking as full advantage as possible. Although I have read that it is spreading in radiology where there is solid evidence the AI models do better, so it might be just a question of medical professionals being conservative and waiting for more solid data...
Doctors and administrators do not have a process for doing what you suggest. They believe the laws "we" passed don't allow them to.
No, it is not telling anything. I used it as convenient shorthand in a comment to not cite the long chain of representative govt that implemented the HIPPA laws.
>> Doctors and administrators do not have a process for doing what you suggest
Yes, they do. Literally dealing with it right now in caring for an aging parent. Yes, some doctor offices/practices are a real pain in the arse about it and do what they can to not cooperate. Others are very efficient and helpful. I understand your frustration about the awful ones, but the rules have little to do with it, they'd likely be just as obstinate without the rules designed to constrain the abuses of insurance companies.
And yes, I've seen physicians dodge the requirements and not release info when you've given all the legal permissions required, but that is on them, not on the regs. They're abusing the regs for their convenience or some other purpose, and it is damn frustrating.
BUT, if your doctor does not want to use an LLM, there is indeed little you can do to convince him/her.
On top of that, you would be utterly foolish to try to force the issue and make them use it when they are both unwilling and unskilled.
As pretty much everyone here knows, all LLMs can make significant and well-disguised errors whilst sounding highly confident in the veracity of their answer. It takes an expert to validate any serious response, and experience with the LLMs and the types of errors they hallucinate counts a lot.
Seems you'd be better off either getting your entire medical history and feeding it in yourself if you aren't concerned abt uploading it. Or, seeking another physician who IS familiar with LLMs and is actively using them.
In any case, I wish you well in quickly finding a workable solution!
Second, that's not a process. You're referring to a form. Process is everything from the doctor being willing, to understanding what they can and can't do, to the administration supporting it.
That process, as far as I can tell, does not exist in any medical system I've encountered.
> that can instantly cure any disease or injury
> on what year this fiction would be real he says "2045-2055"
I bet my left testicle we won't even have wide spread autonomous cars by then
https://en.wikipedia.org/wiki/Gattaca
that's definitely coming and those babies aren't just lucky.