Hacker News Reader

The fact that the discipline is incredibly nuanced is precisely why I want computers doing the heavy lifting, at least when it comes to diagnosis. I find it extremely hard to believe that even the best doctors could come close to touching the accuracy of an expert system. Now, obviously there is still the necessity of a trained and skilled doctor. A doctor should always have the last decision (before the patient, that is), but the doctor should be trained to understand how expert systems work. And of course, doctors and nurses will be needed to gather information from the patient, both through medical tests and verbal questioning.

The fact is, every time I have gone to the doctor for anything other than an obvious diagnosis (e.g. I need stitches in the hand because it is cut and bleeding), I have been appalled at the experience. I experienced fairly serious chest pain at the age of 26, with no history of health problems. I quickly read online about all the things it could be, ranging in seriousness from reflux to a heart attack, and I expected a doctor to have some sort of academic methodology, or do something vaguely resembling science, but instead, they x-rayed me, said everything looks fine, and prescribed me extra strength Tylenol. I wish doctors would at least pretend that they're doing something that a walk-in x-Ray booth attached to a computer system couldn't do.

I wish doctors would at least pretend that they're doing something that a walk-in x-Ray booth attached to a computer system couldn't do.

They are doing more, but won't always feel the need to show their working for minor illness. That doctor will probably have gone through a few checklists in their head, deciding whether this could be a serious illness, and clearly came to the conclusion it was not. Presumably they also asked you a few questions about the location and nature of the pain?

Also, doctors do use computers already, when they're not sure on symptoms, need to look things up, etc. They have the equivalent of expert systems at their fingertips for when that is necessary, but most of the time it won't be.

I find it extremely hard to believe that even the best doctors could come close to touching the accuracy of an expert system.

Accuracy is not what is required, given our extremely poor and fuzzy knowledge of the human body, diagnosis and treatment is more like a tradeoff between different possibilities based on partial symptoms (not all symptoms will be obvious or noticed by the patient), tests which are not always accurate and sometimes harmful, and treatments which are not always effective and have some bad side effects.

An expert system would help with none of the above.

> Accuracy is not what is required, given our extremely poor and fuzzy knowledge of the human body, diagnosis and treatment is more like a tradeoff between different possibilities based on partial symptoms (not all symptoms will be obvious or noticed by the patient), tests which are not always accurate and sometimes harmful, and treatments which are not always effective and have some bad side effects.

I don't understand. That scenario is a perfect example of something an expert system would excel at.

Not really no. Some steps are suitable for codification in an expert system, indeed doctors already use expert systems in some cases. Some things which are hard problems and are not currently solved better by expert systems:

Translating what the patient thinks is happening into a set of symptoms, sometimes those two are far apart

Finding problems the patient came in for but is too embarrassed to talk about or is even unaware of

Explaining risk to patients in terms they will understand, and then choosing an appropriate treatment based on their often fuzzy and emotional response

Weighing patient risk against an individual patient's needs or other ailments which an expert system is not going to be able to query about or be aware of.

etc.

There's certainly a huge and growing place for expert systems in healthcare, but suggesting they can replace doctors is pretty absurd at this point in time.

If an expert system prescribed you extra strength Tylenol would you feel better? I think you are taking a too simplistic view of diagnosis and put to much faith in AI. I have also noticed Watsons good results, but I do not trust them yet, there is a lack of evidence supporting the claim that Watson can make 90% correct diagnosis.

Medical imaging is a big part of diagnosis, especially in many cancer patients. There is a huge effort to automate the segmentation of tumors, and even though algorithms often have higher inter- and intra-observability agreement than radiologists, they often fail miserably for "non-standard" tumors. A fully automated system that scans the patient, make the diagnosis and treatment plans and provides treatment is very far future.

A completely different issue is that the specific combination of a highly developed brain and opposable thumbs are sometimes necessary. Consider cardiac response teams in hospitals, but also general patient care where the ability to fluff a pillow can be just as important as placing a quick IV drop.

I won't argue that machines can't do amazing things. Technology definitely has helped us to live longer and diagnose more accurately.

I don't have any source to backup my arguments and claims, but for many older medical practitioners, including my family (we have been a medical practitioner for several generations), we feel that the younger doctors lack of confident and knee knowledge because they rely on machines. I have a hard time to put my family's words into English because they are Chinese (they are Western medicine practitioner btw), but the basic idea is that they learn to use their fingers and their sensory inputs to gather empirical information about the patients. I won't say they can diagnose a tumor at a precise location or detect the type of cancer precisely, but experts can diagnose to a very good estimate. Back then they didn't have x-ray in every hospital. Only the major one has x-ray in China and my grandma and my father had to use their fingers to determine the location of the tumor, or the type of tumor. When my dad first saw the x-ray machine back in the 70s when he was still in China, it was amazing to him what he could do with him, but he said practical experiences and hands-on experience matter.

Ask how many new practitioners can do that with confident? I supposed few. Because most of them rely on machines.

What if machine is down? What if the machine is not capable of telling the truth? It has limitation.

I am arguing that machines can't help, but doctors should never rely on machine. They need to rely on their experience and knowledge and this is what distinguish an expert from an average practitioner.

To be honest, getting people x-ray all the time is largely because of suing medical mistreatment and then good part of just being efficient (or lazy if you want to be harsh). Hospitals are scared of getting sued so they'd put everyone through an x-ray machine.

Interesting you bring up the chest pain scenario. In 2003, when I was 32, I had crippling chest pains, and, simultaneously, (this is what scared the crap out of me) my entire left arm went absolutely numb - I was certain I was going to die of a heart attack, and so I immediately called Kaiser (in hindsight, 911 would have probably made more sense, but I was overly sensitive to calling 911 for anything other than murder in progress, somebody not breathing, or house on fire).

I was patched through to an intake nurse, described my symptoms, and she asked a grand total of three questions:

"Are you having any shortness of breath" "Do you feel any nausea or light headedness" "Can you press your fingernail on your left hand, and tell me how quickly it goes from white to red"

After I answered those three questions (No, No, right away) - she then said, "It probably isn't anything, but we'll want you to come in some time this week" - and set me up for an appointment a couple days later.

Total interaction time - less than 5 minutes, and that was with an intake nurse, not even a doctor.

Clearly, an expert system had done the assessment, and had decided, based on some baysean weighing (32 year old is almost certain not to be having a heart attack) plus a few expert-tree questions, that it wasn't really an emergency.

(As it turned out, I was starting to suffer from some RSI in left arm - Doc made the assessment in 90 seconds of prodding my arm, suggested an ergonomic keyboard and less typing - His analysis of the chest pains were, "Likely stress related, nothing to be worried about.")

BTW - my doctor didn't even suggest x-rays. Just a standard stethoscope/Deep Inhale test. When I asked a physician friend of mine later, why they hand't run any tests, his answer was, "You weren't in pain when you visited, had no trouble breathing, - there were no obvious indications of distress, you are 32 years old - there was very low return on running tests, and doctors (particularly at HMOs) are taught not to engage in "fishing" (testing for the sake of testing)

I can't see the part in your story where you found out what the chest pain was and that any form of diagnosing entity [sh|w]ould have picked it up.

That's a big part of the frustration. The doctor's conclusion was essentially that it was nothing.

You may well be right, but nevertheless, where medical care can be automated, it should be. Because otherwise medical care never gets more affordable. The economy doubles in size, wages double in the general economy, and ... doctors wages double. As long as we consider medicine an elite profession (so that we don't train an overabundance of doctors and drive the wages down, which most of us wouldn't want to do) employing medical staff will always cost some approximately fixed multiple of the median hourly wage. And medical care is and has always been too expensive to provide proper care for the entire population. If, for instance, we can automate doctors taking blood pressure readings, by giving patients their own sensor, or automate the drawn out process of step-by-step diagnosis by driving down blood cost prices so that 50%, 5% and 1% diagnoses can be explored simultaneously, or have automated machines which allow frail, elderly patients to bathe without needing a nurse, then we can really start to improve care, and decrease cost (which, in any case, will simply be reallocated to dealing with other ailments). Doctors lose nothing, but, step-by-step, the general population gains immeasurably.

Almost none of the prescriptions you offer have any measurable impact on healthcare spend. Blood pressure readings -> performed by $18 Medical Assistants Step by step diagnosis -> office visits are < 4% of total healthcare spend etc.

Now, what actually costs money is getting your spine surgery at Stanford rather than the outpatient clinic which costs more than getting physical therapy. It takes a person (often a doctor) to help guide a scared patient towards the most effective option, not the most expensive one. We, as patients, are not experts, so we choose the expensive option as the best proxy for quality. Computers can't engage in a meaningful discussion.

> Because otherwise medical care never gets more affordable.

They can be affordable but the problem is largely because of the large number of law suits hospitals might have to face, and plus the cost of buying those technology. Then you go with drugs. Those are expensive. Regulations, license, those are costly too.

And salary too, very high. One might argue that doctors deserve to be paid more if they can do the job properly, saving people's life. I won't get into that, but unless you can cut drug $$ down, salary down and law suit down, or at least two of them, you won't get very far from affordable healthcare.

You might ask why isn't Chinese medicine expensive.

You're right of course that inflated drug prices and lawsuits are significant costs, but, for instance, the NHS analyses all proposed pharmaceuticals for cost-effectiveness, and barters with the manufacturers at a national level, and medical litigation in the UK is almost nil. The system is more efficient, but there is a base-line which is impermeable. The only way to go below that is to train many more doctors, and start paying them something closer to an average wage, but, firstly, they'd just move abroad, and secondly, you would not be able to persuade the candidates to train for ten years if they didn't earn a decent professional wage. The only other option is to move some of the burden onto technology.

> And medical care is and has always been too expensive to provide proper care for the entire population.

Which countries are you talking about here? (I agree with your general sentiment, though.)

I know there are issues which are specific to the States, but I would say this applies almost everywhere. Almost everywhere doctors are paid 5-10 times the median wage, because they are chosen from the top few percent of students, and that difference in wages introduces an inherent asymmetry in how many hours of work are necessary to buy an hour of a doctor's time. The majority of the population, certainly as people age, have low level health issues which they manage imperfectly. You don't exercise enough because your knee is a bit gammy (perhaps requiring physiotherapy or knee surgery), your cholesterol is high, cancer is spotted six months later than necessary because you thought your dodgy stomach or sore throat was just something minor, etc, etc. And, you know, even in countries with good health provision for the poor, the middle-class still get much better care, because they talk to their doctor friends, know how the system works, research their options thoroughly, and so on. Perfectly reasonable behaviour, but the fact that it is necessary shows up the gap and the inherent limits in provision. The cheaper healthcare is, the higher the baseline and the average can rise.

Oh, sure, there's an infinite amount of (generalized) health care people might want to consume if the prices are low enough.

You can get there either by lowering the pay of health care people, or by making them more productive per humman-hour, eg via automation or just better business practices.

There are two prices of health care I have to pay. One is the up front cash (or monthly insurance) cost I have to pay, the other is the massive time sink and general pain and inconvenience it is to get it.

I live in country where the cash cost is basically zero, yet I consume as little health care as possible due to the high second cost. In fact the times I do need health care I often opt to pay more up front to avoid as much of the second cost as possible.

As long as you keep the second high enough the demand for health care won't go to infinity even as the first cost goes to zero.

Yes, the demand will only go up in relation to the total cost. (I am not sure making people wait and introducing hassle is a good way to regulate demand, though.)

Whether evidence is easy or hard to gather is a red herring. Patients present a body of evidence. Small or large, that evidence is what doctors and machines both use to generate diagnoses. The places where machines have the potential to excel is situating the evidence within a broader universe of evidence.

The human body is amazingly complex, and it's unfair to hold it against human doctors that they don't have the capabilities to use the enormous amount of information we've collected on it so far. Machines do, though, and they're already on the brink of matching or exceeding human capabilities.

It's true that medicine is more than generating diagnoses. But the years of training and huge personnel costs are caused by the advanced levels of knowledge required to perform with competence. The deprofessionalization of medicine means you can substitute expensive labor with much cheaper labor.

There's also the value that comes from standardization. The best human doctors are probably going to surpass machine doctors in their chosen field of expertise for awhile yet. But very few people have access to the best human doctors, and the best human doctor in one area of specializaton is unlikely to be exceptional in areas outside his or her specialization. The proper comparison isn't against the best doctors, or even the average doctors: it's against the worst doctors who make terrible errors in judgment and doctors working outside their area of expertise, whose work will be most easily replaceable by Watson-like software.

>Whether evidence is easy or hard to gather is a red herring.

No it's not, it's incredibly relevant. How can you take an evidence-based approach to newborn and infant care when no studies have been done? How do you run studies ethically on infants who can't give consent? Are you willing to subject a baby to an untested delivery procedure, that could potentially go very very wrong, or would you rather stick with the devil you know?

>Patients present a body of evidence. Small or large, that evidence is what doctors and machines both use to generate diagnoses. The places where machines have the potential to excel is situating the evidence within a broader universe of evidence.

Such was my point; that a human doctor or nurse, supplemented with computer analysis, will inevitably be a better care provider than either human or computer alone.

> How can you take an evidence-based approach to newborn and infant care when no studies have been done?

What do you call the alternative? And why can't computers be trained to do it, too?

You are overestimating the skills of an average doctor. I'm sure your spouse is very skilled, but there have been other success stories facts besides lung cancer when it comes to the diagnosis and treatment http://arxiv.org/ftp/arxiv/papers/1301/1301.2158.pdf: "The results shown here demonstrate the feasibility of such an approach relative to human decision-making performance. Even in its early stages of development, such an AI framework easily outperforms the current treatment-as-usual case-rate/fee-for-service models of healthcare."

I'm not sure what the difficulties in gathering additional evidence do to support your argument? The doctors have huge difficulties even with following explicit guidelines and existing studies.

I agree. I think this is the same kind of narrow minded mistake that AI made by focusing on getting computers to play chess instead of realising that the hard problem was really building something that could even recognise a single chess piece in different lighting conditions.

A lot of medicine is about perception. Obviously there is a huge amount of physical perception and pattern recognition but also, for example, a brilliant doctor who is not from the same culture as the patient will often miss something vital when taking a history from that patient that a local doctor will pick up straight away. These things are complex and subtle. We need to recognise just how limited technology is if we ever want to really improve it instead of getting too exited about the things it can do.