Jul 27 2017

Computer-Assisted Diagnosis

It is very disheartening for me to listen to the political discussions surrounding health care. I can’t help thinking that the proposed changes amount to rearranging the furniture on the deck of the Titanic.

This is not to say that there aren’t important policy decisions at stake. It is important that everyone has health coverage, as a matter of efficiency and just compassion. However, the health care debate is often framed as an attempt to reduce health care costs. This is where the rearranging the deck furniture metaphor is apt.

There are some minor efficiencies to be gained in how we pay for health care, but that is not going to touch the real driver of rising health care costs – technology. Of course there are multiple factors, but the main one is the fact that we have the technology to deliver more care and more expensive care, and to keep people alive longer. This is combined with a culture that demands expensive care. We want MRI scans for every ailment, and the most aggressive procedures to keep our loved ones going for as long as possible.

More care and higher tech care costs more money.

Fixing those issues, however, is going to require tough decisions and a change in culture. Meanwhile we do need to pick the low hanging fruit, even if it is the smaller portion of rising health care costs. The low hanging fruit consists of the win-wins – changes that lead to better outcomes, cheaper care, and everyone is happy.

Medical Decision-Making

One area that everyone agrees needs to improve is medical decision-making. This consists of making the best diagnosis as early as possible, ordering just those tests that are necessary, and prescribing the optimal treatment. Of course, all physicians already aspire to this ideal. The problem is, it’s really difficult. Medicine is complex and getting more complicated as our knowledge and options increase.

The traditional method for dealing with the complexity of medicine is training, and this will always be necessary. But training and education have their limits, and it seems that we are pushing up against those limits. Further, as you engage in error reduction you get diminishing returns. It takes more and more effort to make smaller and smaller incremental improvements.

Because of our technology, errors and delays are getting more costly.

While I do think we need to optimize medical education and training, it is also clear we need other methods to optimize care and reduce error. Atul Gawande wrote about this in his excellent book, The Checklist Manifesto. He advises using systems, like checklists, to reduce error and improve quality of care. A checklist is essentially an external aid that professionals (not just doctors) can use to reduce error.  Checklists work.

There is another external aid that has the potential to have an even more profound effect on error reduction and care optimization than checklists, but has yet to come into its own – expert computer systems. These are software algorithms that can, for example, take a list of signs and symptoms and suggest possible diagnoses.

The latest issue of Scientific American has an excellent article on these systems, which is worth a read. I will expand upon their main points here.

Essentially physicians engage in two types of thinking when it comes to diagnosis and treatment – intuitive and analytical. Intuitive thinking is experience based and involves a great deal of pattern recognition. The advantage of intuitive thinking is that it is fast and can incorporate a great deal of information. The disadvantage is that it is based upon the quirky (not necessarily representative) experience of the physician, and is subject to a host of cognitive biases.

Every heuristic I have discussed here and elsewhere comes into play when thinking about patients. This is why physicians (and again, any professional) needs to be aware of critical thinking, biases, heuristics, logical fallacies, and the fallacy of memory and perceptions and take them into consideration when making medical decisions.

Analytical thinking is slower and more deliberate, but has the advantage of accounting for specific bits of information in a rigorous statistical manner. Analytical thinking in medicine is critical, but it is very difficult. It involves remembering or having access to a mountain of statistical data and knowing how to properly crunch the numbers.

The ideal clinician blends intuitive thinking and analytical thinking to take advantage of the best of both.

Expert computer systems essentially are a tool for analytical thinking, which is something that computers do much better than humans. Humans are better at intuitive thinking – recognizing a disease by the subtle way a patient looks and moves, for example. Intuitive thinking is also critical for interpreting a patient’s symptoms. Patients don’t complain to their doctors that they have appendicular ataxia. They report what they experience, and that has to be translated into medial phenomena. Further, there is a personality and cultural layer involved. How significant is a symptom, how new is it, etc. ? Because patients are people, we won’t be removing the human element of health care anytime soon.

But computers potentially kick ass at that analytical evaluation. They can sift through thousands of factors, and know what their statistical influence is on possible diagnoses. They will know what tests are necessary, and what treatments are likely to have the best outcomes. This information can then be filtered through the intuitive and personalized evaluation of the physician.

To give you an example of how powerful the analytical approach can be, when coupled with large amounts of data, play with the Akinator. With a surprisingly few questions, some of which may seem arbitrary, this computer algorithm can guess just about any character you are thinking of. It essentially takes an analytical approach to 20 questions, and is better than any human.

Now imagine playing this game except for diseases.

There are applications already in existence, DXplain, VisualDx and Isabel. They work, and studies have shown they improve diagnostic accuracy for rare or obscure diseases. For everyday common diseases, there is less room for improvement. However, even in those situations an expert system may optimize treatment, by, for example, recommending the optimal medication.

Right now we essentially are having an infrastructure issue. These systems need to get better and more comprehensive, but mainly they need to be available at every point of patient care. Further, use of expert systems needs to be incorporated more into medical training.

I would like to see such systems everywhere and routinely used. We should already be at this point, in fact. They we can explore further ways to exploit these systems to optimize care and reduce health care costs.

Again – we do need to improve how we pay for health care to optimize access and reduce waste and inefficiency. But this is small potatoes compared to optimizing the practice of medicine itself. Expert systems, when fully realized and integrated into medical practice, have a far greater potential for improving outcomes and reducing costs.

This is an infrastructure investment worth making.


25 responses so far

25 thoughts on “Computer-Assisted Diagnosis”

  1. Nidwin says:

    Certainly but you’re going to hit a wall at some point in time. The equation is simple, the more IT you integrate the more IT you’re going to need and IT at the top is extremely expensive.

    Big pharma and big medics, like many other big whatever, are at the very beginning, low-end of the IT-chain where a lot of stuff is financially and human resource wise achievable. But in no way can they afford and compete with the financial sector, one if not the biggest consumer for IT and IT folks.

    There’s just an x amount of us around the world and we go and/or stay where we’re paid and treated the best.

    The only way I see this possible for the medical world is through globalisation with a fully centralised IT department and make sure to keep your IT dudes and dudettes very very happy.

    Good article Steve.

  2. Nidwin – I agree, but there are trillions of dollars at stake in health care, investing heavily in IT is ultimately cost effective. It is 1/6 of our economy.

  3. Willy says:

    Dr. Novella, I am curious as to how you see malpractice fear driving costs as regards, say, over-testing?

  4. Defensive medicine plays a role, but it is difficult to quantify because it generally manifests as a bias in making judgment calls.

  5. Charon says:

    Such systems sound interesting and useful, but there are clearly big savings we’re missing unrelated to this – obvious by comparing to other countries. A large part of this seems to be that we pay a lot more in the US for the same things, to drug companies, hospitals, doctors, and administrators everywhere (including insurance companies). Another aspect of this is that medical pricing is an absurd nightmare of opaqueness. Fixing that is tough, since it involves paying a lot less money to a number of powerful actors.

    Also, any other profession that paid as well as being a doctor would attract a lot more people, except that we essentially cap the number of doctors artificially by the number of spots in med school. This seems problematic.

    Better analytics would help with perhaps the other big part of the cost, which is too many tests and unnecessary procedures. But analytics won’t solve that, because it’s a judgement call where to set the bar. The yes/no call depends on cost, probability of improving outcome, degree to which outcome might be improved, severity of problem, etc. And where to draw that line is a decision that must be made by humans. Computers only help us to make it in a more consistent way.

  6. hardnose says:

    I have waited a long time for someone to mention one of the real causes of the health care crisis —
    technological advances. I read an article by Paul Krugman a long time ago, where he said the same thing. But now Krugman only yells about how everyone deserves the best health care, and Republicans are cruel, blah blah blah.

    Patients expecting and demanding miracles is a big factor in rising costs. as Steve N points out.

    Another factor, which he does not mention, is the inadequate supply of MDs. We need medical schools to stop limiting the supply. Also, start using more physician assistants.

    And the idea of hanging desperately on to life, no matter how sick, hoping a cure will suddenly be found — this is not practical, and it costs a fortune. I personally would NOT hang on and “fight” an incurable disease. But lots of people are, understandably.

    As for whether expert medical systems can have a role in lowering costs and improving effectiveness — I don’t know, but I doubt it will help as much as technology lovers expect. Computers are good at statistical analysis and crunching huge amounts of data, but they have absolutely no common sense. And you can’t program common sense into them, so every decision they make will have to be checked carefully. But who knows they might be useful in some cases.

  7. tb29607 says:

    More medically centered IT would be great. The majority of doctors I talk to often type symptoms into Google when faced with a perplexing case. “Googling for a diagnosis” (JAMA 2000 I think) strikes me as highly likely to be problematic.
    I wonder how patients will respond to an algorithm telling them that no, they can’t have antibiotics for their trip over seas “just in case”. Or that the head ache their currently asymptomatic kid had a month ago is not an indication for an MRI no matter how worried they are about a tumor. Anything that helps stop this type of waste would be hugely beneficial.
    I agree completely that the USA needs to confront some uncomfortable truths about the excessive costs of allowing families to insist on squeezing every last second of life out of loved ones (who are terminal, requiring multiple modes of life support, and are likely only experiencing pain and/or air hunger if anything).
    Obviously an IT diagnostic/treatment assistant will base recs on percentages. I am curious how the inevitable malpractice claims would be handled?

  8. BillyJoe7 says:

    Well, if you’re going to rely on your common sense I don’t think you’ll have to wait for an incurable disease to kill you.

  9. JimV says:

    I agree that more IT tools would have an important impact, but from what I read, we in the USA pay about twice as much as other developed countries for about the same results (after correcting for the negative effects of our gun culture; otherwise worse results). I don’t think this is because the other countries use more IT.

    Dean Baker keeps harping on the costs of using patent-protection to pass on the costs of drug research, rather than using government-funded research, which results in high costs for prescription drugs which can be produced for pennies per pill.

    In general, an unregulated capitalist (no, I am not a communist) system in which consumers have little agency will evolve ways to concentrate wealth in the hands of a few at the expense of the many.

    An anecdotal experience which I think illustrates this: I began having trouble breathing due to accumulation of mucus in my lungs. One of the suspected causes was acid reflux, so I was given a prescription for omeprazole. I filled it at a Rite-Aid pharmacy for $2.83 per 20-mg pill. (This was before I had drug-coverage in my health insurance.) Later, I found out omeprazole is available OTC at Rite-Aid for 50 cents a pill (less when it is on-sale).

  10. MosBen says:

    HN, while I agree that that issue isn’t discussed enough, it’s important to remember that discussions about end of life care and planning were turned into ‘death panels’ during the ACA debate. I’m not going to get into the politics of that debate, but it’s always going to be a hard sell to tell people that they should consider limits on care for terminal cases and not have them hear “I want to kill your grandmother”.

  11. RC says:

    As someone who earns as pretty good living writing healthcare software, I agree with this for both selfish, and altruistic reasons.

    The healthcare industry, from a technological standpoint, is absurdly backwards. One of the projects I’ve worked on is the ACA driven push to get hospitals to use electronic prescribing. E-prescribing leads to drastically less medication errors, lower malpractice costs, reduced workload for pretty much everyone involved, and completely eliminates the problem of stolen prescription pads.

    And it was a gigantic hurdle to get doctors to actually use it.

    And its the same thing over and over and over again with them – like pulling teeth. Everything costs more than it should because you’ve got a 65 year old chief of medicine who can’t figure out his iPhone fighting you every step of the way.

  12. BillyJoe7 says:

    “play with the Akinator”

    My character was Richard Dawkins.
    Akinator guessed Stephen Hawking.
    It continued with half a dozen more questions and hit the mark.

  13. bachfiend says:

    The reason why American healthcare is so expensive is because it doesn’t have a single health insurer.

    Australia manages to have better health care at half the price because it has a single state run public insurer (in addition to private insurance which cover admissions to private hospitals and often ‘essentials’ such as naturopathy and chiropractice…).

    Extra technology, such as MRIs, might actually save money. My GP insists on ordering annual PSAs on me (I have a family history of prostatic carcinoma), which exposes me to the risk at some time I’ll have a rising PSA level and potentially be inflicted with prostatic biopsies (PSA is prostate specific, not cancer specific, and my PSA is always below the normal range, so the only way it can go – if it changes – is upwards).

    Radiologists in Australia want to have prostatic MRI included as a scheduled item on Medicare as an investigation following a raised PSA which would eliminate the need for biopsies (which are more expensive since not only is the doctor taking the biopsies need to be paid, but also the pathologist analysing them – and often the biopsies miss the tumour, if there’s one present, requiring repeated biopsies) in many cases and allow directed biopsies in others (instead of the multiple biopsies currently employed). Even without it being on the schedule, it costs around $400, which I’d happily pay.

  14. BurnOut says:

    S. Novella: “This consists of making the best diagnosis as early as possible, ordering just those tests that are necessary, and prescribing the optimal treatment. Of course, all physicians already aspire to this ideal.”

    I no longer believe this to be true. Background – I’m an in-the-trenches primary care physician, not at an academic center. I remember how shocked I was when I watched my first partner after residency prescribe zithromax for literally every patient with a runny nose.

    That was 12 years ago. I have become frustrated with the pressure our administration places upon us (financial pressure, as well as publicly airing poor patient satisfaction scores and comments). So last month, after a shift with another partner of mine, I checked – every patient with cold symptoms got antibiotics.

    It appears to me that the average physician seeing patients every 15 minutes wants to end each visit as quickly as possible while maintaining happy patients – so they give them whatever they want. Antibiotics, xanax, ambien, CT scans, labwork etc etc. (Side note: not so much opiates at any practice I’ve worked at, this has been a notable exception fortunately.)

    Maybe diagnosis/treatment assistants can turn the tide, but I am skeptical until the pressure to satisfy patient demands is removed. And to be honest, this is getting much worse.

  15. tb29607 says:

    BurnOut, well said.

    Although if you want to see opioids prescribed more than diet and exercise just visit a few emergency departments. I am still baffled that patient satisfaction scores use pain management as one of the 8 core assessments, these scores affect reimbursement, and people can’t understand why we have an opioid epidemic.

    As far as “death panels”, every nationalized health system has some mechanism for withdrawing or continuing care in cases where physicians and patients/families disagree. The costs of prolonged “heroic” care are just too obscene to be allowed when resources are not unlimited. $1.75 million on average per month for 4 months is the most expensive single patient I have personally been involved with and I bet others can give much more impressive examples. Personally, I am in favor of any system that has a way to overrule demands for continuing futile care, private or public options. Although I suspect I am in the minority.

  16. Robney says:

    I try to be impartial about this issue, but it seems to me that if the main mechanism for covering health care costs requires a profit to be generated (such as private insurance), then it may not be optimal for the end users’ costs – because their fees are covering an operating cost that otherwise would not be there. So I think there is opportunity for the US to improve its healthcare cost efficiencies.

    True, the personal financial responsibility may incentivise people to change their behaviours and thereby reduce their own healthcare costs, but this is assuming people are capable of withholding short term gratification in the interest of their long-term health – and I would say all the evidence points towards people not being good at this.

    True also that the free market might drive down health care delivery costs, but it may also introduce profit motives over health outcomes – which is not in the interest of public health or health care value for money. Besides this, how health-care services are delivered is a very different question from how they should be funded. Its possible to socialise healthcare revenue collection but privatise health-care service delivery. This would yield the efficiencies of the free market and be a more equitable system for distributing the costs (because an individual’s personal health-care costs are very much a lottery and not aligned with their income or the choices they make, personal private insurance is a very poor way of covering their health-care costs). Extending the revenue base may also reduce the average costs each individual is required to cover.

    I know I’m not saying anything original, but the US system just seems very odd to me.

  17. Robney says:


    In such a case where a reasonable threshold is met and the state withdraws care from an individual, if they family wishes to continue support, could they not take out insurance to insure against such an event so they can fund continue care should they wish to?

    This seems like a reasonable balance between personal rights and the reasonable use of public resources. Everyone is guaranteed a basic level of care but people can take out personal insurance if the basic care does not meet what they consider reasonable or ethical.

  18. bachfiend says:


    ‘It’s possible to socialise healthcare revenue but privatise healthcare service delivery’.

    That’s the Australian system, and it results in better healthcare at around half the costs.

    People also have the option of taking out private health insurance to pay for extras (in many cases, it’s enforced by the government increasing the Medicare levy on people above a certain high income if they don’t have private insurance, as applies to me).

    The private insurance companies are motivated only by profits. Periodically, Australian newspapers report cases in which their customers complain of costs they weren’t expecting. That their policies excluded conditions and necessary treatments.

    I’m not certain how your idea that families should be able to take out insurance to cover ‘heroic’ treatment of terminal illnesses would work. Who would be taking out such a policy? The beneficiary who’d suffer the ‘heroic’ treatment. Or the children of elderly parents who’d be inflicting the ‘heroic’ treatment on their parents?

    I can’t imagine that there’d be much demand for such a policy, unless the family already knows that it’s going to be necessary in their eyes. In which case, the premium would be enormous and unaffordable.

  19. BillyJoe7 says:

    Character: Stephen Novella
    Akinator: Donald Trump junior 😀

    But Akinator got it after about 20 more questions, which surprised me because the questions didn’t seem to be homing in on the answer!

  20. tb29607 says:


    Indeed, the balance you describe seems completely reasonable to me. The U.K. system is the one I hear referenced most often, with the public National Health Service for all and a separate private insurance and care market (although I have not researched this personally).

    Unfortunately funding is not the only limiting factor. Current examples (instead of the dialysis “death panels” of the Bronze Age) include organs for transplant (with the notable exception of Northern California which has no waiting list for kidneys) and many less commonly used medications and devices which are stored in small quantities despite being potentially life saving. Also the people and actual facilities needed to make these things viable (inpatient mental health beds, inpatient rehab beds, and various specialty ICU beds for infants and children come to mind due their frequent lack of availability).

    Perhaps IT assistance could help optimize utilization of these limited resources? Less inappropriate use leads to more availability? I agree the public for all with available private seems like a good compromise and the private option could make funding a nonissue. Limited resources are more commonly problematic in my practice and I would be surprised if that changes significantly between health care models.
    And I would be interested in hearing other people’s experiences, either similar or different.

  21. Robney says:

    I was raised in the UK and now live in Australia so have used both systems.

    In my fairly limited anecdotal experience, the Australian system is superior in terms of service although both have funding challenges. Blair tried to reform the UK system to something closer to the Australian system, with local health authorities tendering services to external contractors to drive down costs.

    It was my understanding that the main driver of increasing costs was an aging population, which very much goes on hand in hand with higher aged-care costs because of newer treatments/technologies. In short, a declining revenue and base compounded by an increasing demand for services and more expensive services.

    It’s kind of a microcosm of the wider economic/demographic problem that is the ratio between working and non-work age people.

  22. Nidwin says:

    “(tb29607):Perhaps IT assistance could help optimize utilization of these limited resources?”

    IT can only assist when people want IT assistance in the first place. And when you read RC’s experience with E-prescribing software it just shows the huge gap between IT and medical end-users.

    We, IT in the financial sector, are still Today constanly fighting with the business side of our industry (our end-users) because they just see us as a cost center that’s always way too expensive, always too late and not enough focussed to the real needs of our end-users, them. Whatever we achieve, often close to miracles as we also have to read business minds, it’s never good enough and always too expensive. And we’re talking here about the IT, the one that’s, functional and efficient wise, decades ahead compared to some others.

    Imagine the clash and incompatibility between IT and big medics, aint going to work till big medics wants or going to truly need our help.

  23. tb29607 says:


    I agree there is great resistance to IT in medicine. Much of that is due to older doctors who are stuck in their ways and resist change. However, my own experience with E-prescribing suggests to me that at least some blame should be assigned to the E-prescribing programs. Of the 3 programs I have used, none consistently deliver the prescription to the correct pharmacy. So now I print the prescription and give the hard copy.

    That is annoying but the biggest problem I have is that pediatric dosing for medications is rarely included in the program’s formulary. So I end up having to manually type in the needed dose which essentially voids any program prescribing error safeguards.

    And if the child requires the med in liquid form I frequently can’t enter that on the computer prescription and it will not print if I leave the medication “Form” space blank. So I have to inaccurately enter “tablet”, print the prescription, scratch out “tablet”, and write in “suspension” and the concentration. So in addition to the frustration of finding a way to make the thing print, much of the prescription is still hand written which voids that benefit as well. If I were not optimistic that the programs will eventually improve I would find the irritation and extra time (for every single prescription) impossible to justify and would go back to hand writing them.

  24. bhob says:

    What do you think about a whole body CT or MRI or both used to baseline for medical health decision making? We have comparison software built in to our X-ray systems that highlight differences in scans as a tool to the operators in the making of security decisions (it’s called ‘Easy Match’ https://www.leidos.com/products/security/vacis-xpl). Blood chemistry baseline would fall into this category also no?

  25. Ivan Grozny says:

    What is exactly the “crisis” here? The fact that people live longer and have more money to spend on their health, while technology advanced beyond anything people 30 years ago could have imagined in the wildest dreams? Instead of celebrating this a an unmistakable sign of progress of our civilization, we have now to cry that this as a “problem” or even worse a “catastrophic crisis” !

    I have a simple solution for this health care cost inflation: prohibit people from spending their money on their health – adopt a single payer system. Actually many countries already “control costs” this way: you pay “nothing” but have to wait six months to see a specialist. Per capita health care costs (the visible ones, at least) will be way down (USA spends 17% of GDP on health care, Canada just 10%: just ban legally 90% or normal private health care transactions (as we Canadians do) and you will have much less private health care transactions, and thereby much lower health care costs per capita. Basic economics).

    And second solution: prohibit any technological change in health care. Since new treatments are always more expensive in the beginning, prohibiting any new treatments will make treatments cheaper and greatly contribute to controlling the health care costs. Health care expenditures in Tanzania are much lower than in the USA since they have very little technological equipment at all.

    If you really think that health care costs are primary problem, then you should not be wasting your time tinkering along the edges with technology and with smarter use of computers: more robust ways are at hand and have already been tried in practice.

Leave a Reply