How many patients do doctors kill by accident?

Paul Kedrosky writes:

There is a longstanding debate in the medical community about how many patients they kill by accident. There are many estimates, all fairly harrowing, but little overall agreement. It’s coming to a boil again, and I’m wondering if you’ve ever looked at the underlying claims and statistical data here.

The most recent paper, Strengthening the Medical Error “Meme Pool”, by Benjamin Mazer and Chadi Nabhan, seems to have a somewhat bizarre argument, that absence of evidence should be evidence of absence, and that “extrapolating” from small samples shouldn’t be allowed given how bad doctors are at determining the actual cause of death.

My reply:

I’m not sure what to think. I’m somewhat sympathetic to the argument presented in that article, although I think the whole “meme” thing adds zero to the value of their discussion. I think the real issue here is that we’ll need some clear definition of “preventable medical error” before talking about their rates.

I do remember a few years ago writing about a ridiculous claim that was made by a data scientist on a similar topic. The data scientist claimed that approximately 75 people a year were dying “in a certain smallish town because of the lack of information flow between the hospital’s emergency room and the nearby mental health clinic.” I don’t know if they guy was making this up, or what, but the “certain smallish town” thing really irritated me because it implied some specific knowledge, but the numbers didn’t make sense. The “smallish town” played the same role in this fake-statistics story as the “friend of a friend” plays in traditional urban folklore.

There are certain things you can say that are automatic crowd-pleasers. One such thing is anything against the U.S. health care system. There’s a lot of things to hate about the U.S. health care system but that doesn’t mean we should believe numbers that people just make up.

24 thoughts on “How many patients do doctors kill by accident?

    • Its cool that the linked blog post exposes sloppy work.

      OTOH, it claims implicitly that criticism is bad because it provides fuel to enemies. That’s a great way to suppress all criticism and ensure that problems never get fixed.

    • 0.77% deaths per medical facility interaction still needs to be mapped to a mortality rate per population. 3rd leading causing of death doesn’t sound outrageous. Maybe someone here has a better feel for this mapping, I’m using what I call “engineering judgement”, an intuition based on knowledge and experience which is probably closer to pure intuition in my case.

    • I love how he just confidantly proclaims it isn’t based on similar quality of evidence as the thing he’s debunking.

      The truth is that medical errors and their consequences are not tracked well, and they should be.

      • Like it or not, the system that tracks medical errors is the legal system. So, if you want a measure of medical errors (that cause injury), you should look to how much the medical industry has paid out in malpractice claims. Those claims are vetted and reflect the uncertainty of determining what constitutes an error. I don’t know if those numbers are available publicly. But, the amount paid in claims would be better than most of these attempts to statistically estimate the number of errors which never involve anyone lookIng at the individual case.

        • Sorry to be disagreeable, but this idea was debunked years ago. The medical malpractice system in the US fails miserably at compensating people who are hurt by medical malpractice. There’s much more malpractice than there is malpractice payout. (There’s a book called, if memory serves, something like “The Malpractice Myth”.)

          Preventable medical error is, I’d think, a different beast from malpractice. Maybe that’s too subtle an argument, though.

          Whatever, my father went into the ER for a problem that’s a fairly common side effect of one of the medications he was on, and didn’t survive the experience. The same hospital dropped my mother out of a bed. My intuition is that there’s a lot more unnecessary medical deaths than those researchers found.

          For example, there was an NPR report on maternal mortality in the US, and it’s way higher than it should be. My reading of it was that the places where such deaths are higher are not doing the basic, well-known, things that prevent such deaths. It seems that “care” is grossly lacking in US medical care. (Another example: My PCP here in Tokyo claims that deaths (per 100,000 population) from influenza and its complications are way higher in the US than Japan. I haven’t checked the numbers, though.)

  1. I agree about the importance of the definition. Saying that a doctor “killed” a patient implies a very active role in that outcome. As in, whoops I slipped and my scalpel punctured her heart. My intuition is that this type of scenario is pretty rare.

    The other scenario is that there were a series of actions which could have been plausibly taken given the information available at the time that would have saved the patients life (at least for a time). I could very well imagine that if you counted all of these they would add up to the third leading cause of death.

    • Yes, I think it’s the latter. Things like “if you had started this person on antibiotics for the pneumonia while waiting for the sputum culture to come back, they’d probably be alive today, but because you waited, they died before the culture even came back…”

      Or whatever. Could be things like “someone prescribed drug X even though it’s listed right there on the chart that the person has adverse reactions to drug X”, or “the current standard of care for thing Y is combined A and B but you gave the old recommendation C instead and that has 35% lower success rate” or “someone switched the X-rays of the patients lungs and you did an operation to remove a non-existent tumor, the aftermath of which is that they contracted pneumonia in the hospital and died”

      all of these kinds of things *do* happen, when you put them all together is it so crazy to imagine that they happen in 0.7% of hospital admissions, and that this leads to 10% of US deaths (see numbers elsewhere in a post that hopefully comes out of the spam filter soon).

      • Good points. Just in my own experience combined with that of friends and relatives, medical misjudgments seem not uncommon (e.g., failing to recognize jaundice). Fortunately, none of them lead to death, but they are still troubling.

  2. The thing that always bothered me about this topic was the lack of specifics. What mistakes are being made? Too many transfusions or too few; too much medicine or too little? Without some specifics, there are no measures to take. We need studies that focus closely on problems. Transfusion medicine is actually an example of improvement from focused study. Wrong patient transfusion was a problem that occurred once in every 5000 when the nurse would hang the blood on the wrong patient through misidentification usually in a multi patient hospital room. Armbands on the patients came in and reduced this to about 1/500,000 as well as wrong medication administration. The residual errors appeared due to human factors in reading long strings of i.d. numbers. Bar codes and bar code readers have cut these errors by a order of magnitude.
    I acknowledge that mistakes are made. I made my share of mistakes. Yes, there are institutional barriers that allow mistakes like favoritism. However, close scrutiny rather than a broad brush is needed. For example, let’s look at flu hospitalizations this winter; was respiratory support started appropriately in these patients? If we just throw around numbers without specifics, people, and doctors are people, will respond by protecting their turf. If a study finds that intubation for respiratory ventilation should be started sooner/later than people can take action.
    We live in a world where blaming someone is more important than solving problems.

    • I thought the original paper was referring to stuff like leaving sponges inside patients after surgery, administering the wrong treatment, etc. Specifically “avoidable” medical errors.

      There is a whole other class of errors where the standard practice is just wrong. Stuff like causing colon cancer with colonoscopies, causing a skin cancer epidemic by telling people to wear sunscreen that prevents sunburn (UVB) but not DNA damage (UVA) until like 5-10 years ago, recommending people eat a high carb diet via the food pyramid, etc.

      I would guess that bad medical advice is in fact the #1 cause of premature mortality in the US.

    • > The thing that always bothered me about this topic was the lack of specifics

      If they were mostly the same error over and over again, then this could be a fine complaint, but if it’s thousands of individual errors then there’s nothing specifically to take action on… it’s just “Be more careful”… Sometimes someone gives a drug with a known bad reaction (it’s already in the chart). Other times, it’s the wrong dose of a medication, other times it’s a failure to properly sterilize an instrument, other times someone ramps up the CT scan machine way too high dose (this was going on at UCLA for quite a while if I remember correctly). Other times it’s switching the meds between patients accidentally, or giving a combination of 3 or 4 meds which interact poorly… etc etc etc

      • The trouble with “be more careful “ is that it doesn’t call for any measurable process modification. Here is an example of a specific action:
        https://sciencebasedmedicine.org/inappropriate-antibiotics-use-is-rampant-in-u-s-pediatric-hospitals/
        A antibiotic usage monitoring system is recommended.
        BTW, I remember overdosing of radiation therapy in a center that did not adjust their linear accelerator for altitude. Mistakes get made, and I believe that action to processes are very useful.

        • Sure, I agree with you that “be more careful” isn’t enough, but basically what “be more careful” looks like is you have lots of checklists and cross checks between different groups and involve the patients when possible. you use computer systems that automatically check dosages and interactions. You continue to remind the employees that taking a little extra time is critical because it’s how we avoid killing people accidentally. You intentionally create processes that slow people down on purpose (obviously unless you’ve got an emergency code situation). For example you have to enter all the drugs the patient is going to get first… Then the computer cross checks them for interactions and dosage sanity. Then it prints out barcodes that also have the patient name on it and dispenses the vial. The nurse has to stick the barcode on the vial before pressing to get the next vial. If they press too quickly it simply ignores the press.

          The nurse reads the names and sticks the barcodes on the vial. Once with the patient, you have to scan that barcode and have the patient read off their name from the barcode scanner machine and audibly confirm their name before injecting the drug.

          Every morning at 5am someone should be in charge of coming into the CT scanner machine and placing a standard “dummy head” or something into the machine. The dummy head has a radiation dosimeter in it. The tech runs a scan on the machine, the dosimeter reads off the absorbed dose and displays it as a fraction of the expected dose. Three different people need to read off the dose and type it into a log. If the dose is outside the range of say 0.95 to 1.05 then the machine shuts down until a technician and manager comes in and each personally signs off that the machine has been recalibrated. A proximity key card is required to unlock the machine. The technicians/managers can be held personally criminally liable if it’s found that they falsified evidence… Things like that.

          There is in fact an entire field known as operations research that studies these kinds of things. How many hospitals employing upwards of 1000 people have even 10 operations research / industrial safety experts on staff with sufficient organizational power to get things done? I’d like to know. Hopefully it’s more than I think.

        • “How many hospitals employing upwards of 1000 people have even 10 operations research / industrial safety experts on staff with sufficient organizational power to get things done? ”

          Good question!

  3. It’s useful to just look at some rawish numbers: https://www.cdc.gov/nchs/fastats/deaths.htm

    (note, below I’m using 123e4 to mean 123 x 10^4 in case someone isn’t familiar with floating point computer notation, also “k” means x 1000)

    Total deaths in 2017 = 2.8e6 which I think is probably a pretty good number. The CDC is supposed to get every death certificate for every death in the US, even if there are some administrative goofs it’s probably getting 99% of them.

    Then we break down by leading causes of death:

    Heart disease = 647k
    Cancer = 599k
    Accidents = 170k
    Chronic respiratory = 160k
    Stroke = 146k
    Alzheimers = 121k
    Diabetes = 84k
    Influenza/Pneumonia = 56k
    Nephritis = 51k
    Suicide = 47k

    Let’s assume that these broad causes are relatively well established, though some people might succumb to more than one (like a cancer patient who contracts pneumonia or something).

    Let’s just back of the envelope give doctors a 10% share of anything that is usually treated in a hospital… So I come up with heart disease, cancer, accidents, respiratory, stroke, diabetes, pneumonia, nephritis… (rounding)

    65k + 60k + 17k + 16k + 15k + 8k + 6k + 5k = 192k/yr

    So, to get the magnitude of numbers claimed we need about 10% of deaths to involve a medical error. Now, obviously some people will live longer than others, so the better way to handle this is QALYs or something… If you are 80 and have a severe stroke and don’t die in the hospital, perhaps you would be expected to die within a few weeks in hospice, whereas if you have an accident and are 40 and don’t die in the hospital you might be expected to live 40 more years, which is about a factor of 1000 difference… so you can’t gloss over that.

    Next, let’s take a look at total hospital admissions: https://www.aha.org/statistics/fast-facts-us-hospitals

    36e6 admissions to hospitals… so 192e3/36e6 = 0.005

    So, if half a percent of people who come into the hospital die involving a medical error, we get the number where 10% of deaths involve a medical error.

    This turns out to be right in line with the order of magnitude from the article linked above where he quotes 0.7% (=0.007 fraction) of patients have preventable harm leading to death.

    So, I for one think that yes something like 100k to 200k people in the US die each year involving preventable medical death *is plausible* but that this doesn’t necessarily mean that say 10% of QALYs lost are due to medical error (because for example cancer patients don’t necessarily have long life expectancies when they enter the hospital)

    • > So, if half a percent of people who come into the hospital die involving a medical error

      Seems high compared to the number of people who die at hospital which seems to be about 2% of admissions. I think your point about QUALYs is a really good one though. You’d really have to do the analysis in terms of QUYLs or something like it to get a sense of how big of an issue medical errors are.

  4. Giving the wrong medicine or the wrong quantity, hey, this ties in with that movie Andrew mentioned recently, “Knives Out”. It’s a ton of fun, and hinges on a plot point involving two nearly identical-looking vials of medicine.

    Here are a few thoughts in no particular order.
    1. Aviation safety took a big step forward when ‘pilot error’ became a disfavored way of characterizing what caused a crash. Donald Norman gives an example of a small plane that had two identical levers that were placed next to each other, one for controlling propeller pitch and the other for controlling the throttle. Sometimes the pilot would intend to move one but would instead move the other. Sure, pilot error, they moved the wrong lever. But it’s really a design flaw. There was a good reason for the levers to be next to each other, but no reason for them to be identical. The manufacturer made some small change, like putting a round knob on the top of one of them and square on the other, something like that, and the problem was greatly reduced. Relatedly, Norman noted with approval that at a nuclear power plant he had visited the technicians had replaced a couple of levers with beer taps, one Budweiser and one Heineken or something, to reduce the risk of pulling the wrong one. I have a feeling a lot of ‘doctor errors’ and ‘nurse errors’ could be better thought of as poor processes or poor design.

    2. As most commenters have already suggested, ‘medical error’ is way to vague. Is misdiagnosis a ‘medical error’? Surely not always, but it can be. Big gray area. On the other hand, getting sick because you visit a doctor’s office or hospital is a real thing: there are lots of germs in those places! Telling an immunocompromised patient “you should come in for a checkout” could be a fatal decision, but would we (ever) call it a medical error?

    3. In spite of the challenges of studying the issue, it’s a good issue to study.

    • “I have a feeling a lot of ‘doctor errors’ and ‘nurse errors’ could be better thought of as poor processes or poor design.”

      I agree.

      “Telling an immunocompromised patient “you should come in for a checkout” could be a fatal decision, but would we (ever) call it a medical error?”

      If a physician did the telling, I’d say it was a medical error.

      • Nosocomial infections are also problematic. Determining point origins can be difficult in a hospital with immunocompromised patients as they are sicker to begin with. That’s where we hope to have enough data points to do good epidemiology but sometimes it’s just because the patient was too sick to fight off opportunistic bacteria/viruses that exist just about everywhere.

    • I have been a fan of Norman’s books for years. I missed the beer taps but vividly remember his description op pilots armed with tools msuch as post-it notes and Styrofoam cups preparing for flight.

      Whenever I see those little brown medical vials for injectionable drugs I shudder. Modern beer bottles look safer.

Leave a Reply

Your email address will not be published. Required fields are marked *