You’re a data scientist at a local hospital and you’ve been asked to present to the physicians on communicating statistical information to patients. What should you say?

Someone who wishes to remain anonymous writes:

I just read your post reflecting on crappy talks . . . I’m reaching out because I’m a data scientist at a local hospital in the US and I’ve been asked to present to our physicians about communicating statistical information to patients (e.g., how to interpret the results of a mammogram). My first pass at an outline is relying heavily on work by Gerd Gigerenzer et al. and will include a discussion about the value of expressing risk using frequencies (1 in 1,000) rather than probabilities (0.1%).

If you were in my position, what additional topics would you include in the talk? If my talk goes well, I may be asked to give more talks with similar themes. So, if you do have the time to share a few suggestions don’t be concerned that I might cram all of them into a single talk. I’m in an explore mode, so the more ideas the better!

(PS – I’m aware that sometimes these kinds of messages can end up being discussed on your blog. In the unlikely event that this message motivates a blog post, feel free to share this message, but please omit my name and Email address.)

Ummm . . . I don’t know! So I’ll throw this one out to you, the blog readers. Please help this person!

78 thoughts on “You’re a data scientist at a local hospital and you’ve been asked to present to the physicians on communicating statistical information to patients. What should you say?

  1. Whatever physicians need to tell patients, it needs to be in simple terms. Anything complicated or hard to understand will cause the patient to freeze up and not be able to take anything in.

    How about including an image with Allison Janney talking to Micky Rooney, saying “On average, women are shorter than men”? (Janney is 6 ft tall (1.82 m), Rooney is 5 ft 2 inches (1.57 m)).

  2. I’ve co-authored a couple of papers about this. I don’t think that communicating probabilities (e.g., with frequencies) is going to help much. A bigger problem is that people don’t know what to do with probabilities even if they are understood. So I agree with the first post. Some examples: “That test we did says that you may have cancer. But practically nobody your age gets that type of cancer, so we’re going to ignore it.” or “Yes, those night sweats are consistent with lymphoma. But the chance that you have lymphoma is about the same as that of your getting hit by a car when you walk out of here.” Or “This drug may have a risk of causing X as a side effect. The result was not statistically significant, but the possibility makes sense biologically.”

    There are many possible misunderstandings. I knew someone who was trying to decide whether or not to have chemo after colon cancer surgery. I looked up the research and told him the stats about 5-year survival, and the side effects of the therapy. I told him that this was all there was to know, and that he should decide. He said he would go to another doctor to get a second opinion. (He made the right decision, but possibly not for the right reasons.)

    In another study by Andrea Gurmankin-Levy (in her dissertation, which I supervised), women who chose to have a BRCA test, and who were told it was negative so that their risk was now 15% (or whatever), were then asked by the researcher “What is your risk?” Many said “50%”. They chose to have this test and then didn’t believe the results. The numbers did not do much for their anxiety that led them to get tested in the first place. (But for others they did allay their fears.)

  3. > A bigger problem is that people don’t know what to do with probabilities even if they are understood.

    Related – remember that conditional probability is hard, and does not come naturally to people.

    And it’s important to discuss the difference between absolute risk and relative risk.

    These are often foundational concepts, where (IMO) people who are familiar with working with them underestimate how fundamental they are for understanding statistical probabilities in the context of real life implications.

    Obviously, how to get them across isn’t easy – but my point is that you can’t just layer statistics on top of a poor foundation. It’s like discussing algebra with someone who doesn’t have a firm grasp on basic number theory: In one ear and out the other.

    • Good points Joshua.

      I can’t decide who is the most talented science communicator among scientists and journalists because each make assumptions that are determined by the constraints imposed by their sponsors. I see this unfolding in the evaluation of the rapid antigen tests. The debate is pretty strident due to the COVID situation in England. Jon Deeks [@Deeks on Twitter] from the University of Birmingham presented a negative appraisal of the antigen testing in Birmingham. He criticized the accuracy of the Innova Antigen Test. We hear that his university has or is developing another antigen test. So it pays to know the background for why this or that expert has his bias.

      • Sameera –

        Yes, I was actually thinking to a large degree about what I’ve encountered with people when taking to them about Antigen testing!

        I have encountered a response many times something like “Why would I take the test that is less accurate.” When I try to discuss the benefits of multiple tests, many people have a hard time grasping how the inaccuracey diminishes – because of the difficulty of conditional probability. An inaccurate test is just an inaccurate test.

        Same with talking to people about the effects of “lockdowns” where people can’t really deal with the conditional probabilities: they just glide past them when they confidently state counterfactuals of what would have been had interventions been absent.

        I have a new working theory that conditional probabilities act like a weak point in scientific questions where political biases can invade and politicize policy debates.

        People have a hard time grasping the conditional probabilities so they just jump past the conditionality and go to simple cause and effect. Thus, for example, COVID interventions “cause” harm (i.e., “lockdown deaths”) – because understanding that if interventions hadn’t been in place those same (or more) deaths might have occurred requires dealing with the conditionality.

        I have long encountered the same obstacle when talking to people about the “cost” of ACO2 mitigation.

  4. Having worked with quite a few physicians on the research side of a hospital, I would like to submit that in many scenarios it is likely that the patient has a better understanding of quantitative methods than the doctor.

    I don’t say this to besmirch the educational attainment of doctors (MDs), just to restate the problem you are facing. That is, you are not teaching doctors how to communicate something they already know/understand to patients who don’t understand; but rather, you are teaching doctors something new AND ALSO how best to communicate this to patients.

    • > Having worked with quite a few physicians on the research side of a hospital, I would like to submit that in many scenarios it is likely that the patient has a better understanding of quantitative methods than the doctor.

      That has obviously always been the case for me. Doctors are often kind of intimidated by the questions I ask, like “how would doing that test change the treatment?” or “You say the worst case scenario for this surgery is a cranial bleed, what’s the rate at which that occurs?” (they’ll often answer something like “it’s very rare” but “1 in 400 million” is very rare, but so is “1 in 100 thousand” and they’re not the same thing at all). But then there are plenty of patients in the LA area for example who are immigrants from Mexico with less than a high school education…

      So your point is very good, but I’m not sure how often the doctor is on the short end of the quantitative knowledge. It could be very rare, and yet your point is still very good. The doctors really often don’t understand the quantitative part either.

    • No, please. Don’t use the number needed to treat, instead just give the absolute risk reduction. See the discussion here and the comments section (https://errorstatistics.com/2018/07/11/s-senn-personal-perils-are-numbers-needed-to-treat-misleading-us-as-to-the-scope-for-personalised-medicine-guest-post/#comment-172214) for a couple of reasons why, and then see the dozens of other papers explaining why the NNT is not a good statistic.

      Duncan, B. W., & Olkin, I. (2005). Bias of estimates of the number needed to treat. Statistics in Medicine, 24(12), 1837–1848. [https://doi.org/10.1002/sim.2076](https://doi.org/10.1002/sim.2076)

      Hutton, J. L. (2000). Number needed to treat: properties and problems. Journal of the Royal Statistical Society: Series A (Statistics in Society), 163(3), 381–402. [https://doi.org/10.1111/1467-985X.00175](https://doi.org/10.1111/1467-985X.00175)

      Kristiansen, I. S., Gyrd-Hansen, D., Nexøe, J., & Nielsen, J. B. (2004). Goodbye NNT. Journal of Clinical Epidemiology, 57(2), 221. [https://doi.org/10.1016/j.jclinepi.2003.09.001](https://doi.org/10.1016/j.jclinepi.2003.09.001)

      Wu, L. A., & Kottke, T. E. (2001). Number needed to treat: caveat emptor. Journal of Clinical Epidemiology, 54(2), 111–116. Retrieved from [https://www.ncbi.nlm.nih.gov/pubmed/11166524](https://www.ncbi.nlm.nih.gov/pubmed/11166524)

      • You are 50 year old patient with 99.5% chance of living another year.

        The doctor says “here is this blood pressure med with ARR of 0.01.” Then your chances of living another year will increase to 99.6% if you take the drug?

        Alternatively they could tell you “the NNT for the medication is 100”. That means there is 1% chance you will be someone who benefits from taking the drug.

        The second is more accurate since the number refers to the population average rather than an individual.

        As to the other critiques, they don’t really bother me because they assume a false precision that is always missing in reality. The NNT is typically just being used as an order of magnitude. You can assume error of at least 1/2-2x whatever number is cited due to differences from study population, crappy studies, etc.

      • No layman will understand absolute risk reduction. NNT will perhaps be understood. We are speaking here about communication with patients, not about scientific precision, don’t we?

        • “Your risk of dying from this disease is X %. If you take this drug, that risk is reduced to Y %. Here are the potential side effects and other costs associated with taking the drug”.

          On what basis do you claim that no layman will understand absolute risk reduction? Do you think it’s easier to understand the abstract concept of NNT? “If we treat X people like you with this drug, then Y out of those people will be helped. We don’t know if you are among those who are helped or not”.

          As a patient I would much rather know what my risk is with and without the drug. I also think that’s much easier to relate to.

        • The study did not show risk to any given individual was reduced by one/ten/whatever percent. Some people could be harmed, others greatly benefit, etc.

          For instance, if the mean score of a group is larger than zero, this does not imply that a random individual of that group is more likely to have a positive score than a negative one (as long as there are more negative scores than positive scores an individual is more likely to have a negative score).

          https://en.m.wikipedia.org/wiki/Ecological_fallacy

        • That’s obvious, but the average effect in the study is still the expected effect of the intervention when applied to an individual patient. Of course one should also consider the uncertainty of that estimate when making a decision in each case.

          NNT is misleading in many ways. For example, how many hypertensive patients need to receive blood pressure lowering medication to prevent one fatal stroke is only interesting if preventing a fatal stroke is the only incentive. If that NNT is 100, that doesn’t mean it’s only 1% chance that each patient will benefit from taking the drug, unless benefit is restricted to preventing a fatal stroke only. Most effects of medical interventions are not dichotomous at all.

        • NNT for blood pressure meds is about 100 for all major health event outcomes I’ve seen. That is why it is ok to round to 100. If there was a relevant outcome where it was ~10 then you would use 10.

          The point is that “you have 1% chance of benefiting” conveys that the average effect applies to patients in aggregate rather than the individual. This is more accurate than “your chances of benefit increase by 1%”.

          Neither are exactly true, but NNT is better. Really the idea of applying an average result to the individual is just fallacious to begin with and is a big problem with RCTs as some kind of gold standard. But that is the type of info the research community has decided to provide doctors…

      • Excellent points and recs!
        NNT may be handy for us docs deciding what to advise or offer, but yes problematic for many reasons.
        ARR much more sensible, though sometimes requires some digging.
        Thanks for the references!

  5. It just so happened that after reading this post I began listening to the current weekly podcast on covid from Michael Osterholm, a well known and experienced epidemiologist, who gave this, from the transcript:
    “Last October, I expressed my concerns nationally on NBC’s Meet the Press. One thing I said was about how we’re communicating. I said we don’t have a consolidated one voice, we didn’t then. It caught the interest of a communications expert by the name of Randy Olson. Randy and I have become very good friends since that meeting in October. He’s a scientist turned filmmaker who has written five books on science communication. Last year, he earned an award from the American Medical Writers Association for his work over the years with organizations like the National Institutes of Health and the Centers for Disease Control and Prevention. A friend connected us and we began a collaboration that has become very important to this podcast. I cherish and value his input very much. In one discussion, I was telling him about the science when he interrupted, and quite abruptly, and said “The science is crucial, but so much of what’s happening is about communication…” So look up Randy Olson.

    • Osterhom is a great communicator.

      Lots of people are going on about science
      communication. So far, no apparent changes in public acceptance or understanding, but it is a fruitful career path

      It’s frustrating as a scientist when ppl don’t do what you want them to do. That doesn’t mean there is a communication prob. Some ppl just don’t care.

      • If you want people to listen to you, then you need to perform impressive feats anyone can understand or make surprising predictions that come true anyone can observe.

        That is how it has always been since Archimedes pulled a ship from the bay by himself with his contraptions.

  6. Why not the truth? “Most treatments increase your chances of avoiding whatever it is you’re here for by only a small amount. They do however have a very large and positive effect on our finances, and for that we thank you.”

  7. Dear anonymous DS

    The task of communicating information assumes you have generate information quality that you now need to communicate. This requires a transformation from numbers, to data (numbers with a context), to findings (through analysis), to knowledge and insights (thorough information).

    In fact, I believe you task is to show how to generate information quality. Communication is one of the eight dimensions that affect it. Operationalization is another. The ability to generalize is yet another. The Real Work of Data Science is a (short) book about the role of data scientists in organizations. It gives you a checklist and lots of examples, not necessarily from healthcare: https://www.wiley.com/en-us/The+Real+Work+of+Data+Science%3A+Turning+data+into+information%2C+better+decisions%2C+and+stronger+organizations-p-9781119570707#:~:text=The%20Real%20Work%20of%20Data%20Science%20explores%20understanding%20the%20problems,helping%20companies%20become%20data%2Ddriven.

  8. When I was a young doctor, I was often frustrated by presenting a clear, factual discussion that somehow led the patient in a direction other than what I thought best. I blamed them for coming to me with ingrained opinions that were resistant to facts. Gradually it dawned on me that I needed to form a relationship with the patient and understand them before they could understand me. I realized, not always successfully, that the medical encounter does not take place in some abstract mathematical space but in the nitty-gritty world of human relations. When I was able to make that human connection, the talk became much easier. Most of oncologists are obviously educated, and many of us are smart, Patients have no trouble recognizing that. We need to make patients understand that we are on their side, and you can’t fake that. Patients still made choices that were contrary to my thoughts, but they knew that this didn’t affect our relationship. A decision to undertake treatment that results in ten extra weeks of life at serious risk of toxicity does not belong to the doctor. In breast cancer, I often saw a rush to get treatment or not; my biggest effort was to get the patients to take the time to make a decision and weigh the facts. I succeeded about 80% of the time.

    • I’d agree with this assessment, but for me “succeeding about 80% of the time” sometimes means *not doing* what I think would be the best from a purely medical standpoint. I treat people for chronic, blinding conditions. Every day I see some patients who would probably choose to go blind (in one eye) rather than undergo a straightforward surgery. Transportation, pain, COVID, cost, and (most importantly) the vision of the other eye, all play into the decision. Sometimes, not doing the surgery that has an 80% chance of preserving the patient’s remaining vision for 3+ years can be the correct decision.

      • Can you give any references for making decisions about cataract surgery in this time of COVID? My regular ophthalmologist said at my regular checkup about a year ago that the time had come for cataract surgery. So I made an appointment with the cataract surgeon she referred me to, but then COVID appeared, and the appointment I made with the cataract surgeon was canceled with no notice to me.I don’t know when I’ll be able to get a. COVID vaccine — I’ve managed to get on a couple of waiting lists, but so far haven’t found a vaccine provider that has any vaccine.

    • Agree 100%!
      It is very difficult to engage in shared decision making without pretty clearly revealing your preference. I try to be honest about that aspect. Just difficult. (I’m hospitalist, not H/O.)
      Then if you ask 2 docs about the same issue, you get at least 3 distinct answers.

  9. I’ve thought about this as a patient, an analyst, and a physician (through my wife, a physician). My answers vary, depending on which hat I put on. As an analytics, and as discussed often on this blog, I think it is important to convey the uncertainty in the estimated effects of the diagnosis and treatment options. Gigerenzer has much good advice about better ways to convey probabilities in terms that are more easily understood. As a patient, I’d prefer the probabilities, but most patients would not know what to do with that. The physician’s hat is complicated – on one hand, experience clinicians will tell you that many patients require definitive advice – they don’t want to hear too much uncertainty. Of course, there is the added factor that some physicians don’t want to convey uncertainty because it diminishes their status. Disentangling these reputational effects (which I don’t respect much) from the real need of clinicians to have the trust of their patients, is not easy. And, the answer will differ depending on who the patient is. Unfortunately, the business pressures of modern medicine leave less and less room/time for these nuances to be considered. If a physician is compensated on a measure of efficiency (such as patients treated per day), then it is difficult to tailor an appropriate message for each individual patient. Then, we have the even more complex problem that most of the data concerns average effects, and what the patient really wants to know (and needs) is more personalized assessment – which may not be supported by the available data. Sounds like a complex problem! No wonder so little time is spent in medical education on statistics.

    • Excellent comment. I think that there should be as much discussion about how to assess which patients want which information as there is about how to convey specific types of information.

      As Andrew said in the last post, it’s important to know your audience.

    • What I have summarised from the comments so far:
      1. People (clinicians and patients) tend not to like and/or understand uncertainties.
      2. Options for summary statistics: percentages of survival/side effects, 5-year survival rate, rate in every 1000 people, NNT.
      3. The variation in local cultures and expectations (clinician and patient alike) influence the outcome and the choice of the communication. Some cultures are more paternalistic, some involve the patients in the decision making. It also depends on available resources such as: how much time do the clinicians/patients have? Is there any resources to facilitate ongoing communication/continuous education for both?

      To summarise:
      It’s hard and it depends on how much effort do you want to put into it.

      Something to ponder:
      Maybe we can start from: “how much do you already understand?” And “how much do you want to know more?”

      This will allow the patients to communicate what they already know and what they want to know.
      A. If they already know probabilities then it’s easier for the analyst.
      B. If they don’t know and want to know, then it becomes a two-way effort on how to communicate.
      C. If they don’t know and don’t want to know, then use the paternalistic approach (with downsides of course but pragmatic).

      Having said that, be mindful that patients are humans too.
      As oncodoc said, building relationship with the patients are very important.

    • Very good comment.

      I have some chronic health issues for which I have struggled to obtain a definitive diagnosis, and it is extremely frustrating to deal with doctors, in general. When I was in the Kaiser system, the process was a mechanistic “the most probable (/least expensive for Kaiser) cause of your issues is X, then Y, then Z, so we’re going to try the cheapest treatment for X, then the cheapest treatment for Y, then the cheapest treatment for Z”, even when I told them I thought Y made no sense and the combination of X and Z seemed possible. That said, at least I understood what they were doing and why, even if they mostly didn’t bother to explain it, and even if I didn’t agree with it.

      Since then I switched jobs and switched to a PPO to get more flexibility, and while I do feel the doctors are more likely to at least pretend to listen to my input, I still cannot get most doctors to speak to me like I understand probability and uncertainty, even when I ask. The one exception, interestingly, was my allergist, whose bedside manner sucks but who would at least say things like “80% of people who get allergy shots experience at least a 50% reduction in symptoms.”

      That said, most people are not like me. They want to be told what to do, not do a risk assessment. It’s a tough problem.

      • > I still cannot get most doctors to speak to me like I understand probability and uncertainty, even when I ask.

        It is very likely that the doctors don’t really understand probability and uncertainty well enough to do a good job of what you’re asking them to do.

        • “It is very likely that the doctors don’t really understand probability and uncertainty well enough to do a good job of what you’re asking them to do.”

          I agree.
          It may be that the best you can hope for from the average physician is something like, “This treatment sometimes has this disadvantage, and that treatment sometimes has that disadvantage. Which would you prefer?”

        • Exactly. In a lifetime of dealing with doctors, I only encountered one who knew enough about probabilities to discuss them with even minimal confidence. And I sincerely doubt that the others just avoid probabilities to avoid confusing patients. They are generally not even interested enough to know the probabilities themselves. In those cases where I had a genuine decision to make, I generally asked the doctor for a link to the research that underlies their recommendation. It is exceeding rare that they can provide even that.

          The exception was a urologist who had in his head percentages of side effects for various medications and, when I checked, he actually knew them!

        • I’ve had some discussions with docs about some fairly complicated questions of probability and uncertainty that suited my needs. I’d guess that your own knowledge and accordingly your standards are outliers.

        • I think some of these comments are unfair to, and too dismissive of, physicians. Some may understand probabilistic reasoning, at least well enough to speak in those terms. However, there are a number of reasons they may be reluctant to do so (some of which I am sympathetic to, and some not). They may know that the probabilities from published studies are too aggregate to be of much use in your individual case. They may feel that the evidence basis for the probability is so weak that citing the relevant probabilities might do more damage than good. Those are reasons I sympathize with. On the other hand, they may assume all patients are the same and will not understand probabilities so they don’t/won’t speak of them. Or they feel necessary to be the “expert” and that means never having to say your uncertain (to rephrase the wonderful t-shirt that says “statistics means never having to say you are certain”). Those reasons I do not respect.

          And then there is the reason to being more precise (by referring to probabilities) might potentially open them to lawsuits. On that account, all I can say is “sigh.”

        • > They may know that the probabilities from published studies are too aggregate to be of much use in your individual case. They may feel that the evidence basis for the probability is so weak that citing the relevant probabilities might do more damage than good. Those are reasons I sympathize with.

          Again, a gvery good comment. I have, at times, dug deeper with docs on the question of probabilities to find that part of what was going on was that I was trying to squeeze more certainty our of situations where there’s just a lot of unknowns. I think that the easy access to Google search “probabilities” can contribute to the problem where people come in to discussions about probabilities w/o enough background to really understand their in-context application

          > On the other hand, they may assume all patients are the same and will not understand probabilities so they don’t/won’t speak of them. Or they feel necessary to be the “expert” and that means never having to say your uncertain (to rephrase the wonderful t-shirt that says “statistics means never having to say you are certain”).

          Yes, and yes.

          > And then there is the reason to being more precise (by referring to probabilities) might potentially open them to lawsuits. On that account, all I can say is “sigh.”

          I haven’t encountered anything that resembles that. I wonder how much it really manifests, in comparison to the other reasons you suggest.

          Related to some of what you discussed:

          https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2563369/

        • To be fair, I think lots of statisticians seem not to understand probability well enough to do a good job, so I can’t see how Doctors could possibly be better. (see the fact that we’re still teaching NHST in the stats textbooks for example)

        • Dale said,
          “On the other hand, they may assume all patients are the same and will not understand probabilities so they don’t/won’t speak of them. Or they feel necessary to be the “expert” and that means never having to say your uncertain (to rephrase the wonderful t-shirt that says “statistics means never having to say you are certain”). Those reasons I do not respect.”

          Adding insult to injury, my experience is that some physicians tend to give reasons such as “That’s the way I’ve always done it,” or even “That’s the way I like to do it.”

    • Dale said,
      “If a physician is compensated on a measure of efficiency (such as patients treated per day), then it is difficult to tailor an appropriate message for each individual patient. Then, we have the even more complex problem that most of the data concerns average effects, and what the patient really wants to know (and needs) is more personalized assessment – which may not be supported by the available data. Sounds like a complex problem! No wonder so little time is spent in medical education on statistics.”

      While what you have said means that there is a need to spend more time on it!

    • Well said. Much of every encounter is a tacit negotiation of how best to communicate.
      Further, I believe at least some research has confirmed we are not especially good statisticians.
      Even when we know the applicable research findings and we believe the individual patient in front of us fits the studied cohort, estimating pre-test probabilities is inexact (or worse).
      Then there’s the question of individual assessment. We are “measured” for compliance to guidelines. So discussing the non-$ costs of mammography and PSA exams might land us in trouble with CMS. The science that goes into the guidance is dichotomized. Implementation becomes virtually Manichean.

  10. I like this source for communicating risk in general: “Communicating uncertainty about facts, numbers and science” by
    Anne Marthe van der Bles, Sander van der Linden, Alexandra L. J. Freeman, James Mitchell, Ana B. Galvao, Lisa Zaval
    and David J. Spiegelhalter https://royalsocietypublishing.org/doi/10.1098/rsos.181870

    Abstract: Uncertainty is an inherent part of knowledge, and yet in an era of contested expertise, many shy away from openly communicating their uncertainty about what they know, fearful of their audience’s reaction. But what effect does communication of such epistemic uncertainty have? …

  11. What I as a patient want to hear is how much experience *this* doctor (who is talking to me *now*) has with *this* (or that) treatment; how successful it is with *his* patients; and would he take it *himself*. I do not need the granular preciseness of the probabilities; I do not need decimals. I need to know first of all if *this* doctor is really leveling with me. I have to rely on my ability to get along with and “read” other people; I have to have formed a relationship, in other words, with *this* doctor. If some jack-in-the-box in a white coat pops up and starts reciting to me population statistics that read like they came from a package-insert, that contributes *nothing* to my sense of trust; in fact, for me, that’d be a deal-breaker. That does not mean I do not consult the package-insert, I do, and it must be consistent with the stipulations of the doctor as to his own clinical experience. But if the doctor simply defers to the boilerplate and is circumspect about his own clinical experience; then I find another treating doctor. (Please excuse all the ‘*’ for emphasis — they must do in lieu of italic type)

    • Fascinating. That’s precisely the opposite of what I want. I consider my own doctor’s direct experience to be too small a sample to be worth considering. And if we are discussing a surgical procedure, I absolutely want to talk to some white coat with population statistics, not to the surgeon. To each our own.

      • What was implicit in the above is that I want to know that this doctors direct experience *is* significant. Not once in a while; but e.g. he’s been doing the procedure 4 x/week for xx years. The treating doctor must have sizeable experience base giving the treatment.

  12. I wonder if the OP is asking the wrong audience–or at least not including the right people. People here can think about what that is valid that can be said. Patients might be able to convey what they can hear–and what they want to hear and what they can understand reliably.

    There’s a methodology called action research that might be of us. It tries to create knowledge of the form, “In this setting and with this goal, I am usually more successful with trying approach X.” Of course, math and statistics may become part of the answer.

  13. I think scientific literacy is a problem for the masses to understand. Statistics and Probability are specifically worst because they are more complicated, and not so understandable until for technical professions. Any news could be a bad news. 0.1% chance to die is a chaos to sensible persons, being said in more understandable form or not, still using metaphors or examples more understandable to try represent a reality more favorable, when it is possible. So, I think that… why to complicate? Why to try being more understandable? It’s lost time.

  14. Years ago my wife and I consulted a sequence of obstetricians in search of one who we thought could manage a difficult pregnancy.

    We settled on one who said “the chances aren’t good, but we will wait and see and try a healthy diet (or some such nonintervention)”. I pushed him hard to give me probabilities. He did not want to. Finally he relented, and said “five percent chance that it will be OK.”

    It was OK. But what I understood as soon as I heard his answer was that I really had no more actionable information than I did before. For us it was baby or no baby, not five percent.

    • It sounds pretty actionable — for most people there is some range of P low enough to change their expectations and some lower one at which they change their behavior. It’s not as actionable as a probability distribution of “how well or badly can this pregnancy unfold” in all its ramifications, but small P and tiny P seem different enough to make some differences even if one is determined to see things through to the end no matter what.

  15. When my wife was pregnant with our first child, we had genetic screening for some recessive diseases (she is Ashkenazi Jewish so Tay-Sachs and a few others were possibilities).

    There was an interview with a screening counselor who was not a doctor but some sort of nurse who worked in that office. I asked about false positive and false negative rates of the test and she rattled them off easily and immediately. We had some discussion (not detailed or numerical) of the possibility of followup tests or a retest in the event of a positive, given the false-positive rate. The math questions were unusual for her, apparently most people do not ask, but not unheard-of and the answers were straightforward and satisfactory.

    The message I took from this is that although the nurses are probably not the world’s most highly trained medical personnel and certainly no experts on statistics, it is very much possible for them to become fluent in the quantitative considerations related to their work for purposes of explaining to patients.

    • mr 80s said, “The message I took from this is that although the nurses are probably not the world’s most highly trained medical personnel and certainly no experts on statistics, it is very much possible for them to become fluent in the quantitative considerations related to their work for purposes of explaining to patients”

      My impression is that nurses’ training puts much more emphasis on communicating with patients than does physicians’ training, so your observation does not surprise me.

  16. Dear Someone who wishes to remain anonymous (and everyone else):

    I’d encourage you to take a step back and ask if the fact that you are a data scientist or statistician or math geek or Stan Man (or Woman) qualifies you in any way to explain how to communicate complex probabilities to the average person.

    Less than 35% of Americans have a college degree. I’d guess that less than 10% of Americans overall have had more than one college level math class, and the one they had was designed specifically so they wouldn’t fail.

    Given these facts, let’s take a look at how “anon e mouse'” allergist described the benefit of allergy shots:

    “80% of people who get allergy shots experience at least a 50% reduction in symptoms.”

    It seems like a pretty straightforward statement but it’s not at all that way. The “80%” part most people will get. The “50%” part is mostly meaningless. 50% reduction in what? Frequency? Intensity? Duration? Presumably it has some specific meaning in the context of some study about how to quantify and average these components of “symptoms”. But most people getting this information do not have even the most remote concept of how that’s done. To them it kind of sounds like “some relief”.

    Here’s my crack at how to communicate this information: “Most people feel a lot better after getting the shots”

    That’s what I recommend you tell physicians: boil it down as best you can to a one liner with qualitative quantification. Go any further than that, and most people will just turn you off. For the few people who want more information than that, Google is their friend.

    • jim –

      > Here’s my crack at how to communicate this information: “Most people feel a lot better after getting the shots”

      I’d guess a lot of patients might have follow-up questions, which would require addressing aspects of probability beyond your summary.

    • Please define “most” and “much.” Of course, you are saying that could be the follow-up question. But I suspect your statement will be sufficient for many, if not most, people. What do you suppose they think the information means? And do you think that “most” people will have the same understanding of the “most people” statement?

      I think a statement like that is an admission that it is hopeless to give people meaningful information about the actual effectiveness and side effects of allergy shots. If said in the right tone, it is simply a push to get people to take the shots because the communicator is the expert.

  17. I have not participated in this conversation until this point, and I acknowledge that more mathematics, more probabilistic modeling, etc, does not qualify me in any way to communicate probabilities to people with no math background. Please do not assume that’s what I assumed… but hey gotta start and get your feet wet somewhere. Doing nothing doesn’t help one become a better communicator. Nor is being a condescending prick behind closed doors or privately make you a better communicator.

  18. Anon, I would focus the talk more on setting up a dialogue then trying to give guidelines for the doctors to solve the problem themselves. The reason I say this is because you’re asking this forum, with it’s broad and varying advice for help. This is not a solved problem and may not have a solution. I think that you and your colleagues would be better off communicating on the topic on an ongoing basis and discussing individual cases as they come up. Then you can gradually generate a culture of using your services in a highly productive manner.

    To that end, you could use the Gigerenzer advice in a a few examples, and some of the other advice here as it becomes more applicable. You could have a straight up relation of frequencies on the one hand and then also have another example where you show how you did a search on various outcome probabilities and found something similar to what the patient needs to understand (e.g. probability of being struck by lightning or winning the lottery. or Red in roulette). And, if they sent you some probabilities they cared about you could generate a table with examples.

    • Psyoskeptic:

      I love, absolutely love, your idea of throwing it back at the audience, involving them in the goal-making and problem-solving instead of presenting them with an answer, being a facilitator rather than a guru.

  19. From a patient’s viewpoint: I am mystified as to why doctors would never give quantitative answers as to the probably our babies would be born on time, early, or late. This is an extremely common question, of great importance even though not life and death. The doctors themselves have the data at hand in the local hospital.
    Suppose they did collect their hospital’s data for the past five years of baby arrival relative to predicted due date. How would this best be communicated to patients?
    Would a histogram of bars indicating frequency be a good way? Would bars made up of 100 stick figures be better? Visualizations are far far better than numbers for pretty much everybody, from the illiterate to statistics PhD’s (remember the Chernoff faces anyone?).
    Or would it be better to say, “We had 300 births over the past few years in this hospital. 100 were early, 50 were late, and 150 were within one day of the due date”?

  20. Back in covid spring I put together a gigerenzer frequency box web app for probability you have a disease if your test says you do. I am not much of a python programmer, so it is not complicated code and readers might be able to improve on it, including any doctor who ever took a python class in college (I never took any coding). See

    https://www.rasmusen.org/blog1/understanding-bayess-rule-test-mistakes-and-false-positives-with-frequency-boxes-http-www-rasmusen-org-cgi-bin-bayesbox-htm/

Leave a Reply

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