• Speeding up learning in the health care system

    The current NEJM has published a clinical trial with a statistical commentary that is really exciting. (How is that for a sentence that you never expected to read?) Bear with me while I present the background problem and explain why the trial may mark a breakthrough in how quickly we can change medicine.

    The interminable battle over the ACA suggests that the only significant problem in health care is how we finance it. That’s wrong: the real problems are that the care isn’t good enough, and yet it costs too much. The most important issues in health care are improving the patient experience of care, reducing the per capita cost of care, and — above all — improving the health of populations. You can’t achieve these goals through legislation.

    These goals can only be achieved by learning how to deliver more effective care at lower cost. How does such learning happen? Partly through advances in basic science that lead to better treatments. Advances in basic science, however, do not translate directly into effective treatments. You have do a lot of clinical research to develop a new therapy and eventually you have to test that therapy on patients in clinical trials. The rate at which you can run clinical trials is therefore a rate limiting step on how quickly the health care system can learn and improve.

    It’s valuable to think about the health care system as an adaptive learning system. We can then look for ways to reorganize the system so that it can learn more quickly. At the end of the day, the clinical trials process is more or less trial and error learning. And the only way to speed up trial and error learning is to run a lot of trials and run each trial more quickly.

    Here we reach a deep problem: clinical trials are slow and expensive.* Trials take years to organize and once started they take years to recruit enough patients to adequately test the new treatment. Each patient costs tens of thousands of dollars to run, much of it in the cost of finding and recruiting the patient, and the cost of the data collection to assess the patient and the outcomes of her treatment.

    There are other problems with clinical trials. Trials are frequently designed to recruit groups of nearly homogenous patients. Using groups of highly similar patients makes the data cleaner. Clean data maximizes the statistical power of the trial, which helps keep the cost down. But the consequence is that the trial is run on a subpopulation of patients that is unlike the more heterogeneous patients who are actually seen in clinics. That means it is hard for a doctor to know whether the result of the clinical trial actually applies (‘generalizes’) to the patient she is treating.

    And now we get to the study in the NEJM. The TASTE study (Thrombus Aspiration during ST-Segment Elevation Myocardial Infarction) was testing a tweak (thrombus aspiration) in the technique for percutaneous coronary intervention (PCI), a common treatment for heart attack patients. The news about thrombus aspiration is that it doesn’t work. Unless you are an interventional cardiologist, that’s not interesting. What’s interesting is how they did the trial.

    The trial was conducted in Sweden, where there is a national registry of angiography and angioplasty patients. That is, all the patients who received these cardiological treatments are enrolled in a database. Records about them are kept in a standard way across the country. There is routine follow up with standard measures so that Swedish doctors know what eventually happens to all these patients. This is incredibly valuable for studies of the cost and quality of care, for assessing drug safety, and for retrospective analyses of the outcomes of care. The Swedes have registries for all kinds of procedures and it helps them achieve high quality of care.

    What the TASTE study did was to run the trial using the national registry data. That is, they recruited patients them from the list of patients in the registry. Patients were randomized to either standard PCI or PCI with thrombus aspiration. The trialists then used the registry data to describe the clinical and personal histories of the patients. They used the outcomes data collected by the registry as the trial endpoints. Using the registry meant that the trialists could find patients faster because they just had to look them up. They could run each patient much more cheaply because most of the critical data were already being collected. And they knew exactly how representative the clinical sample was because the entire patient population is in the registry. So the study was done quickly and cheaply, and the results are generalizable.

    There is, of course, an up front cost to set up the national registry. Before anyone complains about this, be aware that care is not only better in Sweden, it’s a lot cheaper.0006_health-care-oecd-cropThe costs of establishing and running a registry are mostly getting everyone to keep records in a standard way and regularly following up patients to measure clinical endpoints. The thing is, we should be doing these things anyway.

    The lesson from the TASTE study is that we should implement registries throughout the US and Canadian health care systems and use them to run quick and efficient clinical trials. That will help us adapt our way to a health care system that works well at an affordable cost.


    * See Avik Roy here for a criticism of the FDA’s role in slowing the pace of clinical trials research.

    • Clinical trials may be slow and expensive, but does Roy’s point apply to what you’re talking about? They’re about slightly different things: Roy is talking about the approval process for new technologies whereas you’re interested in generalized learning, particularly about already-existing treatments, no?

      I think the distinction is important. Because even with such stringent requirements for approval, we still see technologies with low benefits and high cost. Perhaps a laxer approval process would allow companies to price closer to their actual medical and economic value, but this doesn’t necessarily seem clear. It also seems possible that laxer approval process would simply besiege the system with low-value/high-cost technologies.

      • Darius,
        I am talking about generalized learning, although I don’t think it applies only to existing technologies.

        Your concerns about Avik’s argument may be valid. I’m not arguing for laxer regulation — retweets are not endorsements — and I know very little about the FDA approval process. I’m mostly linking to Avik because I like the guy and I think his views are worth discussing.

    • Given Americans deep seated distrust of government I don’t see how this could happen here within a realistic time frame (that it may happen 75 or 100 years from now puts it in the realm of science fiction). Given that even doing the most basic comparative effectiveness research led to a lot of talk about death panels I can only imagine the crazy stuff that would be said about a widespread registry (something like they’re trying to find out who has unaffordable conditions so they can deny coverage and do unspeakable experiments on them seems likely). Trumped up privacy concerns would also be a major issue.

      While this is common sense, I don’t see how any organization but the government could implement it in a sufficiently widespread manner and I don’t see how it is possible to overcome Americans’ deep seated distrust of government to make this possible.

      Of course, the Canadians are rather more sensible and I could see this happening there, at least at the province level anyway. Maybe the best we could hope for in the US is that California or New York goes ahead with the idea, still unlikely but maybe not impossible.

      • Tzmiskes,
        You are probably right. There’s probably a post to write about a pragmatic strategy for bringing this about.

        My guess is that it could start in either a progressive state, as you suggest, or a big Accountable Care Organization. If it works there, it might spread.

        I live and work in Canada and I hope we can pull something like this off, at least on a provincial scale.

    • Registries exist, ACC being one, but this just out yesterday:

      Getting docs to enroll patients–as the registries tend to hew towards hardware and procedure tracking the first step. And not an easy one from what I am told. Not the FDA so much as provincial thinking, markets, self interest, and need for culture change.

      I am wondering, what drew you to this study Bill Not your usual fare, nor is the study all the far reaching in application.

    • You are so right that we need to focus on how much we’re paying, not just who pays.

      I’ve long thought drug (and other) trials would be more effective if the drug companies paid the FDA and the FDA contracted the work out and required publication of the results. As I understand it, drug companies contract out the work, the only change would be that the FDA would be channeling the money and the researcher wouldn’t work for the drug company. Plus the drug company would have no control over the report (except, perhaps, for the ability to appeal).

      Also, that the drug companies have to report the negative findings in as large a type and the same rate as other findings.

    • This is a great idea but I echo Tzimiskes point that this would never ever ever happen in America for a few decades. Back when the IRS thing was in the news, one of the points to prove that the IRS was targeting political opponents was that some conservatives had been audited. In a country with hundreds of thousands of audits, people of all political persuasions will get audited but that didn’t stop people from using this as evidence.

      Now, imagine what happens when the same kind of person finds out that they were put into the control group. “My grandfather didn’t get the right kind of heart medication because he voted for the wrong person.”

      The other issue with this kind of database is that the health officials may not let researchers get access to it. Taiwan has a similar registry, but the only people who can access it (even for anonymized data) are people working for the Ministry of Health. Doctors, scholars, epidemiologists, etc. still have to collect their statistics independently.

      • Matthew,
        That story about Taiwan is distressing.

        • I found this out because I had a professor talking about doing survey work on education. He made some comment about not being able to be sure about the actual age distribution which I found ridiculous because Taiwan has a 98% enrollment rate with the NHI. With that info, you could look up the age of every single person on the island since it’s in the records.

          He said that you could, in theory, but that very few people were allowed to access that data set.

    • Meaningful use mandates the exchange of medical information in a standard format. While less effective than a registry it can be a starting point for a registry as well as making it easier to conduct trials since the patient’s medical history will be readily available.

    • i’m all in favor of learning, and compiling databases of good and useful information, but the unpalatable (for many) facts are that the swedish health care system is basically government-run and government-funded with few private doctors and (apparently) no private insurance:


      and doctors make what looks like a comfortable but certainly not outrageous income:


      i don’t know about canada but here in the us doctors, hospital executives, and insurance company executives make vastly more than do almost everybody who utilizes their services. either cutting provider incomes in half, or doubling the median us household income would go much further toward making health care “affordable” than would trimming a little bit here and there from utilization.