I just read a piece in Quartz, an international online news reporting service. The article is entitled “A new study of 250 million patients shows medicine is still full of guesswork” and reports on a study published in Proceedings of the National Academy of Sciences. The study makes some assumptions they don’t tell us about.
What the study does is aggregate the medical data from 11 sources in 4 countries in a standard way. It’s a great Big Data project and may tell us much. This study of that data though is a study of treatment pathways, and they found interesting things comparing the treatment pathways of diabetes, hypertension and depression. Really the point of this study was to show that study of this issue in an international way is possible and I applaud them for that, but they draw conclusions that go way beyond the data.
Here’s a quote from the article “The pathways revealed that the world is moving toward more consistent therapy over time across diseases and across locations, but significant heterogeneity remains among sources, pointing to challenges in generalizing clinical trial results.” They go on to point out that in the case of diabetes over 90% of people get the same first line medication, but that things aren’t “so good” in hypertension. The basic assumption in the study is that what should be happening is that everyone with one of the complex chronic illnesses, all three of which have multiple causes, should get the same first line treatment. And in fact the study shows that over time, more and more people with these three illnesses are being treated with the same first medication. The authors feel this shows an improvement of medical insight. I wonder how many of them have actually practiced medicine.
When I read that 90% of people in four countries with adult onset diabetes were started on the same medication my first thought was, “Wow, which one? I bet they have a great marketing department.” It might seem like because Type II Diabetes is a single illness it should have a single treatment, but it doesn’t have a single cause. Illnesses for the most part are final common pathways of multiple pathways. We cannot assume that everyone with the same diagnosis needs the same treatment. What the study actually shows is that medicine is being practiced more and more by fiat of large organizations. Over time doctors are picking the treatment the insurance company wants you to start with, or the one with the best advertising, or the one the government endorses. This was not the case for thousands of years.
Dr Osler said that it’s more important to know the kind of patient that has the disease than the kind of disease the patient has, and it isn’t less true today. The authors of the study felt that there were too many different anti-depressants being used first line and that this shows some lack of consensus. When I looked at the data I noticed something completely different. I saw that in some places only a few medications were used for everyone, and in other places over 10 medications were used first. If I had to pick where to go, I’d pick the place with the 10. To me that doesn’t show a lack of understanding or a lack of consensus. It shows a greater depth of understanding of each individual patient. There was nothing wrong with any of those 10 medications. The place that only used a few was leaving tools in the toolbox. I’d rather go to doctors that think than doctors that look good to researchers.
Another assumption they made was that the first drug worked. In the case of diabetes, more than half of those people that got put on the same first drug had to switch to one of the others. That might seem okay from an algorithm standpoint, but if you were the patient wouldn’t you like to skip the medication that wasn’t going to work and get right to the one that would work for you? Is there “guesswork” in medicine? You might call it that, and early on in training it is mostly guessing. But over time and with thousands of repetitions physicians start to get an intuitive sense of things. This is hard to measure because it is based in unconscious pattern recognition that even the physician doesn’t know they’re doing. Once you’ve seen a couple of thousand depressed people, your unconscious mind remembers patterns and outcomes and there starts to be an accurate intuition. If we go for the reductionistic algorithms we’ll lose that. And that would be a shame.
I actually think algorithms can help, but only by trying to recreate the complex pattern recognition that physicians do naturally after thousands of repetitions. I think this is possible. In the ones we’ve created for addiction we have inputs no one else looks for consciously. That kind of algorithm hold promise, but by reducing every person with the same diagnosis to the same treatment sequence we’re making it easier on the researchers and administrators and harder on the patients.