Despite steady progress in medical research and innovation, medical decision making remains extremely challenging, even for experts. Chicago Booth’s Devin G. Pope and his coauthors find that small factors—things as simple as the weather on the day a test is performed—can make a big difference for treatment decisions. Understanding these factors, Pope says, can help make doctors’ jobs easier and patient outcomes better.
So my coauthor, Ziad Obermeyer, and I got a data set that has information on about 5 million people and all of the blood tests that they received over a several year period. And we looked at the temperature on the day that you got your blood test and we find that it has a pretty remarkable effect on the results from several of these tests.
So some of the things are not very surprising: if it was a really warm day when you had your blood test, then you look like you’re a bit more dehydrated than usual. So for example, your creatinine level, your pH level in urine, and your urea nitrogen all look as it should if you’re a bit more dehydrated. But what’s perhaps more interesting is we see that the temperature on the day of your blood test can have effects on other blood-test measures that maybe we wouldn’t have predicted.
So you find that even in the second week of March, if it happens to be a warm day versus a cold day, your cholesterol might look very different if you have your blood taken. We think this is fascinating because because we can predict where these errors actually occur, you could imagine trying to correct for this. And the errors, in addition, or the bias is large enough that it can have an impact on how doctors would treat you. So for example, on a really warm day, your cholesterol levels look better than on a really cold day. And you find that you’re about 5 percent less likely to be prescribed statins or cholesterol medication, if your blood was tested on a warm day than if it was tested on a cold day. So the effects are large enough to actually impact the type of medication you might be prescribed. And this suggests that we should be thinking about the weather on the time of these blood tests and whether or not a correction might be necessary.
In a related project, Ziad Obermeyer but then also Armando Meier, Kevin Volpp, and I look at the same blood test data and we’re interested in how doctors treat people who had a blood test that was right below a guideline versus right above a guideline. All right, so let’s think about this for a minute. Imagine that you have a blood test done that measures your prostate-specific antigen, or PSA for short. Your PSA is a blood test that allows you to see if you might have prostate problems or if you’re at the beginning stages of prostate cancer. Now there’s a guideline out there that says if your prostate-specific antigen or your PSA is more than 4.0, then that puts you in kind of an abnormal range and makes you more likely that you might have an issue with your prostate. If you’re less than 4.0, so let’s say you have a 3.9, you’re in the normal range, although a 3.9 is pretty close to the abnormal range as well. So we look to see how doctors treat patients who fall just below a guideline versus just above a guideline—a 4.0 versus a 3.9.
And what we find is that they’re treated very differently. So a patient that is just above the guideline is, could be two or three times more likely to have a biopsy done, to have a prostatectomy done, to be retested—many things that would come with additional attention and care change discontinuously at that guideline. We argue that this is not necessarily an efficient outcome. It could be that many of these people that have a 3.9 outcome because of their age or family history should potentially be treated as well. And maybe some of the people who had a 4.0, just were barely abnormal, because of their age or family history, maybe shouldn’t be treated. Ideally doctors would smooth out those discontinuities such that there wouldn’t be a discrete break right at the guideline, and yet we see discrete breaks.
We see this for other medication and other guidelines as well. So for example, if you have your TSH value tested, your thyrotropin, which is a test for whether you have a thyroid issue, we find that the guidelines . . . if you’re right above versus right below, you’re treated very, very differently in terms of, say, your probability of getting prescribed thyroid medication.
We don’t see this with all guidelines. There are some guidelines where doctors seem to do a pretty good job actually of smoothing out people that are just below versus just above the guideline. And so we hope that future research and some of the stuff we’re working on can better understand when is it that doctors treat guidelines very discontinuously and when do they smooth out patient care through the guideline?
All right, so what are the implications for both of these research projects? We’re not saying that doctors are dumb or that they’re doing a poor job with treating patients or that they’re being neglectful in any way. We’re certainly not trying to say that. We are trying to suggest that being a doctor is hard, and that there’s a lot of things that one has to look at and analyze in order to make proper diagnoses and treatment decisions. And some things like the weather at the time the test was taken, or whether you’re just below or right above a guideline can be very hard to pay attention to if you’re a doctor that has many, many patients and you’re trying to pay attention to all sorts of things.
Given that these things matter, what could we do to help doctors try to correct for some of these issues that might play out due to just lack of attention or lack of ability to focus on every patient with a lot of time? Well, there are a couple of ideas. So one could create decision aids for doctors that allow a doctor to plug in a bunch of information about a patient, and then the decision aids use perhaps some machine-learning algorithm that allows them and suggests outcomes to them or suggests treatments to them that they may consider. This might help a doctor, for example, if someone has a blood-test result that’s just barely below a guideline, so it appears normal, but maybe they have a bunch of blood tests that are all just below a threshold and they maybe really should be treated more seriously, this decision aid could indicate, hey, this person is someone who we should consider with giving more treatment to.
There are other things that maybe could help doctors. So take, for example, the blood-test form that is often given to both doctors and patients after you get a panel of blood-test results. Sometimes these are very hard to read, and it could be easy to not notice when your blood test is just below a guideline but very close to being in the abnormal range. Could we redesign these blood-test forms to help both doctors and patients better understand what their blood-test results are actually saying, rather than just saying, you know, it’s abnormal or putting a star next to the value? We could give something that allows them to better understand the overall level of your blood-test results, whether it’s really abnormal or just barely abnormal. These are all ideas that we look forward to exploring and we hope can be taken seriously, and in the end could help improve the lives and health of many patients.
Research explores how algorithms influence our social perceptions.
What Are We Learning When We’re on Social Media?
A different framework could help confront misinformation.
We Need a New Way to Talk about Lying
The debate has raged for 300 years and counting.
What Is the Line Between Self-Interest and Selfishness?Your Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.