Readers of this blog will know I actually enjoy kind of obscure research. A new paper at the National Bureau of Economic Research explores physician assessment of health risk and subsequent treatment decision-making. (NBER Paper) The authors used the case of testing for heart disease and attacks. A lot of concern has been expressed about overuse of testing when not needed, but they explored underuse, as well–the failure to test when it was appropriate. Obviously, deciding whether to test for a potential heart attack is a critical decision, with serious consequences for the patient. And you might expect some overuse, given those serious consequences. But the symptoms that could indicate a heart attack are also related to many other, less serious conditions. The best test, a stress one and/or catheterization, to determine if a heart attack has occurred is invasive and expensive, so some caution in its use is warranted. So how do physicians make the decision to test and how accurate are they in those decisions? The authors use Medicare claims and other data to attempt to shed some light on these questions. They assess the objective risk that a patient actually had a heart attack and therefore should have been tested.
As expected, they find a number of patients who probably did not need to be tested, because the relevant factors and subsequent diagnosis information would suggest very low risk that a heart attack occurred. From a cost-effectiveness perspective, if a year of life is worth $150,000, over 50% of tests would be eliminated because their benefit was over that threshhold. At the extreme end, the testing that occurred in the lowest ten percent of the risk scale cost over $600,000 to obtain an additional year of life. But there were a large number of patients who didn’t get tested but probably should have, according to the author’s risk prediction algorithm. Failure to test, and change treatment based on the presumably positive test result, caused death and other morbidity for patients. By almost any cost-effectiveness measure, testing these patients is justified. Under the authors approach, about 18% more patients would be tested, which would ultimately save about $228 million a year net of the testing costs, and about 52% of tests would not be performed, saving around $300 million, after considering the cost of missing a diagnosis by the reduced testing at the low risk end.
So it appears that physicians both over-test and under-test and that simply telling them to test less reduces both necessary and unnecessary tests, which also supports the idea that their risk assessment isn’t always highly accurate. And day of decision-making can be an issue, on the weekends there is less testing of both high and low risk patients, because the tests usually aren’t available on the weekend. Two other factors the researchers identified which appear to be sources of physician error are over-reliance on demographic factors–over-testing in high-risk demographic groups and under-testing in low-risk ones, and less testing in patients with prior pneumonia diagnoses, as that disease can have similar symptoms to a heart attack. Of course, some over-testing undoubtedly has an economic source, people make money from testing. But that is likely declining as more physicians and health systems are at risk for the costs of care. Creating better algorithms to help doctors decide who really needs a test would seem to be the best approach and this paper helps achieve that goal.