Assessing the Overuse and Underuse of Diagnostic Testing

Low-value health care—care that provides little health benefit in light of its costs—is a central concern for policy-makers. Diagnostic testing is a particularly important example: while the use of high-cost diagnostic tests has skyrocketed,1 for many tests the “yield”—the frequency with which tests identify new diagnoses or trigger effective interventions—appears low. This is typically viewed as evidence that physicians are systematically over-testing low-risk patients, and has led to multiple high-profile campaigns to curb overuse of testing.
Looking at the average yield of a test for all patients tested might suggest over-testing, but the challenge is to identify and reduce low-yield testing—not to reduce testing across the board. So there is a clear need to identify marginal patients: those who were tested but have low chance of benefitting. But identifying these patients ex ante is difficult, both for doctors and for policy makers. Describing this pool would help quantify the extent of overuse and provide guidance on how we might reduce it. This hinges on having an accurate prospective prediction of each patient’s likely benefit.
To this end, we aim to develop methods that draw on recent advances in machine learning algorithms to predict risk for Medicare patients who are likely to benefit from specific diagnostic tests: stress testing for myocardial infarction (MI); CT pulmonary angiography (CT-PA) for pulmonary embolism (PE); and spine MRI to detect treatable causes of back pain (e.g., fracture, cancer). These tests seek to detect treatable medical conditions with important health implications, but involve substantial costs, and often risks to patients. Building on early evidence that machine learning algorithms combined with massive datasets can make highly accurate predictions, we will predict risk using data available to the doctor at the time the test is ordered. This will allow us to predict yield prospectively, and assess the extent to which doctors are (a) using tests in patients with predictably low benefit or (b) failing to use them in patients with predictably high benefit. This can elucidate root causes of variations in testing, and lay the groundwork for interventions to improve the value of health care.