LAUSR

Invited Commentary The Electronic Health Record as Practice-Improvement Coach Unnecessary diagnostic imaging is a prevalent form of medical overuse, and leads to unnecessary invasive procedures, radiation exposure, emotional stress, and nosocomial infection. Research has focused on how to identify clinicians who are more likely to overuse diagnostic imaging and how to intervene. For example, studies have identified the number of magnetic resonance images (MRIs) that physicians ordered for uncomplicated low back pain1 and the number of antibiotic prescriptions for upper respiratory tract infections.2 In this issue of JAMA Internal Medicine, Chong and colleagues3 consider the overuse of computed tomographic pulmonary angiography (CTPA) in the diagnosis of pulmonary embolism. They report on the use of “diagnostic yield” (the number of positive studies/total number of studies ordered) as a metric to help identify overuse. Low diagnostic yield results from performing CTPA in patients at low risk of pulmonary embolism. Their study found marked variation in diagnostic yield between providers, with the absolute number of CTPAs ordered being more predictive of low diagnostic yield than physician characteristics such as age, specialty, and years of experience.3 These results are in accord with other research showing that a large number of CT scans are being performed in low-risk patients,4 and suggest that diagnostic imaging is being disproportionately ordered by a select group of clinicians. Chong et al3 make the case that low diagnostic yield can help identify physicians who may be ordering too many CTPAs in lowrisk patients. It is important to note, however, that a high diagnostic yield implies that a provider may only be ordering CTPA for the highest-risk patients, thus potentially missing the diagnosis of pulmonary embolism in some. Given that there are good estimates of the prevalence of pulmonary embolism in various clinical situations, as well as good data on the sensitivity and specificity of CTPE, an expected diagnostic yield could be calculated, and used as a standardized point of comparison.5,6 Even in situations where there is inadequate evidence to calculate an expected diagnostic yield, the mean diagnostic yield from a large group of physicians might suggest a reasonable starting point for analysis. What might this approach add to other implementation strategies for value-based care? One possibility is to use the electronic health record to develop real-time performance metrics, and then to use these metrics to target interventions based on the electronic record, such as decision support tools and peer comparisons toward individual clinicians. Electronic medical record platforms are sophisticated enough to keep a running calculation of a range of quality metrics per physician, such as dollars spent per diagnosis, and trends in the prescription of various analgesic classes over time. Such clinical data can then be used to provide personalized support to clinicians who are either outliers among their peers or whose metrics are far from the expected value. Such personalized approaches to feedback may be more impactful than more general low-value care alerts, such as the obligatory used of a decision support tool based on Wells’ risk-stratification criteria before ordering CTPA. They also may be more visible to clinicians than peer comparisons that are communicated through an online dashboard or email. Further, targeted interventions through the electronic health record could potentially reduce alert fatigue by decreasing the overall number of reminders for physicians. As a metric for appropriate use of imaging technologies, diagnostic yield has limits. The use of sensitive imaging modalities may result in high numbers of incidental findings that increase diagnosticyieldandthecostofcarewithoutaffectingclinicaloutcomes. This is the case with the use of carotid ultrasound in the initial work-up of syncope, an established form of medical overuse. In patients with uncomplicated musculoskeletal pain, imaging findings are often poorly correlated with clinical outcomes. Thus, the meaning of a high diagnostic yield would be unclear.7 Nonetheless, the study by Chong et al3 is thought provoking, and demonstrates how the electronic health record can be used as a performance coach to help physicians improve patient care.