<div>Physician prescriptions might contribute to the opioid epidemic. We study the relationship</div><div>between average patient case complexity (mean HCC risk score) and physician opioid</div><div>prescription propensity. Among 100 physicians at one Minnesota hospital, North et al. (2017)</div><div>find evidence of a positive, linear relationship. However, using 2016 Medicare Part D data on</div><div>over one million physicians across the U.S., we find evidence of a relationship that often</div><div>increases within North et al.’s range of mean HCC scores, but then decreases and levels off</div><div>among higher mean HCCs that North et al. do not observe. Overall, we document wide variation</div><div>in this relationship across the most frequent specialties and by geography. We then introduce a</div><div>novel methodology for diagnosing each top specialty’s outlier states, whose relationships differ</div><div>noticeably from the specialty’s baseline model. Using a hypothesis-testing approach based on a</div><div>hierarchical linear model, we flag outlier states that vary widely by specialty. State-level</div><div>evidence suggests that unusual physician propensity transcends one region such as the Rust Belt.</div><div>Future research should build upon our model by using a more granular geographic level.</div>