This Algorithm Can Help Docs Better ID Pain in Black Patients
By Bapu Jena, MD, PhD
April 2, 2021
Bapu Jena is an associate professor of health care policy at Harvard Medical School, a physician in the Department of Medicine at Massachusetts General Hospital and a member of the 2021 Tradeoffs Research Council. His research interests include the economics of physician behavior and the physician workforce.
Pain is largely subjective, making its diagnosis and treatment especially vulnerable to bias, disparities and inconsistencies. Underserved populations and particularly people of color have historically reported higher levels of pain for the same medical condition, and these higher levels of pain often remain even after researchers account for objective differences in severity of the disease. Sometimes, the presumption is that these racial differences in reported pain stem from external factors, like stress or socioeconomic conditions, rather than true differences in disease severity.
A recent study published in Nature Medicine sought to examine whether these pain disparities were in fact more physical, and not just environmental. Researchers Emma Pierson, David Cutler, Jure Leskovec, Sendhil Mullainathan and Ziad Obermeyer looked in particular at osteoarthritis of the knee, a condition that Black patients are more likely to report causes severe pain than white patients, even when X-rays appear similar to radiologists. The authors applied a machine learning algorithm to X-ray images of the knee to identify which imaging features predict self-reported pain by patients. This algorithm, they found, was far more accurate at predicting pain levels in Black patients than current assessments by radiologists.
Current measures of disease severity account for 9% of racial disparities in pain, whereas predictions from the researchers’ algorithm accounted for 43%. In other words, this algorithm was able to identify physical predictors of pain in X-rays of Black patients that radiologists regularly miss. (While the differences were most pronounced across racial lines, the new algorithm also reduced disparities across income and education gaps.)
One possible explanation suggested by the authors is that the current standard radiologists use to read knee X-rays of people with osteoarthritis was developed more than 60 years ago using white British populations. The new algorithm, on the other hand, was “taught” using a much more diverse patient population. This has not always been the case for health care algorithms, several of which have been shown to exacerbate racial disparities.
These findings imply that racial disparities in pain may arise, in large part, from clinical factors within the knee that are theoretically observable in radiographic images but go unnoticed by human eyes. The authors hypothesize that radiologists’ failure to see these physical explanations for pain in Black patients (as well as lower income and less educated patients) could make these patients less likely to receive joint replacement surgery and more likely to receive opioid prescriptions for their “unexplainable” pain. External factors, such as stress and other non-medical conditions, may still play an important role in how pain is perceived by individuals, but this research suggests that those factors may have received outsized interest historically.