Until recently, economists, policy makers and workforce experts have relied on outdated and inaccurate snapshots of the U.S. physician workforce, making it especially difficult to predict the need and availability of health care services across the country. Data about each physician's area of specialty is collected at the beginning of their career and is rarely updated, increasing the potential for outdated information about who is providing care for our nation's population. In this study, Wingrove et al examine how machine learning algorithms may allow for more real-time, accurate descriptions of the medical workforce, including professions that do not formally collect specialty data like physician assistants and nurse practitioners. Algorithms also can identify physicians in new and evolving interdisciplinary positions. One such learning model from the Robert Graham Center and the University of Pittsburgh was trained to identify a majority of medical specialties with 95 percent accuracy. The model was fed data from clinical encounters in the form of procedures and prescriptions billed by Medicare from 2014 to 2016. The models were less accurate at predicting some specialties, like neurosurgery and physical medicine and rehabilitation. But overall, the model correctly predicted 70 percent of physician's practice type within five percentage points of their actual count, including primary care and specialties such as emergency medicine, cardiology, gastroenterology and radiology.
Using Machine Learning To Predict Primary Care and Advance Workforce Research
Peter Wingrove, et al
University of Pittsburgh, School of Medicine, Pittsburgh, Pennsylvania and the Robert Graham Center for Policy Studies in Family Medicine and Primary Care, Washington, DC