Closing the Gap Between Machine Learning and Clinical Cancer Care-First Steps Into a Larger World
JAMA Oncol . 2020 Sep 24. doi: 10.1001/jamaoncol.2020.4314. Online ahead of print.
John Kang, Olivier Morin, Julian C Hong
Although recent years have seen an exponential rise in machine learning (ML) articles in health care, the majority of promising studies unfortunately do not undergo prospective validation or affect patient care. One reason for this low conversion rate lies in the challenge of determining appropriate clinical decision points and their accompanying interventions. In oncology, reaching an accurate prognosis for patients is critical to determining appropriate and compassionate care that straddles a fuzzy line between benefit and toxicity. Current prognostication methods depend on clinical judgment, incorporating elements such as multidisciplinary discussion, disease status, and semiobjective performance status assessments. Patients at high risk for short-term death may benefit more from supportive or palliative measures as opposed to aggressive and potentially toxic therapies. In this issue of JAMA Oncology, Manz et al1 report the prospective validation results of their previous ML framework for binary prediction of 180-day mortality in patients with cancer seen in the outpatient setting. This work represents an important early step in health care ML, though there remains a long road ahead for the field.
Read Full Article Here: https://doi.org/10.1001/jamaoncol.2020.4314