Effect of Machine Learning–Directed Clinical Evaluations on Acute Care Visits During Outpatient Therapy
In a single-institution study reported in the Journal of Clinical Oncology, Hong et al found that mandatory twice-weekly evaluations reduced the need for unplanned acute care visits among patients undergoing outpatient radiotherapy or chemoradiation who were identified as being at high risk for such visits by a machine-learning algorithm.
"In this prospective randomized study, machine learning accurately triaged patients undergoing radiotherapy and chemoradiation, directing clinical management with reduced acute care rates vs standard of care."— Hong et al
Tweet this quote
The quality improvement study (System for High-Intensity Evaluation During Radiation Therapy, or SHIELD-RT) involved machine-learning algorithm evaluation of 963 outpatient adult courses of radiotherapy and chemoradiation started from January to June 2019 at Duke Cancer Center. The study objective was to determine whether machine learning can identify high-risk patients and direct mandatory twice-weekly clinical evaluation to reduce acute care visits (emergency department visits or unplanned hospitalizations) during treatment.
A total of 311 courses identified as high risk (> 10% risk of acute care visits) were randomly assigned to standard once-weekly clinical evaluation (n = 157) or mandatory twice-weekly evaluation (n = 154). Both groups were permitted to have additional evaluations at clinician discretion. The primary endpoint was the rate of acute care visits during radiotherapy.
Among high-risk courses, rates of acute care during treatment were 12.3% for twice-weekly evaluations vs 22.3% for once-weekly evaluation (relative risk [RR] = 0.556, 95% confidence interval [CI] = 0.332–0.924, P = .02). In analysis extending to 15 days after treatment, rates of acute care were 22.1% vs 32.5% (RR = 0.68, 95% CI = 0.468–0.987, P =.04).
For low-risk courses, the rate of acute care visits was 2.7%.
Overall, machine learning showed good prospective predictive performance. The binary discrimination of high and low risk had a receiver operating characteristic area under the curve value of 0.820 across all courses. Among patients who underwent standard once-weekly evaluation, the area under the curve value was 0.851.
The investigators concluded, “In this prospective randomized study, machine learning accurately triaged patients undergoing radiotherapy and chemoradiation, directing clinical management with reduced acute care rates vs standard of care. This prospective study demonstrates the potential benefit of machine learning in health care and offers opportunities to enhance care quality and reduce health-care costs.”
Julian C. Hong, MD, MS, of the Department of Radiation Oncology and Bakar Computational Health Sciences Institute, University of California, San Francisco, is the corresponding author for the Journal of Clinical Oncology article.
Disclosure: The study was supported by the Duke Endowment and the Duke Department of Radiation Oncology. For full disclosures of the study authors, visit ascopubs.org.The content in this post has not been reviewed by the American Society of Clinical Oncology, Inc. (ASCO®) and does not necessarily reflect the ideas and opinions of ASCO®.