The machine learning (ML)-based methods to predict future symptom deterioration among patients with cancer using data from electronic health records (EHRs). In this study, researchers utilized internal EHR data from the Princess Margaret Cancer Centre in Toronto, Canada, for 2269 patients who received a total of 14,697 treatments. The researchers then built ML-based algorithms predicting deterioration in 9 patient-reported symptoms within 30 days following receiving treatment for various aerodigestive cancers. Overall, the best-performing ML systems developed in this study were able to predict symptom deterioration in the internal testing cohort.
In analysis of the external validation cohort, AUROCs with the model ranged from 0.67 (95% CI, 0.66-0.68) for anxiety to 0.73 (95% CI, 0.72-0.74) for drowsiness. However, there was significant heterogeneity identified across the centers in terms of performance for analyses of 6 symptoms. The symptom associated with the broadest range of AUROC performance across centers was appetite (range, 0.32 to 0.77), and the symptom with the smallest range was drowsiness (range, 0.64 to 0.80). ML systems can predict future symptoms among people with cancer from EHR data, although systems trained at one cancer center may not generalize consistently to others.
【MORE】