Automatic Pain Detection Algorithm for Patients with Cancer Pain Using Wristwatch Wearable Devices.

Pain assessment becomes challenging for patients unable to self-report, given the subjective nature of pain. This study introduces an automatic pain detection model utilizing biological signals from wristwatch wearables and time series data from patients with cancer experiencing pain. Biological signals and pain data were obtained from 10 patients with cancer pain for 7 days during their hospitalization. A total of 73,154 minutes of data and 407 pain reports were obtained. We developed automatic classifiers to detect moderate or severe pain and pain above the personalized pain goal by several machine learning algorithms using per-patient and mixed data sets. The best-performing algorithm achieved an F1 score of 0.87, with enhanced performance using the personalized pain goal as the cutoff. While the generalized model requires improvement, the study demonstrates the feasibility of automatic pain detection using extended real-world patient data.