1570P A deep learning model for prediction of chemotherapy infusion related symptoms using continuous heart rate variability

We hypothesized chemotherapy infusion related reaction might be predictable according to continuous monitoring of heart rate recorded by a wearable device. In this study we propose deep learning model that predicts patient level symptoms per minute during infusion. prospectively enrolled patients with cancer who will receive and at high risk for reaction. During infusion, 1 interval was from smart band self-recorded symptom log obtained. Patient's status every classified into 4 categories; normal, mild without intervention, serious which needed patient's voluntary movement. The prediction used the time windows 120 minutes predict over current time. trained 1500 epochs Adam optimizer initial 1e-3. multi-percentron layer based model. losses were calculated cross-entropy. validated using holdout method 20% datasts. performance assessed receiver operating characteristic curves (AUROC) different events. collected 1-minute record logs 4,716 35 cycles administration, December 2021 February 2022. median age 60(range 35-85), breast cancer(68.6%) most common diagnosis. Among minutes, 277 (5.87%) labeled as non-normal status. Our showed AUROC 0.92 normal status, 1.00 symptoms, 0.82 0.97 proposed shows 97.38% accuracy validation set. This could recognize associated variability obtained device in real world receiving chemotherapy. Further studies predictive capability setting are warranted.