Artificial intelligence system for predicting hand-foot skin reaction induced by vascular endothelial growth factor receptor inhibitors.

Hand-foot skin reaction (HFSR) is a common adverse effect of vascular endothelial growth factor receptor (VEGFR) inhibitors that significantly impacts patients' quality of life. Prevention and management of HFSR require individualized approaches, but risk factors remain unclear. This study aimed to develop artificial intelligence (AI) models to predict grade ≥ 2 HFSR using clinical data and foot sole images from 93 instances of VEGFR inhibitor administration in 76 patients. Image-based, clinical information-based, and ensemble AI models achieved areas under the curve of 0.550, 0.693, and 0.699, respectively. At a high-specificity cutoff, the ensemble AI had a positive predictive value of 0.824, suggesting potential clinical utility for identifying high-risk patients. Feature importance analysis revealed heavier weight, good performance status, lack of prior VEGFR inhibitor exposure, and baseline skin toxicity as risk factors. These findings represent the first AI-based HFSR prediction models and provide insights for preventive interventions, but further accuracy improvements are needed.