Development and exploratory evaluation of a prediction model for oral mucositis in patients receiving CapeOX chemotherapy: a multicenter retrospective study

Abstract

Abstract Oral mucositis (OM) is a clinically significant toxicity in patients receiving CapeOX chemotherapy. However, a prediction model tailored to this population is unavailable. In this multicenter retrospective study, we developed a pragmatic, exploratory and feasibility-based prediction model for OM using routinely collected variables from 1,100 patients with colorectal cancer. Among seven prespecified logistic regression models, the model incorporating CapeOX cumulative exposure (≥ 5 cycles), sex, age, body mass index, smoking history, and platelet count demonstrated the highest discriminative performance in the derivation cohort (ROC–AUC, 0.664). Temporal validation showed limited performance in the validation cohort (ROC–AUC, 0.570), indicating that these routinely available variables captured only part of the inter-individual variability in OM risk. These findings suggest that routinely available clinical variables alone may be insufficient for pretreatment or clinically actionable OM risk prediction, and future models may benefit from integrating oral health-related and treatment-specific factors to enhance predictive accuracy. This study provides exploratory evidence that may inform the future development of more comprehensive and risk-adapted supportive care strategies for patients receiving CapeOX chemotherapy.