Artificial Intelligence–Based Prediction of Chemotherapy Toxicity Using Clinical and Molecular Data.

Abstract

Background: Chemotherapy-related toxicity remains a major cause of dose reduction, treatment delay, hospitalization, and treatment discontinuation. Conventional risk-assessment methods may not adequately capture complex interactions among clinical, laboratory, treatment-related, and pharmacogenomic factors. Artificial intelligence (AI) offers an opportunity for multidimensional prediction of treatment toxicity. Materials and Methods: This methodological original-research framework evaluated 600 adult patients receiving cytotoxic chemotherapy. Clinical characteristics, baseline laboratory parameters, treatment-related factors, and pharmacogenomic variables, including DPYD and UGT1A1, were incorporated into predictive models. Severe toxicity was defined as grade ≥3 chemotherapy-related toxicity. Logistic regression, random forest, gradient boosting, and extreme gradient boosting (XGBoost) models were compared. Model performance was assessed using area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, specificity, and F1 score. Results: Severe chemotherapy toxicity occurred in 188 (31.3%) patients. Patients experiencing severe toxicity were older and demonstrated poorer performance status, lower hemoglobin, absolute neutrophil count, serum albumin, and renal function. XGBoost demonstrated the highest predictive performance, with an AUROC of 0.89, accuracy of 84%, sensitivity of 79%, and specificity of 86%. Incorporation of molecular variables increased AUROC from 0.85 to 0.89. Conclusion: Integration of clinical and molecular data using AI may improve individualized prediction of chemotherapy toxicity. External prospective validation is necessary before clinical implementation.