Prediction of myelosuppression risk in postoperative cervical cancer patients undergoing concurrent chemoradiotherapy
Qingkai Wang1, Yaling Li2, Liran Shen1
1Department of Laboratory Medicine, Shanxian Central Hospital, Heze, China.
Objective:
By integrating clinical features and pelvic bone marrow dose-volume parameters, this study developed and compared multiple machine learning models to predict the risk of bone marrow suppression (BMS) in postoperative cervical cancer patients receiving concurrent chemoradiotherapy (CCRT).
Methods:
A total of 415 cervical cancer patients who received postoperative CCRT at Siyang Hospital and Shanxian Central Hospital between March 2022 and April 2025 were retrospectively enrolled. The primary outcome was BMS. Candidate predictors included baseline clinical characteristics, laboratory indices, and pelvic bone marrow dose-volume parameters. Feature selection was performed using the Least Absolute Shrinkage and Selection Operator (LASSO). Logistic regression (LR), random forest (RF), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), support vector machines (SVM), and artificial neural networks (ANN) were constructed. Model performance was evaluated in an independent test set using the area under the curve (AUC), calibration curves, and decision curve analysis, and model interpretability was assessed using SHAP.
Results:
Among the 415 patients, 220 (53.0%) developed BMS. Significant differences were observed between patients with and without BMS in lymph node boost irradiation, body mass index (BMI), aspartate aminotransferase (AST), serum albumin (ALB), and low-dose pelvic bone marrow irradiated volume parameters (V10, V20, and V30). LASSO retained BMI, AST, ALB, V10, V20, and lymph node boost irradiation as key predictors. Among the six models, the RF model achieved the best performance (AUC = 0.799). Calibration curves and decision curve analysis demonstrated good calibration and potential clinical net benefit of the RF model.
Conclusion:
Machine learning models integrating routine clinical indicators and pelvic bone marrow dose-volume parameters can effectively predict the risk of myelosuppression in postoperative cervical cancer patients receiving CCRT. The random forest model demonstrated optimal performance and may serve as a practical tool for individualised risk stratification and early intervention.
