Prognostic utility of systemic immune-inflammation markers in locally advanced cervical cancer undergoing radical radiotherapy

PURPOSE: To improve survival prediction in locally advanced cervical cancer (LACC) post-radiotherapy, we developed and compared a nomogram and machine learning (ML) models using immune-inflammatory markers. MATERIAL AND METHODS: Clinical data from patients with LACC who received radical radiotherapy at our department between 2016 and 2019 were retrospectively analyzed. Key variables were selected using Spearman correlation and Boruta analysis. A Cox regression-based nomogram was developed and validated. ML models were also used to predict survival, with performance evaluated by accuracy, the area under the receiver operating characteristic curve (AUC), precision, recall, and the F1-score. The performance of the optimal ML model and the nomogram was compared using time-dependent predictive accuracy. RESULTS: A cohort of 300 patients with LACC was randomly allocated to training and validation cohorts at a 7:3 ratio. The nomogram and ML models incorporated immune-inflammatory markers and clinical variables. The nomogram demonstrated strong discriminative ability, calibration, and clinical utility, as evidenced by AUC, calibration plots, and decision curve analysis. Among ML models, logistic regression achieved superior performance (accuracy = 0.933, recall = 1.000, F1-score = 0.938). Time-dependent accuracy analysis revealed dynamic predictive performance over 12-60 months. The nomogram's predictive accuracy improved over time, reaching 0.611 at 53-59 months, while logistic regression peaked earlier (0.911 at 12-24 months) and consistently outperformed the nomogram. CONCLUSION: Both nomogram and ML models provide valuable prognostic insights for LACC. The logistic regression ML model showed superior predictive accuracy compared to the nomogram during 1-5 years of follow-up.