A clinico-radiomic nomogram for preoperative prediction of guideline-defined indications for postoperative adjuvant therapy in early-stage cervical cancer

Abstract

To develop and validate an integrated clinico-radiomic nomogram for the preoperative prediction of guideline-defined indications for postoperative adjuvant therapy in patients with FIGO stage IB–IIA cervical cancer. A total of 200 patients who underwent radical hysterectomy were retrospectively included and divided into training ( n = 141) and validation ( n = 59) cohorts. Radiomic features were extracted from preoperative contrast-enhanced CT images and selected through a multistep process including reproducibility assessment, correlation analysis, and LASSO regression. Clinical variables were identified using univariable and multivariable analyses. Three models—a clinical model, a radiomics model, and a combined model—were developed using multivariable logistic regression. Model performance was evaluated in terms of discrimination, calibration, and clinical utility. The combined model achieved AUCs of 0.901 (95% CI, 0.851–0.952) in the training cohort and 0.865 (95% CI, 0.765–0.965) in the validation cohort. Although these AUCs were higher than those of the clinical model, the differences were not statistically significant in either cohort (training: ΔAUC = 0.024, P = 0.079; validation: ΔAUC = 0.033, P = 0.253). However, integrated discrimination improvement (IDI) favored the combined model in both the training cohort (IDI = 0.061, P = 0.002) and the validation cohort (IDI = 0.095, P = 0.019), while decision curve analysis suggested favorable net benefit across a broad range of evaluated threshold probabilities. The clinico-radiomic nomogram showed favorable overall predictive performance for preoperative estimation of guideline-defined indications for postoperative adjuvant therapy. However, the incremental improvement in AUC over the clinical model was modest and not statistically significant. The added value of radiomics may lie primarily in refining individual risk classification and decision support, as suggested by IDI and decision curve analysis, and requires confirmation in external cohorts.