The ICPPE score: a multimodal machine learning framework for pretreatment stratification of PD-L1 expression in cervical cancer

Programmed cell death ligand-1 (PD-L1) expression guides the selection of patients with cervical cancer for immune checkpoint inhibitor therapy; however, its assessment currently relies on immunohistochemistry of invasive biopsy specimens and is constrained by intratumoral heterogeneity and sampling bias. This study aimed to develop and validate a multimodal machine learning framework (In-silico Cervical Cancer PD-L1 Positivity Expression [ICPPE] Score) for pretreatment stratification of PD-L1 expression in cervical cancer. This retrospective-prospective study analyzed 386 patients divided into training ( n = 270) and temporally independent prospective validation ( n = 116) cohorts. Radiomic features extracted from multiparametric MRI were integrated with clinicopathological data. Five machine learning algorithms (RF, SVM, DT, XGBoost, and KNN) were evaluated to identify the optimal architecture for deriving the ICPPE score. The performance was assessed using receiver operating characteristic (ROC) curves, decision curve analysis (DCA), net reclassification improvement (NRI), and integrated discrimination improvement (IDI). The multimodal XGBoost architecture, incorporating the pathological grade, parametrial infiltration, ADCmin, and radiomics signature, showed favorable discrimination among the algorithms evaluated. It achieved areas under the curve (AUCs) of 0.830 (95% confidence interval [CI]: 0.755–0.905) and 0.820 (95% CI: 0.741–0.899) in the training and validation cohorts, respectively. The derived ICPPE score demonstrated a significant incremental value over the standard models, with substantial NRI (0.450–0.485, P < 0.05) and IDI (0.138–0.152, P < 0.05). The DCA indicated that the ICPPE framework provided the highest net clinical benefit across the relevant thresholds. The ICPPE score serves as a robust, multimodal tool for pretreatment PD‑L1 probability estimation, which may support risk stratification and help prioritize patients for confirmatory PD‑L1 immunohistochemistry.