The Construction and Validation of KCP-Based Chemoresistance Predictive Model for Cervical Cancer

Abstract Objective: This study aimed to determine the diagnostic threshold of serum KCP for predicting chemoresistance in cervical cancer and to establish and externally validate a relevant prediction model. Additionally, computational drug virtual screening was performed to identify candidate compounds that may reverse KCP-mediated chemoresistance. Methods: This ambispective cohort study enrolled patients with FIGO 2018 stage IB3–IVA cervical cancer who received paclitaxel–platinum neoadjuvant chemotherapy. The retrospective cohort (January 2013–January 2023; n = 158) served as the training set, and the prospective cohort (January 2023–November 2025; n = 130) served as the external validation set. Serum KCP levels were measured by ELISA. Predictive performance was evaluated using receiver operating characteristic (ROC) curve analysis, Kaplan–Meier survival analysis, Cox proportional hazards regression, and nomogram construction with decision curve analysis (DCA). Computational drug screening was performed using DrugReflector based on transcriptomic signatures of chemoresistant versus chemosensitive cervical cancer states from the TISCH single-cell database. Results: A total of 288 patients were enrolled (185 chemosensitive, 103 chemoresistant). The optimal diagnostic threshold of serum KCP for predicting chemoresistance was 24.59 ng/L (sensitivity 88.9%, specificity 77.8%, AUC = 0.707) in the training cohort. Independent risk factors for recurrence after neoadjuvant chemotherapy included serum KCP levels, SCCA, tumor shrinkage, depth of stromal infiltration, lymph node metastasis, and degree of tumor differentiation. A predictive nomogram integrating these variables achieved an AUC of 0.762 in the training cohort but showed substantially reduced performance in the validation cohort (AUC = 0.535, 95% CI: 0.414–0.656), indicating limited external validity. Drug virtual screening identified 9,597 candidate compounds, with the top-ranked drugs exhibiting high logit scores and predicted probabilities, providing a foundation for future experimental validation. Conclusion: KCP was identified as a promising serum biomarker for paclitaxel resistance in cervical cancer with an optimal cutoff of 24.59 ng/L. However, the KCP-integrated nomogram demonstrated insufficient external validity, highlighting the necessity of prospective validation and consideration of population heterogeneity in predictive model development. Computational drug screening identified candidate compounds that may inform future therapeutic strategies for KCP-mediated chemoresistance.