An interpretable machine learning model for predicting symptomatic pelvic lymphocele after pelvic lymphadenectomy in
Yiyue Wang1, Nenghuan Tang1, Weimin Zhang1
1Department of Obstetrics and Gynecology, Affiliated Hospital of North Sichuan Medical College, Nanchong, Sichuan, China.
Objective:
To develop and validate an interpretable machine learning model using retrospective analysis of clinical characteristics and laboratory data from patients with cervical cancer to predict symptomatic pelvic lymphocele (SPL) following pelvic lymphadenectomy.
Methods:
Clinical data were collected from 221 patients with cervical cancer who underwent pelvic lymphadenectomy at the Affiliated Hospital of North Sichuan Medical College between January 2023 and September 2024. SPL occurred in 44 patients, whereas 177 had asymptomatic lymphoceles or no lymphocele. Univariate analysis identified risk factors for SPL. Four predictive models including Logistic Regression (LR), K-Nearest Neighbors (KNN), Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (XGBoost) were developed. Model performance was assessed using the Area Under the Receiver Operating Characteristic Curve (AUC), Decision Curve Analysis (DCA), and Brier Score. The SHapley Additive exPlanations (SHAP) method was applied to determine feature importance.
Results:
The incidence of SPL among patients with cervical cancer was 19.9%. Among the four models, the KNN model demonstrated relatively better predictive performance in both the training and test sets (training set AUC = 0.952, 95% confidence interval (CI) 0.924-0.980; test set AUC = 0.832, 95% CI 0.692-0.972; Brier Score = 0.118). DCA indicated favorable clinical utility for the KNN model. SHAP analysis identified the most predictive features for SPL as diabetes, surgical approach, preoperative monocyte-to-lymphocyte ratio, preoperative fibrinogen level, and tumor size.
Conclusion:
This interpretable model identified key features associated with SPL following pelvic lymphadenectomy in patients with cervical cancer, yielding preliminary insights that support hypothesis generation for future research.
