Identifying patients with apparent early-stage ovarian cancer at very low risk of lymph node metastasis: a multicenter preoperative prediction model.

OBJECTIVE

Systematic pelvic and para-aortic lymphadenectomy is recommended for surgical staging in presumed early-stage epithelial ovarian cancer, although lymph node metastases occur in only a minority of patients and the procedure is associated with significant morbidity. We aimed to develop and validate a predictive model based on variables available prior to lymphadenectomy within a two-step surgical strategy for identifying patients at very low risk of lymph node metastasis.

METHODS

We conducted a multicenter retrospective study including patients with presumed early-stage high-grade epithelial ovarian cancer treated in 12 French centers between January 2018 and December 2023. All patients underwent comprehensive surgical staging including systematic pelvic and para-aortic lymphadenectomy. Three prediction algorithms were evaluated: logistic regression, random forest, and extreme gradient boosting (XGBoost). Model performance was assessed using repeated stratified 5 × 5 cross-validation. A predefined probability threshold of 0.98 defined very low risk of nodal metastasis (≤2%).

RESULTS

A total of 219 patients were included. Lymph node metastases were identified in 26 patients (12%). Among the evaluated algorithms, XGBoost achieved the best trade-off between safety and clinical applicability. The final model included five preoperative variables: cytology, bilateral ovarian involvement, age ≥60 years, BMI ≥25, and CA-125 ≥ 100 IU/mL. Using the predefined threshold, the model classified 42% of patients as very low risk of lymph node metastasis with only one false-negative prediction.

CONCLUSIONS

This multicenter study supports the feasibility of a simple preoperative model to identify a substantial subgroup of patients with apparent early-stage ovarian cancer at very low risk of lymph node metastasis. If externally validated, this approach may help refine patient selection and reduce unnecessary lymphadenectomy.