A LASSO-derived prediction model for assessing the risk of lymph node metastasis in T1 and T2 epithelial ovarian cancer: an international retrospective cohort study.

Abstract

The necessity of lymphadenectomy remains controversial in patients with early ovarian cancer. This study aimed to estimate the prevalence of lymph node metastasis (LNM) in patients with apparent local epithelial ovarian cancer (EOC) and to develop a risk model for the prediction of LNM in local EOC with machine learning algorithms. A cohort of 4110 patients who underwent lymphadenectomy with apparent local EOC were retrospectively evaluated. In the model development, a least absolute shrinkage and selection operator (LASSO)-derived Cox regression with internal 10-fold cross-validation was used to identify the risk factors of LNM. Clinical performance was assessed using the decision curve analysis, and a nomogram-based analysis was used to identify the risk factors for LNM. The incidence rate of LNM was rare in our cohort, at 7.4% (213/2885) in the primary set and 4.8% (59/1225) in the validation set. After feature selection, 10-fold internal validation, and external validation, our risk model performed well in both the discrimination (area under the curve=0.790 in the primary set and area under the curve=0.752 in the validation set) and the Hosmer-Lemeshow tests (P=0.873 in the primary set and P=0.380 in the validation set). Additionally, decision curve analysis demonstrated the superiority of our model for clinical interventions in daily practice. LNM is rare in patients with local EOC. In our study, which is based on machine learning algorithms, we developed a prediction model to identify patients with local EOC who could benefit from lymphadenectomy.