Use of clinical variables for preoperative prediction of lymph node metastasis in endometrial cancer

Objective: Endometrial cancer is the most common gynaecological cancer, and most patients are identified during early disease stages. Noninvasive evaluation of lymph node metastasis likely will improve the quality of clinical treatment, for example, by omitting unnecessary lymphadenectomy. Methods: The study population comprised 611 patients with endometrial cancer who underwent lymphadenectomy at four types of institutions, comprising seven hospitals in total. We systematically assessed the association of 18 preoperative clinical variables with postoperative lymph node metastasis. We then constructed statistical models for preoperative lymph node metastasis prediction and assessed their performance with a previously proposed system, in which the score was determined by counting the number of high-risk variables among the four predefined ones. Results: Of the preoperative 18 variables evaluated, 10 were significantly associated with postoperative lymph node metastasis. A logistic regression model achieved an area under the curve of 0.85 in predicting lymph node metastasis; this value is significantly higher than that from the previous system (area under the curve, 0.74). When we set the false-negative rate to ∼1%, the new predictive model increased the rate of true negatives to 21%, compared with 6.8% from the previous one. We also provide a spreadsheet-based tool for further evaluation of its ability to predict lymph node metastasis in endometrial cancer. Conclusions: Our new lymph node metastasis prediction method, which was based solely on preoperative clinical variables, performed significantly better than the previous method. Although additional evaluation is necessary for its clinical use, our noninvasive system may help improve the clinical treatment of endometrial cancer, complementing minimally invasive sentinel lymph node biopsy.