Direct Comparison of Logistic Regression and Recursive Partitioning to Predict Lymph Node Metastasis in Endometrial Cancer

Abstract

The purpose was to compare logistic regression model (LRM) and recursive partitioning (RP) to predict lymph node metastasis in early-stage endometrial cancer. Three models (1 LRM and 2 RP, a simple and a complex) were built in a same training set extracted from the Surveillance, Epidemiology, and End Results database for 18,294 patients who underwent hysterectomy and lymphadenectomy for stage I or II endometrial cancer. The 3 models were validated in a same validation set of 499 patients. Model performance was quantified with respect to discrimination (evaluated by the areas under the receiver operating characteristics curves) and calibration. In the training set, the areas under the receiver operating characteristics curves were similar for LRM (0.80 [95% confidence interval [CI], 0.79-0.81]) and the complex RP model (0.79 [95% CI, 0.78-0.80]) and higher when compared with the simple RP model (0.75 [95% CI, 0.74-0.76]). In the validation set, LRM (0.77 [95% CI, 0.75-0.79]) outperformed the simple RP model (0.72 [95% CI, 0.70-0.74]). The complex RP model had good discriminative performances (0.75 [95% CI, 0.73-0.77]). Logistic regression model also outperformed the simple RP model in terms of calibration. In these real data sets, LRM outperformed the simple RP model to predict lymph node metastasis in early-stage endometrial cancer. It is therefore more suitable for clinical use considering the complexity of an RP complex model with similar performances.