Multifactorial construction of low-grade and high-grade endometrial cancer recurrence prediction models.

Abstract

Objective: To analyze independent risk factors for endometrial cancer (EC), a common female cancer globally, and construct individualized prediction models for EC recurrence.

Methods: The EC patients from the medical record system were divided into low-grade (n = 392) and high-grade (n = 183) groups. Immunohistochemical expression of estrogen receptor, progestin receptor, Ki67, and L1 cell adhesion molecule (L1CAM) was detected. Univariate Cox regression, LASSO regression, and stepwise Cox regression were applied for identifying independent risk factors for EC recurrence. The predictive value of the model was verified by using receiver operating characteristics curves, bootstrap method, calibration curves, and decision curve analysis curves.

Results: Multivariate Cox analysis revealed that FIGO (the International Federation of Gynecology & Obstetrics) Stage, progestin receptor, lymphovascular space invasion (LVSI), and tumor size were independent risk factors for low-grade EC recurrence-free survival (RFS), and FIGO Stage, L1CAM, LVSI, and pelvic lymph node status were independent risk factors for high-grade EC. The areas under the curves at 1-, 3-, and 5-year RFS in low-grade and high-grade groups were 0.881/0.825, 0.888/0.853, and 0.807/0.832, respectively. Calibration curves were close to the diagonal, and the decision curve analysis curves were located mostly above the All and None lines in both groups.

Conclusion: The prediction model demonstrates accurate discriminative ability and strong calibration capability. It has high clinical application value and provides decision making information regarding RFS for both low-grade and high-grade EC patients. This may assist in formulating personalized treatment plans, monitoring follow-up strategies, and implementing lifestyle intervention measures.