Fertility-sparing treatments decision in patients with endometrial cancer based on machine learning

Although many studies have been recently performed on fertility-sparing treatments in patients with endometrial cancer (EC) and endometrial atypical hyperplasia (EAH), most of the corresponding studies were retrospective and small sample research. However, it is essential to more thoroughly assess the necessity of hysterectomy in EC patients using current accumulated experience. With the development of machine learning (ML), it has been gradually integrated into oncologic research but seldom applied to predict the efficacy of hysterectomy due to an insufficient number of patients who did not undergo hysterectomy, leading to a learning imbalance. Thus, the commonly used machine learning models cannot provide satisfying performance. In this study, we aimed to develop ML models to predict whether hysterectomy is necessary for EC patients and help gynecologists determine the possibility of fertility-preserving treatment in EC patients. A clinical dataset of 1534 women with EC was analyzed. The Borderline -SMOTE algorithm was employed to solve imbalanced learning issues. Then, the Adaptive Boosting (AdaBoost) algorithm, which is less susceptible to overfitting than other machine learning algorithms, was used to build a high-performance ensemble classification model. The findings indicated that the method outperformed conventional machine learning models and provided a realistic strategy to make fertility-preserving treatment decisions. The proposed model provides a platform for physicians to precisely predict the efficacy of fertility-sparing therapy in EC patients, allows gynecologists to select the optimal treatment for a patient, and reduces resource waste and risks of overtreatment.