Prediction Models for Girls with Suspected Central Precocious Puberty Using Machine Learning Algorithms

Abstract

Central precocious puberty (CPP) in girls is a common pediatric endocrine disease, which seriously affects the physical and mental development in childhood, and significantly increases the risk of cervical or breast cancer in adulthood. Since children of false puberty often present with similar clinical symptoms as those of CPP, it is important to differentiate CPP from false puberty at diagnosis. Gonadotropin releasing hormone (GnRH) or GnRH analogue (GnRHa) stimulation test is used to diagnose CPP. However, they are expensive and make patients uncomfortable with repeated blood sampling. Although previous studies have made great efforts to solve this problem, it is still open. Our study aims to combine multiple CPP-related features and construct machine learning models to replace the GnRHa stimulation test. We leveraged clinical and laboratory data of 1,757 girls performed with GnRHa test, to develop XGBoost and Random Forest classifiers for prediction of response to GnRHa test. Meanwhile, the local interpretable model-agnostic explanations (LIME) algorithm was used to the black-box classifiers to increase their interpretability. The XGBoost classifier with 19 variables achieved the optimal performance with a specificity of 85.39%, a sensitivity of 77.94% and an AUC of 0.89. Basal serum luteinizing hormone, follicle-stimulating hormone, insulin-like growth factor-I, prolactin, growth hormone, height, uterine and ovarian volumes are factors with high feature importance. In the interpretable models of LIME, above variables are demonstrated to have high contributions to the prediction probability. The prediction models we developed can help diagnose CPP in a certain extent. Funding: This work was supported by Guangzhou Institute of Pediatrics, Guangzhou Women and Children’s Medical Center (Grant NO. KCP-2016-002). Conflict of Interest: The authors have declared that no competing interests exist. Ethical Approval Statement: The study design was approved by the appropriate ethics review boards.