JMIR Publications

ABSTRACT

Background:

Central precocious puberty (CPP) in girls seriously affects the physical and mental development in childhood. The diagnosis method, Gonadotropin releasing hormone (GnRH) or GnRH analogue (GnRHa) stimulation test is expensive and makes patients uncomfortable with repeated blood sampling.

Objective:

We combined multiple CPP-related features and constructed machine learning models to predict response to the GnRHa stimulation test.

Methods:

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.

Results:

Both of the XGBoost and Random Forest models achieved good performance in distinguishing positive and negative responses, with the AUC ranging from 0.88 to 0.90, the sensitivity from 77.91% to 77.94% and the specificity from 84.32% to 87.66%. Basal serum luteinizing hormone, follicle-stimulating hormone, and insulin-like growth factor-I are the three most important factors. In the interpretable models of LIME, above variables are demonstrated to have high contributions to the prediction probability.

Conclusions:

The prediction models we developed can help diagnose CPP and may be used as a pre-screening tool before GnRHa stimulation test.