Endocrine Abstracts

JOINT1887

Objective: This study aims to develop a diagnostic prediction model for girls with rapidly progressive central precocious puberty (CPP) using machine learning algorithms.

Methods: CPP girls admitted to the department of endocrinology, genetics and metabolism of the hospital without treatment intervention for a follow-up period of more than 3 months from June 2017 to June 2023 were included retrospectively. These girls were randomly divided into training (70%), validation (15%), and test set (15%). The training set data were used to screen the predictors based on Lasso regression. Validation set data assisted the training set data to adjust the model parameters to establish the rapidly progressive CPP girl diagnosis prediction model based on five machine learning algorithms including Logistic regression, support vector machine, random forest, limit gradient lifting and artificial neural network, and carried out internal validation. The test set data were used for model external validation.

Results: A total of 277 CPP girls were analyzed, of whom 141 (50.9%) were diagnosed with rapidly progressive CPP and 136 (49.1%) were diagnosed with slowly progressive CPP. Lasso regression included five predictors: BMI (kg/m²), group of basal luteinising hormone (LH) (LH<0.20 IU/l, 0.20 IU/l≤LH<0.52 IU/l, LH≥0.52 IU/l), group of insulin like growth factor 1 (IGF-1) SDS (IGF-1 SDS<0.35, IGF-1 SDS≥0.35), and advanced bone age (y). Five machine learning algorithm prediction models were developed. The artificial neural network model has the best performance, with AUC vulues of 0.774 in internal validation and 0.725 in external testing cohort, and the accuracy rate was 76.2% and 69.1%, respectively. When the prediction probability was greater than 0.80 as the diagnostic threshold of rapidly progressive CPP, the specificity of the model in the validation and the test set data were 100.0% and 95.5%. When using a prediction probability less than or equal to 0.30 as the diagnostic threshold for slowly progressive CPP, the model demonstrated sensitivities of 96.2% and 95.5% in the validation and test set, respectively.

Conclusion: This study developed prediction models based on machine learning algorithm for the typing of CPP girls. Internal validation and external testing ensured a good degree of discrimination and calibration of the models, which could assist in the diagnosis of rapidly progressive CPP on the basic of clinical data.

Key words: central precocious puberty, girl, machine learning, prediction model.