Exploring Machine Learning for Predicting Peripheral and Central Precocious Puberty Through Cross-Hospital Validation

Area of Science:

• Pediatric Endocrinology
• Machine Learning in Healthcare
• Clinical Decision Support Systems

Background:

• Precocious puberty (PP), encompassing Peripheral Precocious Puberty (PPP) and Central Precocious Puberty (CPP), poses diagnostic challenges in pediatric endocrinology.
• Delayed diagnosis of PP can lead to suboptimal treatment outcomes.

Purpose of the Study:

• To develop and validate machine learning models for predicting and differentiating between PPP and CPP.
• To assess the generalizability of different machine learning models across diverse datasets.

Main Methods:

• Utilized Random Forest (RF), Gradient Boosting Machine (GBM), and Extreme Gradient Boosting (XGB) models.
• Extracted 12 clinical features from electronic medical records (EMRs) for model training and validation.
• Performed internal validation on TMUH data and external validation on WFH data.

Main Results:

• XGB achieved the highest sensitivity (0.88) and AUC (0.86) in internal validation.
• RF demonstrated superior generalizability in external validation, with a sensitivity of 0.91 and AUC of 0.89.
• RF showed robustness for cross-hospital implementation.

Conclusions:

• Machine learning models show significant potential for improving the early diagnosis of precocious puberty.
• The Random Forest model is a robust choice for predicting and differentiating PPP and CPP in real-world clinical settings.