Height growth trajectory in precocious puberty treated with gonadotropin-releasing hormone agonist

Height growth is a critical prognostic indicator in children with precocious puberty receiving gonadotropin-releasing hormone agonist (GnRHa) treatment, yet it requires continuous monitoring and remains difficult to predict prospectively. We aimed to identify height growth trajectories in children with precocious puberty receiving GnRHa treatment and develop explainable prediction models using machine learning.

This longitudinal study included 1,561 children with precocious puberty who received GnRHa treatment at a children's hospital in Hunan Province. Trajectory modeling was used to identify height growth patterns. Logistic regression analysis was used to assess the associations between baseline characteristics and height growth trajectories. Ten machine learning methods were used to construct prediction models.

Two distinct height growth trajectories were identified as declining (n=1,161) and rising (n=400). When the declining trajectory was used as the reference group, multivariable logistic regression analysis revealed that age (OR = 1.19, 95% CI: 1.02-1.39), estradiol >13.5 pg/mL (OR = 1.58, 95% CI: 1.15-2.17), basal follicle-stimulating hormone >2.75 mIU/mL (OR = 2.15, 95% CI: 1.57-2.95), recombinant human growth hormone (rhGH) treatment (OR = 5.10, 95% CI: 3.44-7.58), and other medications (OR = 1.55, 95% CI: 1.18-2.05) were independently associated with the rising trajectory. Among the machine learning models, the neural network demonstrated optimal predictive performance with AUCs of 0.764 (training set) and 0.762 (validation set).

Our findings provide a predictive framework to help achieve a trajectory consistent with the expected height growth for children with precocious puberty during GnRHa treatment. The height growth trajectories constructed using height Z-scores in this study cannot be used to capture final adult height.