SAT-200 From Prediction To Personalization: Machine Learning Vs Traditional Predictors For GH Response In Pediatric GHD — A Prospective Indian Cohort Study

Abstract Disclosure: K. Gupta: None. B. Sharma: None. S. Saran: None. S.K. Mathur: None. Predicting Growth hormone (GH) therapy response remains challenging in diverse real-world pediatric populations. While the Ranke formula is widely used, its generalizability and accuracy in Indian children remain untested. This prospective observational study included 45 GH-naïve children with GHD from a tertiary Indian center. Baseline and 12-month data included height SDS, growth velocity, bone age, IGF-1, weight SDS, pituitary volume (MRI), and GH dose. The primary outcome was growth velocity (≥10 cm/year); Δ height SDS (>0.5) was evaluated as a secondary outcome. Three predictive models were compared using AUC and RMSE. GH dose-response simulations evaluated individualized dosing potential. The results showed a mean height SDS improvement from –3.15 ± 0.98 to –2.28 ± 0.89 (Δ 0.87 ± 0.35). The mean growth velocity was 10.4 ± 1.9 cm/year. Good response (≥10 cm/year) was observed in 89% children. IGF-1 SDS and bone age delay were significantly associated with height gain (p < 0.01). Younger age and higher GH dose were also predictive of a better response (p < 0.05). Pituitary volume was not independently predictive. Model comparisons showed Random Forest: AUC = 0.84, RMSE = 0.35 SDS (1.78 cm/year); Linear Regression: AUC = 0.74, RMSE = 0.43 SDS (2.21 cm/year); Ranke Formula: AUC = 0.72, RMSE = 0.46 SDS (2.41 cm/year). Dose simulations showed minimal benefit from escalation suggesting standard GH doses were adequate in most cases.This is the first prospective Indian study comparing ML, traditional regression and published formula-based models in GHD. ML outperformed other models in predicting GH response. IGF-1 SDS, bone age delay, and GH dose emerged as key predictors. These findings support the use of AI based tools trained on local data for early prognostication and personalization of GH therapy in children with GHD. Presentation: Saturday, July 12, 2025