Machine Learning Prediction of Growth Hormone Response in Children Non-Growth Hormone-Deficient Short Stature

Area of Science:

• Pediatrics
• Endocrinology
• Biostatistics

Background:

• Height gain under recombinant human growth hormone (rhGH) treatment shows significant variability in children with short stature.
• Accurate prediction of treatment response is vital for tailoring individualized pediatric care plans.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting rhGH treatment response in children with Idiopathic Short Stature (ISS) and Small for Gestational Age (SGA).
• To identify key predictors of rhGH response and determine the optimal time point for accurate classification.

Main Methods:

• Utilized routinely collected clinical data from Bambino Gesù Children's Hospital.
• Developed and evaluated ML models using data updated at baseline, end of year one, and end of year two of rhGH treatment.
• Classified responders into poor, mild, or good categories based on height gain.

Main Results:

• Response classification accuracy was highest at the end of the first treatment year.
• Key predictors included perinatal size, age at treatment initiation, mid-parental target height, baseline stature, growth trajectory, and initial rhGH dose.
• ML models demonstrated effectiveness in categorizing treatment response.

Conclusions:

• The first year of rhGH therapy is a critical checkpoint for predicting long-term response in children with ISS and SGA.
• ML-driven predictions can guide clinical decisions regarding rhGH dose optimization, adherence support, and further evaluations.
• This approach moves beyond simple treatment continuation or discontinuation, enabling more personalized pediatric endocrine care.