Machine learning approaches for enhancing adult height prediction in girls with early-onset and rapidly progressing puberty undergoing GH and GnRHa therapy: a scoping review

Background: Accurate prediction of adult height in girls undergoing treatment for early-onset or rapidly progressing puberty with growth hormone (GH) and gonadotropin-releasing hormone agonist (GnRHa) is essential for guiding clinical decisions and optimizing outcomes. Traditional statistical models, such as Bayley-Pinneau and Tanner-Whitehouse, were developed in homogeneous cohorts and often fail to capture the heterogeneous growth trajectories seen in treated girls aged 7−10 years. This review examines the potential of machine learning (ML) approaches to improve predictive accuracy and clinical applicability in this underrepresented population. Methods: A scoping review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) framework. PubMed, Scopus, and IEEE Xplore databases were searched for peer-reviewed studies published between January 2007 and May 2025. Eligible studies included predictive models of adult height or closely related outcomes in pediatric endocrinology, with a focus on early puberty and GH/GnRHa interventions. Data were extracted on study design, sample characteristics, treatment regimen, baseline risk markers (e.g., bone-age advancement, predicted adult height methods), and predictive methodology (traditional versus ML). Results: Twente studies were included. Traditional models demonstrated variable accuracy, with systematic overestimation in girls with rapidly progressing puberty. ML models, including random forest (RF), extreme gradient boosting (XGBoost), and deep learning, consistently reduced prediction error compared with conventional methods [root mean square error (RMSE) <3.5 cm in some cohorts]. However, most studies were limited to East Asian and European populations, with minimal representation of Middle Eastern cohorts. In addition, variation in treatment timing, adherence, and genetic background complicates direct comparison across studies. While ML demonstrates superior feature integration and adaptability, questions remain regarding optimal algorithm selection for the 7-10 age group, interpretability for clinical trust, and validation in ethnically diverse, real-world settings. Conclusions: ML approaches show promise in enhancing adult-height prediction for girls undergoing GH/GnRHa therapy, but evidence remains constrained by limited diversity and short-term validation. Future research should prioritize ethnically representative cohorts, interpretable models, and prospective clinical trials to establish predictive validity and improve real-world decision-making in pediatric endocrinology.