ECEESPE2025 ePoster Presentations Growth Axis and Syndromes (132 abstracts)
Development of AI-based growth prediction models for children with growth disorders: a 3-year analysis using the lg growth study
Young Suk Shim 1 , Hae Sang Lee 1 , Young-Jun Rhie 2 , Hyun Wook Chae 3 , Jaehyun Kim 4 , Young Ah Lee 4 , Yoo-Mi Kim 5 , Ja Hye Kim 6 , Moon Bae Ahn 7 , Yong Hee Hong 8 & JIYEON CHOUNG 9
1Ajou University Medical Center, Department of Pediatrics, Suwon, South Korea; 2Department of Pediatrics, Korea University College of Medicine, Ansan, South Korea; 3Department of Pediatrics, Yonsei University College of Medicine, Seoul, South Korea; 4Department of Pediatrics, Seoul National University College of Medicine, Seoul, South Korea; 5Department of Pediatrics, Chungnam National University Sejong Hospital, Sejong, South Korea; 6Department of Pediatrics, Asan Medical Center Children’s Hospital, University of Ulsan College of Medicine, Seoul, South Korea; 7Department of Pediatrics, College of Medicine, The Catholic University of Korea, Seoul, South Korea; 8Department of Pediatrics, Soonchunhyang University Bucheon Hospital, Soonchunhyang University College of Medicine, Bucheon, South Korea; 9Life Sciences. Corporate Innovation. DX Team(AI Development Part), LG Chem, Seoul, South Korea
JOINT521
Background: Growth hormone (GH) treatment is effective in improving growth outcomes in children with various growth disorders, including growth hormone deficiency (GHD), idiopathic short stature (ISS), small for gestational age (SGA), and Turner syndrome (TS). However, individual responses to GH therapy vary, necessitating predictive models to guide personalized treatment strategies. Previous models have primarily focused on short-term outcomes using regression methods. This study aimed to develop machine learning-based models to predict growth outcomes up to 3 years after treatment initiation and improve accuracy through ensemble learning, using data from the LG Growth Study (LGS). Turner syndrome patients were limited to females.
Methods: Prepubertal children with GHD, ISS, SGA, and female children with TS were included in this study. Clinical and demographic features were collected at the screening visit, including baseline height and weight, age, sex, mid-parental height, bone age, diagnosis, and initial GH dose. Machine learning models (TabNet, XGBoost, LightGBM, CatBoost) were used to predict growth outcomes at 1-, 2-, and 3-years post-treatment. A Weighted Ensemble model was constructed using root mean squared error (RMSE)-based weights. Model performance was evaluated using mean absolute error (MAE), mean squared error (MSE), RMSE, mean absolute percentage error (MAPE), and the coefficient of determination (R²).
Results: The Weighted Ensemble model achieved superior accuracy for 1-year predictions with an RMSE of 1.95 and R² of 0.983. TabNet demonstrated better performance for mid-term predictions, achieving RMSEs of 2.975 (R² 0.961) at 2 years and 3.674 (R² 0.937) at 3 years. The Weighted Ensemble maintained overall stability and consistent performance. However, predictions beyond 3 years showed a decline in accuracy across all models.
Conclusion: This study demonstrates that AI-based models can provide accurate growth predictions for children with various growth disorders over the first 3 years of GH treatment. TabNet showed the highest performance in mid-term predictions, while the Weighted Ensemble offered overall stability. Future research should aim to improve long-term prediction accuracy by integrating additional clinical data and advanced time-series methods.
Volume 110
Article tools
My recent searches
My recently viewed abstracts
Authors
Endocrine Abstracts
ISSN 1470-3947 (print) | ISSN 1479-6848 (online)
© Bioscientifica 2025 |
Privacy policy |
Cookie settings
BiosciAbstracts
Bioscientifica Abstracts is the gateway to a series of products that provide a permanent, citable record of abstracts for biomedical and life science conferences.
Find out more