LG Chem has built an AI model that predicts how much a child will grow in the first year of growth hormone treatment, with a margin of error under 2 centimeters.

The Korean company unveiled the findings at the Joint Congress of the European Society for Paediatric Endocrinology (ESPE) and the European Society of Endocrinology (ESE), held May 10 to 13 in Copenhagen, Denmark.

The goal is to bring clinical certainty to a treatment space where individual growth outcomes are often unclear and difficult to explain to parents, LG Chem said in a Wednesday release.

Trained on data from 3,045 children and tested on 550 more, the model uses a weighted ensemble of four machine learning algorithms including TabNet, XGBoost, LightGBM, and CatBoost to predict individual responses to growth hormone therapy across four conditions: growth hormone deficiency, idiopathic short stature, small for gestational age and Turner syndrome, a genetic disorder that affects only girls.

The AI model delivered a root mean squared error of 1.95 centimeters in predicting first-year height gain and achieved a coefficient of determination, or R², of 0.98. That level of precision significantly outperformed traditional regression-based approaches, according to LG Chem.

While TabNet delivered the strongest performance in predicting outcomes at years two and three, the ensemble model showed more consistent accuracy across the full three-year period. 

Initial height, weight, diagnosis, and sex emerged as the most predictive variables across all cases.

“By combining clinical records with ensemble machine learning, we’ve made it possible to forecast short-term growth with far greater accuracy,” said Professor Shim Young-suk of the pediatrics department at Ajou University Hospital, who co-developed the model with Choung Ji-yeon, LG Chem’s lead data scientist in its Digital Transformation unit.

The data powering the model comes from the LG Growth Study, a non-interventional registry launched in 2012 to track the safety and efficacy of LG Chem’s Eutropin (somatropin). The study aims to enroll 10,000 Korean children by 2032 and remains the country’s largest pediatric rhGH database.

LG Chem said it plans to further refine the model for use in clinical settings, giving doctors a tool to visualize growth percentile changes and help families understand what to expect.

“This model is especially meaningful because it reflects the realities of Korean pediatric data,” said Yoon Soo-young, head of corporate innovation and life sciences at LG Chem. “It’s built for the realities of real patients and real decisions.”