Helena-Jamin Ly: Advanced technology can improve growth hormone therapy for children
Children with growth hormone deficiency are at risk of impaired growth. The administration of growth hormone is a key aspect of treatment, though it poses challenges. This thesis explores improvements in diagnostics and the potential of prediction models, artificial intelligence (AI), and machine learning to optimize treatment for these children.
HELENA-JAMIN LY
Dissertation defense: 4 April 2025 (click for details)
Doctoral thesis: Applying prediction models and AI to optimize growth hormone therapy in children with short stature
Research area: Pediatrics
Sahlgrenska Academy, Institute of Clinical Sciences
Diagnosing growth hormone deficiency in children can be complex, especially in milder cases. Currently, no standardized diagnostic method exists, meaning that assessments rely on an overall evaluation of multiple tests - a process that is often time-consuming and subjective.
"The treatment is costly and requires daily injections, making it crucial to identify the right children for treatment. The next challenge is optimizing the dosage by administering the appropriate amount of growth hormone, as individual sensitivity varies," says Helena-Jamin Ly, pediatrician at the Pediatric Endocrinology Clinic, Queen Silvia Children's Hospital and a doctoral student at the Institute of Clinical Sciences.
Children’s sensitivity to treatment varies, and incorrect dosing can have consequences - excessive doses may pose risks, while insufficient doses result in inadequate growth. This thesis examines how prediction models (mathematical models that forecast outcomes based on data), artificial intelligence (AI), and machine learning can optimize treatment for children receiving growth hormone therapy.
Laying the foundation for more personalized care
Research evaluations show that prediction models effectively identify children who benefit from growth hormone therapy.
"A Nordic study showed that 28 percent of children with growth hormone deficiency had a poor response to treatment. However, with the help of prediction models, we were able to reduce this percentage to below two percent. We have also developed new machine learning-based models to optimize dosing, enabling a more precise, safe, and cost-effective treatment."
These results could lead to improved growth outcomes and increased safety for patients.
A rewarding and challenging doctoral project
"Exploring AI has been exciting—it's an entirely new world for me, and I see great potential for its application in medical research. The biggest challenge has been learning and understanding AI, a complex and unfamiliar field, but at the same time, it has been a valuable skill for developing better clinical tools."
Text: Susanne Lj Westergren
Above: Figure 9, page 27 in thesis: The plot illustrates how Explainable Boosting Machine (EBM), used in this dissertation, visualize a variable’s impact on prediction with uncertainty intervals. This highlights AI’s potential for transparent and interpretable clinical decision-making..