ESPE Abstracts

Background: Vitamin D deficiency is a prevalent health concern in children, potentially leading to various health issues. Early identification of children at risk of vitamin D deficiency is crucial for timely intervention and prevention of associated complications. This study aim ed to develop a deep learning algorithm to predict the risk of vitamin D deficiency in children aged 0-12 years using simple laboratory tests.

Methods: A dataset comprising more than 5000 children aged 0-12 years was utilized in this study. The dataset included laboratory tests such as T4, IGF-1, height, bone age, Zn, growth hormone, BW, BA/CA, Age, BMI, and age at diagnosis. Three deep learning models, namely Random Forest Tree, XGboost, and Transformer, were employed for building and training the predictive models. The performance of these models was evaluated based on their accuracy in predicting vitamin D deficiency.

Findings: The deep learning models achieved accuracies of 0.82, 0.82, and 0.84 for Random Forest Tree, XGboost, and Transformer, respectively. The SHAP (SHapley Additive exPlanations) model was applied to interpret the contribution of each laboratory test in predicting vitamin D deficiency. The SHAP analysis revealed that T4, IGF-1, height, bone age, Zn, growth hormone, BW, BA/CA, Age, BMI, and age at diagnosis were the most influential factors, with T4, IGF-1, and height being the three most important measures.

Interpretation: The developed deep learning algorithm demonstrates promising accuracy in predicting the risk of vitamin D deficiency in children using simple laboratory tests. The SHAP model provides a clear and intuitive interpretation of the contribution of each laboratory measure, enabling physicians and patients to understand the key factors influencing vitamin D status. These findings suggest that the proposed algorithm could serve as a valuable tool for early identification of children at risk of vitamin D deficiency, facilitating timely interventions and improving overall health outcomes.