A predictive model for low bone mass in pediatric and adolescent patients with transfusion-dependent beta-thalassemia

Patients with transfusion-dependent beta-thalassemia (TDT) frequently experience osteoporosis, and low bone mass (LBM) is particularly common. However, there is no accurate predictive model for LBM risk in the pediatric and adolescent TDT cohort. This study aimed to create a predictive model to assess LBM risk in this specific population.

Retrospective demographic and laboratory data of pediatric TDT patients were analyzed. The dataset was divided into training and test sets at an 8:2 ratio. Independent predictors of LBM were identified through logistic regression analysis and subsequently incorporated into six machine learning models to develop risk prediction models. Cross-validation was employed to evaluate the models' generalizability, while the test set was used to assess the effectiveness of the final optimal model. The predictive performance of the models was evaluated using receiver operating characteristic curves, calibration curves, and decision curve analysis.

A total of 389 TDT patients were analyzed, and age, insulin-like growth factor 1 (IGF-1) below -2 standard deviation (-2SD), and hypogonadism were identified as predictors of LBM. The Gaussian Naive Bayes (NB) model was optimal, showing areas under the curve (AUCs) of 0.732, 0.734, and 0.730 in training, validation, and test sets, respectively. An online tool was developed to calculate the precise probability of LBM in this population.

We develop and validate a risk prediction model for LBM in pediatric and adolescent TDT patients, which facilitates the efficient identification of high-risk individuals in clinical practice and enables early intervention to prevent disease progression.