Machine learning approach to standard clinical and biochemical parameters for prediction of pathological Transcranial Doppler in pediatric sickle cell disease

Background: Children with sickle cell disease (SCD) are at risk for cerebrovascular complications, with abnormal transcranial Doppler (TCD) velocities serving as predictors of stroke. Reliable laboratory correlates of pathological TCD remain uncertain. This study applied a machine learning approach to identify hematologic and biochemical predictors of abnormal TCD velocities in pediatric SCD, aiming to improve risk stratification and clarify underlying pathophysiologic mechanisms. Methods: . We conducted a retrospective study including 133 pediatric SCD patients (HbSS, HbS/β 0 , HbS/β + , HbSC) monitored between January 2015 and March 2024. A Gradient Boosting Binary Classification (GBbinclass) model was implemented using the H2O framework on the complete dataset, including demographic, hematologic, biochemical, and treatment variables. Model performance was evaluated by five-fold stratified cross-validation. Predictive accuracy was assessed by the area under the receiver operating characteristic curve (AUC), area under the precision–recall curve (AUCPR). The Shapley additive explanations (SHAP) method was used for visualizing model characteristics. Results: . During 3190 patient-years of observation, 447 TCDs were analyzed, of which 45 (10.1%) showed pathological velocities (≥170 cm/s). The optimized model achieved an AUC of 0.776 and an AUCPR of 0.360, indicating moderate discrimination and improved precision relative to baseline prevalence (0.11). Reticulocyte and neutrophil counts, hemoglobin, fetal hemoglobin, platelet count, and age were the most influential predictors, while ongoing therapy exerted a protective effect. Conclusions: . This exploratory study supports the feasibility of machine learning to model cerebrovascular risk in pediatric SCD, offering clinically interpretable insights that may complement standard screening and guide individualized prevention strategies.