A model based on dynamic hematologic parameters to predict short term clinical events in pediatric acute...

Acute lymphoblastic leukemia (ALL) treatment is frequently complicated by infections, emergency visits, and therapy interruptions, yet early prediction of short-term clinical deterioration remains challenging. Traditional prognostic markers rely on static laboratory values, whereas dynamic hematologic fluctuations may provide earlier warning signals. This study presents and internally validates a clinically applicable prediction model based on dynamic hematologic parameters and clinician-documented symptoms for predicting short-term (7-day) clinical events in children with ALL.

Included in this retrospective study were 44 pediatric ALL patients treated with Berlin-Frankfurt-Münster-based protocols between January 2023 and June 2025. Weekly observation units were created by aggregating complete blood count values and clinician-documented symptoms. Dynamic hematologic indices included mean absolute neutrophil count (ANC), coefficient of variation (ANC-CV), and time in target range (ANC-TTR). The composite outcome was defined as any of the following occurring within 7 days: unplanned emergency visit, ≥48-h chemotherapy interruption, or infection requiring systemic antibiotics. Mixed-effects logistic regression was used to account for within-patient clustering. Model performance was assessed using discrimination, calibration, decision curve analysis, and bootstrap internal validation.

A total of 1136 weekly observations were analyzed. Composite clinical events occurred in 32.3% of weeks. Event weeks demonstrated lower ANC, higher ANC-CV, reduced ANC-TTR, lower hemoglobin levels, and higher symptom burden (all p <0.01). In the hematology-only model, ANC, ANC-CV, ANC-TTR, hemoglobin levels, and platelet counts were independent predictors (AUROC = 0.77). Adding the symptom score improved discrimination (AUROC = 0.83) and calibration. Decision curve analysis demonstrated greater net clinical benefit for the combined model across threshold probabilities of 10-40%.

Dynamic hematologic trajectories and clinician-documented symptoms enable accurate early prediction of short-term clinical events in pediatric ALL. This low-cost, accessible prediction model may support individualized risk stratification and proactive supportive care.