AI-driven transcriptomic biomarker discovery for early identification of pediatric deterioration in Acute Care

Abstract

Early detection of juvenile clinical deterioration in acute care settings remains a significant problem in modern healthcare. This paper presents an AI-powered predictive analytics platform that combines transcriptome biomarker signals with structured vital signs, laboratory data, and unstructured clinical notes to improve early warning capabilities. The system uses ClinicalBERT to extract insights from clinical narratives, XGBoost to analyze tabular clinical information, and long short-term memory (LSTM) networks to simulate temporal dynamics. A meta-classifier combines multimodal data to produce real-time risk ratings for clinical deterioration. The performance evaluation utilizing five-fold cross-validation showed great accuracy, with an AUROC of 0.91, AUPRC of 0.83, and an average early warning lead time of 5.6 hours. Predictive markers included higher lactate levels, heart rate patterns, SpO₂ variability, and transcriptome signals indicating systemic inflammatory activation. Ablation investigations proved the importance of multimodal data fusion in increasing prediction robustness. The suggested strategy provides a scalable, interpretable, and high-performing hospital integration system that enables biomarker-informed, precision-based pediatric intervention options.