From Diagnosis to Clinical Decision Support: An AI Framework for Pediatric ICU Using Electronic Health Records
Year: 2026
Type: dissertation
Abstract: Iatrogenic withdrawal syndrome (IWS) is a common and clinically significant complication in pediatric intensive care units, often arising from prolonged exposure to opioids and sedatives. Current assessment methods rely on manual, intermittent scoring systems that are labor-intensive and may delay timely intervention. This dissertation presents a unified artificial intelligence framework for automated diagnosis, early prediction, and clinical decision support for IWS using electronic health records. First, a machine learning–based computable phenotype is developed to estimate real-time withdrawal risk from structured clinical and medication data. Second, an explainable deep learning model, FIR-LSTM, is proposed to forecast near-term IWS risk using longitudinal patient trajectories. Finally, these models are integrated into a safety-gated clinical intelligence system that combines predictive modeling, deterministic calculations, and retrieval-grounded large language model outputs to support clinician decision-making. Together, this work advances the development of interpretable, real-time, and clinically deployable AI systems for pediatric critical care, enabling earlier detection of withdrawal risk and supporting safer, data-driven treatment strategies.
Source: VTechWorks (Virginia Tech)
Author: Song, Haoqiu
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Children's National Hospital
OpenAlex ID: W7161357768
