A Multi-Modal Machine Learning Framework for Real-Time Clinical Decision Support and Patient Outcome Prognostication

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

Critical care triage decisions in resource-constrained U.S. hospital systems face mounting pressure from rising patient volumes, workforce shortages, and the inherent subjectivity of conventional scoring systems. Current triage approaches, including the Emergency Severity Index and similar protocols, demonstrate significant inter-rater variability and limited capacity to integrate the full spectrum of clinical and operational data available within electronic health records. This study addresses the gap between available data and actionable clinical intelligence by developing and validating a multi-modal machine learning framework that synthesizes structured vital signs, unstructured clinical narratives, and temporal physiological trajectories to support real-time triage decisions. The proposed framework, evaluated on a retrospective cohort of 18,633 unique patients from a large U.S. hospital system, achieved a C-statistic of 0.796 for predicting the composite outcome of ICU admission, emergency response team activation, and mortality . The framework demonstrated a 33% reduction in median time from arrival to initial care area and improved critical care identification from 78.8% to 83.1% compared to conventional triage methods . These findings establish that multi-modal machine learning frameworks can meaningfully enhance triage accuracy, reduce cognitive burden on frontline clinicians, and optimize resource allocation in resource-constrained environments, offering a 

• PDF

Copyright (c) 2026 Abiodun Okunola (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.