Consciousness assessment in prehospital deterioration risk stratification: A foundation for automated AVPU evaluation

Abstract

Early recognition of clinical deterioration in prehospital settings is essential to ensure timely escalation, reduce adverse events, and optimize emergency medical resources. The National Early Warning Score (NEWS) is a widely adopted framework to support deterioration risk stratification based on routinely collected physiological parameters.Within NEWS, the level of consciousness, commonly assessed using the AVPU scale (Alert, Verbal, Pain, Unresponsive), has a high impact because any deviation from alertness carries a strong escalation weight. Nonetheless, the AVPU assessment is observational and influenced by operational constraints, particularly in the dynamic context of Emergency Medical Services (EMS).This dissertation presents a proof of concept designed to evaluate the role of AVPU in prehospital deterioration risk stratification, using NEWS as the reference clinical framework. Two complementary approaches were developed**: a** video-based methodology**,** defined through clinically grounded scenarios with model patient interactions**, and a sensor-data methodology**. The videos were created and annotated under a medical rationale to enable the assessment of computer vision methods for recognizing responsiveness patterns aligned with AVPU categories. The sensor-data methodology was developed to analyze physiological measurements, their association with consciousness**,** and the NEWS escalation thresholds.Building on these complementary models, the dissertation proposes a modular multimodal architecture integrating video-derived responsiveness cues and physiological analytics within existing EMS workflows.Rather than claiming full automation of consciousness assessment, this work provides a structured methodological and architectural foundation for future prospective validation and the responsible integration of multimodal decision-support systems in prehospital emergency care.