Algorithmic Triage: AI-Assisted Decision Support in Mass Casualty and EMS Operations

Mass casualty incidents expose every weakness in the emergency medical system simultaneously. Communication degrades. Resources are overwhelmed. Triage decisions get made in seconds by providers operating under physical and psychological stress that impairs cognitive performance. The question is not whether better tools are needed for MCI management — it is whether AI-assisted decision support can actually deliver in the field.

What follows is a grounded assessment of the current state of AI-assisted triage and decision support in civilian EMS and military medical operations — anchored in available evidence and informed by operational experience.

The Triage Problem: Why Human Performance Fails Systematically

START and SALT remain the operational triage standards, but their limitations are well-documented. Inter-rater reliability is imperfect — studies using simulated MCI scenarios have demonstrated meaningful disagreement on triage category even among experienced providers. Cognitive load degrades performance in a predictable, nonlinear fashion: the provider triaging the 30th patient in a 50-casualty incident is making decisions in a fundamentally different cognitive state than at the outset.

The physiologic inputs that drive triage decisions — respiratory rate, pulse quality, capillary refill, level of consciousness — are inherently rapid assessments subject to estimation error. In darkness, chemical contamination scenarios, or high-noise environments, that error compounds substantially.

AI-assisted triage tools address this by inserting objective, sensor-derived physiologic data into the triage process at the point of care. Wearable sensor arrays combined with machine learning algorithms can generate continuous vital sign streams, aggregate deterioration signals, and flag patients whose physiologic trajectory is worsening — even when their initial triage category assigned lower priority. This is the use case with the most compelling evidence base for field AI application.

Current State of Field AI: What Is Actually Deployed

DARPA and DoD-funded programs have advanced wearable biosensor platforms capable of transmitting continuous physiologic data to a command-level dashboard, giving the medical commander real-time patient status across an entire casualty collection point. Systems integrating Tactical Combat Casualty Care protocols with forward telemedicine platforms represent the current military leading edge of this capability.

In civilian EMS, AI-assisted dispatch prioritization is the most widely deployed application. Platforms that analyze caller characteristics, address history, and real-time call content to optimize unit deployment and predict call severity are operational in several large urban systems. The outcome evidence is early but directionally positive.

Predictive deterioration models for interfacility transport represent an emerging high-value application. This is a population at elevated risk for en-route decompensation, and the ability to identify, before departure, which patients are most likely to deteriorate has direct implications for crew configuration, equipment selection, and destination decisions. Several critical care transport programs are currently piloting this capability.

Command and Medical Decision Architecture

From a medical command perspective, the value proposition of AI in MCI operations is fundamentally about information management. The medical commander's challenge in a large-scale incident is not a scarcity of information — it is an excess of it, arriving faster than it can be processed and acted upon. AI that filters, prioritizes, and presents actionable signals from that information stream has genuine operational value.

AI does not replace the medical commander’s judgment. It changes what that judgment is applied to — shifting the cognitive task from sorting raw data to making decisions from synthesized intelligence. That is a meaningful distinction, and experienced commanders should embrace it.

The military medical community has long understood that the TCCC algorithm is a decision support framework, not a substitute for trained judgment. The same principle applies to AI in mass casualty operations. The provider or commander who understands what the algorithm is doing, what data it draws on, and where its confidence degrades will use it more effectively than one who treats it as an oracle.

Rules of engagement for AI in the field must be developed with the same rigor applied to medical equipment credentialing. What is this tool approved to do? What are its documented failure modes? Who bears responsibility when its output contributes to an adverse outcome? These are not theoretical questions — they are the operational and medicolegal realities that medical directors for EMS and HEMS programs need to resolve before deployment, not after.

Looking Forward: AI and the Future of Prehospital Care

The convergence of wearable sensor miniaturization, edge computing, and low-latency satellite communications is building the technical infrastructure for genuinely capable field AI. Within this decade, it is plausible that a combat medic or paramedic will have access to continuous AI-assisted clinical decision support that aggregates patient physiology, injury pattern recognition, available resources, and transport time to generate real-time treatment recommendations.

The question is not whether that technology will exist. It is whether the medical professionals who deploy it will be prepared to use it effectively, critique it appropriately, and maintain the clinical competency to override it when necessary. That preparation must begin now — before the technology outpaces the doctrine.

—
Dr. Chester “Chet” Shermer, MD, FACEP is a Professor of Emergency Medicine, Medical Director for Critical Care EMS and State Surgeon in the Army National Guard. He is the founder of Global MedOps Command and the creator of AI in Emergency Medicine: Becoming AI Bulletproof.