Conversational AI in Tactical Casualty Care: Baseline GPT-4o Improves Combat Medic Decision-Making (FieldAI)

The aim of the project is to investigate whether the integration of artificial intelligence (AI) support, specifically through the GPT-4 model, enhances the decision-making processes of military medical first responders within the framework of Tactical Combat Casualty Care (TCCC). The study focuses on AI's ability to assist in ventilator settings for injured individuals in combat scenarios, emphasizing improved accuracy and decision-making speed. The project tests the hypothesis that the use of AI can positively impact outcomes without compromising the autonomy of first responders. The results have the potential to optimize patient care in challenging conditions and contribute to the advancement of combat medicine.

Study Overview

• Status: Completed
• Conditions: Artificial Intelligence
• Intervention / Treatment: Other: Combat Medic Decision-Making with and without artificial intelligence assistance

Detailed Description

This study investigates the potential of conversational artificial intelligence (AI), specifically GPT-4, to enhance clinical decision-making in Tactical Combat Casualty Care (TCCC) scenarios. The primary objective is to evaluate whether AI support improves the accuracy and efficiency of ventilator management decisions for combat medics in high-pressure environments without compromising their autonomy.

A prospective, randomized, within-subject study design will be employed. Thirty combat medics from the Czech Armed Forces will participate. Each participant will complete 10 simulated TCCC scenarios: five with AI assistance and five without. Scenarios will be matched for complexity and randomized to control for order effects. Participants will use ChatGPT on handheld devices to simulate real-time AI-assisted decision-making.

In scenarios involving AI assistance, medics will query GPT-4 for support in optimizing mechanical ventilator settings based on patient data, including blood gas results, vital signs, and ventilator parameters.

The primary outcome is the accuracy of ventilator settings as categorized into "excellent," "acceptable," or "failing" based on predefined TCCC standards. Secondary outcomes include decision-making speed and participants' perception of AI's utility, measured through post-scenario surveys.

The findings aim to determine the feasibility of integrating large language models (LLMs) into combat medical care to optimize patient outcomes and support medics under combat conditions. The study seeks to advance the understanding of AI's role in military medicine, providing a foundation for future deployment of fine-tuned AI solutions in TCCC and other critical care scenarios.

This study offers a proof-of-concept evaluation of LLM applications in combat casualty care, with the potential to improve decision-making and inform the development of specialized AI tools for military use.

• Study Type: Interventional
• Enrollment (Actual): 42
• Phase: Not Applicable

Contacts and Locations

Study Locations

• Czechia
  • Praha, Czechia, 16209
    • Military University Hospital Prague

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: Yes

Description

Inclusion Criteria:

• Combat medics actively serving in the Czech Armed Forces
• Completion of standardized Tactical Combat Casualty Care training modules and e-learning on ventilator settings and blood gas interpretation
• Successful passing of pre-tests to ensure a uniform baseline knowledge level.
• Willingness to participate and provide informed consent.
• Availability to complete the full study protocol, including 10 simulated scenarios.

Exclusion Criteria:

• Failure to pass the pre-tests or complete TCCC and ventilator management training
• Prior advanced training or professional certification in critical care or mechanical ventilation that could bias results
• Refusal to provide informed consent or inability to commit to the study schedule

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Other
• Allocation: N/A
• Interventional Model: Single Group Assignment
• Masking: None (Open Label)

Number of Arms

1

Arms and Interventions

Participant Group / Arm

Intervention / Treatment

Experimental: Combat Medic Decision-Making with and without AI Assistance; All participants will complete 10 Tactical Combat Casualty Care scenarios: 5 with AI assistance using GPT-4 for ventilator management and 5 without AI assistance. The crossover design ensures each participant experiences both conditions.

Other: Combat Medic Decision-Making with and without artificial intelligence assistance; Participants will complete 10 simulated Tactical Combat Casualty Care (TCCC) scenarios, with 5 scenarios conducted using AI assistance (GPT-4) and 5 without AI. In AI-assisted scenarios, participants will use GPT-4 to query and optimize ventilator settings based on patient data, while non-AI scenarios rely solely on their clinical judgment.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Accuracy of ventilator settings	Accuracy of ventilator settings as categorized into "excellent," "acceptable," or "failing" based on predefined TCCC standards. Excellent means 2 points, acceptable 1 point and failing 0 point.	1 hour

Other Outcome Measures

Outcome Measure	Measure Description	Time Frame
Perception of artificial intelligence's utility	perception of artificial intelligence's utility, measured through post-scenario survey	1 hour

Collaborators and Investigators

• Sponsor: Charles University, Czech Republic
• Collaborators: Czech Technical University in Prague
• Investigators: Principal Investigator: Michal Soták, M.D., Ph.D., Charles University, Czech Republic

Publications and helpful links

General Publications

• Nemeth C, Amos-Binks A, Rule G, Laufersweiler D, Keeney N, Flint I, Pinevich Y, Herasevich V. TCCC Decision Support With Machine Learning Prediction of Hemorrhage Risk, Shock Probability. Mil Med. 2023 Nov 8;188(Suppl 6):659-665. doi: 10.1093/milmed/usad298.
• Preiksaitis C, Ashenburg N, Bunney G, Chu A, Kabeer R, Riley F, Ribeira R, Rose C. The Role of Large Language Models in Transforming Emergency Medicine: Scoping Review. JMIR Med Inform. 2024 May 10;12:e53787. doi: 10.2196/53787.

Study record dates

Study Major Dates

• Study Start (Actual): February 2, 2025
• Primary Completion (Actual): March 31, 2025
• Study Completion (Actual): April 30, 2025

Study Registration Dates

• First Submitted: January 22, 2025
• First Submitted That Met QC Criteria: January 22, 2025
• First Posted (Actual): January 28, 2025

Study Record Updates

• Last Update Posted (Actual): May 13, 2025
• Last Update Submitted That Met QC Criteria: May 12, 2025
• Last Verified: May 1, 2025

More Information

Terms related to this study

• Keywords: mechanical ventilation; artificial intelligence; decision-making; combat medicine
• Other Study ID Numbers: FieldAI

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: Individual participant data will not be shared due to concerns regarding participant confidentiality, data privacy, and the sensitive nature of the study involving military personnel. Aggregate data and study findings will be shared through peer-reviewed publications and presentations.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No