Transforming Combat Casualty Care Training: Generative AI-Enabled Adaptive Learning in Forward Medical Settings

Oct 15, 2026

Rapid training for far-forward combat casualty care demands speed and reliability in austere, resource-limited environments. This white paper addresses how variability in baseline medical knowledge can be mitigated through adaptive learning without the heavy cost of manually mapping question-answer (QA) sets to curricular objectives. It presents an AI-assisted, human-in-the-loop workflow that generates QA sets aligned with NBME standards, validates quality using LLM judges, and compares performance to human rater outcomes.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages20

Reading Time50 min

AccessFree to download

Executive Summary

Background

Method

Results

Interrater Reliability

Discussion

Study Limitations and Methodological Implications

Conclusion

References

Appendix A: Time and Cost Advantages of LLMs in Medical QA Set Creation

Appendix B: Author and Judge Prompt Key Components

TRANSFORMING COMBAT CASUALTY CARE TRAINING: GENERATIVE AIENABLED ADAPTIVE LEARNING IN FORWARD MEDICAL SETTINGS Alan D . Koenig, John J . Lee, Eric Savitsky, Gabriele Nataneli, Karson Lindstrom, David L. Schriger, and Tyler Savitsky Copyright © 2025 The Regents of the University of California. To cite from this report, please use the following as your APA 7th edition reference: Koenig, A. D., Lee, J. J., Savitsky, E., Nataneli, G., Lindstrom, K., Schriger, D. L., and Savitsky, T. (2025) . Transforming combat casualty care training: Generative AI-enabled adaptive learning in forward medical settings. [White paper] . UCLA/CRESST. Table of Contents Executive Summary ............................................................................................................... iv Background............................................................................................................................. 1 Method .......................................................................................................................................4 Results ........................................................................................................................................4 Interrater Reliability ................................................................................................................... 5 Discussion ............................................................................................................................... 7 Study Limitations and Methodological Implications ..................................................................8 Conclusion ............................................................................................................................ 10 References ............................................................................................................................ 11 Appendix A: Time and Cost Advantages of LLMs in Medical QA Set Creation......................... 13 Appendix B: Author and Judge Prompt Key Components ...................................................... 14 Transforming Combat Casualty Care Training: Generative AI-Enabled Adaptive Learning in Forward Medical Settings Alan D. Koenig,a John J. Lee,a Eric Savitsky,b,c Gabriele Nataneli,c Karson Lindstrom,d David L. Schriger,e and Tyler Savitskyc a CRESST/University of California, Los Angeles b Emergency Medicine/University of California, Los Angeles c SonoSim, Inc. d Caylent e Emergency Medicine/University of California, Los Angeles Executive Summary Introduction Rapid training of military and civilian first responders for far-forward combat casualty care in austere, resource-limited environments is a NATO priority during Large-Scale Combat Operations (LSCOs) . Variability in baseline medical knowledge complicates this training. Although adaptive learning can bridge these gaps, creating and mapping traditional questionanswer (QA) sets to curricular objectives is a resource-intensive process. Recent advancementsin generative artificial intelligence (AI) have the potential to overcome these barriers. Here, we describe how we leveraged AI-driven techniques to generate QA sets consistent with National Board of Medical Examiners (NBME) standards, validated their quality using large language models (LLMs) as a judge, and compared their performance against that of a human rater. This study presents a process that can be integrated into online medical training platforms, utilizing a methodology that is generalizable to other educational domains. It features a human-in-theloop framework, where expert supervision is applied to AI-assisted educational assessment. Method A medical subject matter expert (SME) rated four LLMs’ (OpenAI’s GPT-4, GPT-4o, o1-mini, and o1) performance generating 320 medical QA sets using a three-tier scale: “High” for complete task coverage,“Medium” for minor omissions, and “Low” for critical omissions. Subse

FAQs

What time and cost reductions were reported compared with human SMEs?

A human SME took about 24 minutes to author one high-quality QA set at an estimated cost of $100. In contrast, each LLM generated a QA set in roughly 20 seconds at a fraction of the price, with large reductions reported in the paper’s results and discussed in the appendix.

Oct 15, 2026

Uploaded by Taylor Morgan