Trauma triage often unfolds in fast-moving, high-pressure environments where clinicians must make life-altering decisions with limited time and incomplete information. This project advances computerized trauma triage decision support designed to improve the accuracy, fairness, and trustworthiness of these critical decisions. Building on preliminary evidence that richer data and more sophisticated models can identify severely injured patients far more accurately than current guideline-based approaches, this work seeks to develop and validate complex machine-learning models using national trauma data, evaluate and mitigate potential algorithmic biases, and creates EMS-oriented explanations that make model recommendations understandable and trustworthy. By integrating technical innovation with clinician-centered design, the project aims to lay the foundation for intelligent, sensor-aware triage support systems that will enhance patient outcomes in trauma care and inform future tools for other time-critical emergencies.

Funding: Research reported on this page was supported by the National Library of Medicine of the National Institutes of Health under Award Number R15LM013824. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Project Aims

Aim 1: Assess the accuracy of complex trauma triage models using the National Trauma Data Bank. Our preliminary experiments measured the accuracy of a single model using data from a single state with limited exploration of algorithmic optimization. Thus, this specific aim seeks to validate and extend those experiments by more rigorously measuring the accuracy of complex models built using multiple machine learning algorithms with a larger and more diverse data set.

Aim 2: Assess the fairness of the complex trauma triage models and compare the performance of models designed to mitigate bias. Given the many documented problems with bias in models learned from data, it is important to ensure that we identify and address issues of fairness in our learned models. Thus, this specific aim will assess the fairness of models explored in Aim 1 and assess the ability of multiple bias-mitigation approaches to maintain accuracy and improve fairness.

Aim 3: Design, generate, and assess EMS-oriented explanations from selected complex trauma triage models. Engendering trust in more complex models through meaningful explanations and through model consistency is important to mitigate the skepticism associated with hard-to-understand, complex models. Thus, using selected complex models, this specific aim will engage EMS personnel to help design and assess explanations.

Associated Publications

2026

• Talbert DA, Talbert S, Atkins N, Phillips KL, Patterson N, and Kamal M. A Human-Centered Approach to Identifying the Challenges of Automatic Generation of Clinically Comprehensible Trauma Triage Explanations. International FLAIRS Conference, 2026.

• Kamal M, Talbert DA, Patterson N, Atkins N, and Hough C. Accuracy Is Not Enough: Rethinking Model Selection for Clinical Machine Learning. International FLAIRS Conference, 2026.

2025

• Brown KE, Talbert S, Talbert DA. Derivation and Experimental Performance of Standard and Novel Uncertainty Calibration Techniques. In AMIA Annual Symposium Proceedings 2025.

2024

• Talbert DA, Phillips KL, Brown KE, Talbert S. Assessing and addressing model trustworthiness trade-offs in trauma triage. International Journal on Artificial Intelligence Tools, 2024

• Brown KE, Talbert S, Talbert DA. A QUEST for Model Assessment: Identifying Difficult Subgroups via Epistemic Uncertainty Quantification. In AMIA Annual Symposium Proceedings 2024.

2023

• Talbert DA, Phillips KL, Brown KE, Talbert S. Group bias and the complexity/accuracy tradeoff in machine learning-based trauma triage models. International FLAIRS Conference, 2023.

2021

• Talbert DA, Talbert S. Trauma Triage in an Information Rich Environment. In: Proceedings of the American Medical Informatics Association Annual Symposium. 2021.

Research Team

PIs: Dr. Doug Talbert (dtalbert@tntech.edu) and Dr. Steve Talbert (steven.talbert@wvumedicine.org) 

Year 1 student research team:

• Matthew Beech (undergraduate researcher)

• Ethan Owens (undergraduate researcher)

• Katherine Phillips (undergraduate researcher)

• Katherine Brown (PhD mentor)

Year 2 student  research team:

• Omar Abdelsalam (undergraduate researcher)

• Vaughn Gavigan (undergraduate researcher)

• Sam Schmahl (undergraduate researcher)

• Moumita Kamal (PhD mentor)

Year 3 student research team:

• Nick Atkins (undergraduate researcher)

• Celia Hough (undergraduate researcher)

• Nol Patterson (undergraduate researcher)

• Tate Wieber (undergraduate researcher)

• Moumta Kamal (PhD mentor)

Year 4 student research team:

• Nick Atkins (undergraduate researcher)

• Celia Hough (undergraduate researcher)

• Brence Moore (undergraduate researcher)

• Katherine Phillips (PhD mentor)