Army Demographics and Injury Dataset

This data was used to create a synthetic population in order to analyze the effectiveness of the START, SALT, and BCD Sieve triaging protocols. Along with this analysis, we also performed a study using the ALIGN system to evaluate the effectiveness for LLMs to triage injuries using these protocols.


Evaluation of Triage Tagging Protocols Using a Synthetic Dataset Representative of Battlefield Injury Profiles	Aligning a Large Language Model Protocol-Based Medical Triage Decision Making

The Casualty Evaluation Dataset was created to meet the following army distribution requirements.

	Male	Female
Percent [6]	84.6%	15.4%
Height (cm) [5]	mean: 177.0 stdev: 7.1	mean: 163.5 stdev: 7.7
BMI [5]	mean: 26.4 stdev: 3.4	mean: 24.7 stdev: 2.8

Normalized Age Distributions [6]

Injury severities use the AIS (Abbreviated Injury Scale) severity score. AIS values are:

1.0: Minor
2.0: Moderate
3.0: Serious
4.0: Severe
5.0: Critical
6.0: Fatal

Location	Type	Proportion	AIS
Head / Neck		36.2% [2]
	Airway Obstruction	18% [3]	mean: 4.00 stdev: 0.70 **
	Superficial	60%	100% are 1.0 **
	Traumatic Brain Injury	22% [1]	mean: 3.50 stdev: 0.65 [3]
Thorax		8.6% [4]
	Fracture	51.2% [4]	mean: 2.85 stdev: 0.85 [4]
	Hemorrhage	34.6% [4]	mean: 2.85 stdev: 0.85 [4]
	Hemothorax	30% [4]	mean: 2.85 stdev: 0.85 [4]
	Pneumothorax	51.8% [4]	mean: 2.85 stdev: 0.85 [4]
	Pulmonary Contusion	50.2% [4]	mean: 2.85 stdev: 0.85 [4]
	Spinal	14.6% [4]	mean: 2.85 stdev: 0.25 [4]
Abdomen		6.9% [2]
	Hemorrhage	34.6%*	mean: 2.85 stdev: 0.85 *
	Laceration / Contusion	65.4%	mean: 2.85 stdev: 0.85 *
Extremity		49.4% [2]
	Burn / Nerve	5% [7]	56% are 2.0 23% are 3.0 17% are 4.0 7% are 5.0 [7]
	Contusion / Sprain / Strain	20% [7]	56% are 2.0 23% are 3.0 17% are 4.0 7% are 5.0 [7]
	Fracture / Dislocation	22% [7]	56% are 2.0 23% are 3.0 17% are 4.0 7% are 5.0 [7]
	Hemorrhage	52% [7]	56% are 2.0 23% are 3.0 17% are 4.0 7% are 5.0 [7]

*Assumed to match statistics for Thorax Hemorrhage
** Derived from [3]

Disclaimer

This research was performed in connection with the U.S. Army Contracting Command - Aberdeen Proving Ground (ACC-APG) and the Defense Advanced Research Projects Agency (DARPA) under contract number W912CG- 24-C-0011. The views and conclusions are those of the authors and should not be interpreted as presenting the official policies or position, either expressed or implied, of ACC-APG, DARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation hereon.