Data Desert: Military Medicine's Artificial Intelligence Implementation Barriers

Funding Source: None

Disclaimer: The views expressed in this manuscript are those of the author and do not necessarily reflect the position or policy of the Department of Defense, the Department of Veterans Affairs or the United States government. This manuscript has received PAO/OPSEC approval for publication. No personal or financial conflicts of interest to report.

Acknowledgements: None

Abstract

Artificial Intelligence possesses the potential to revolutionize military medicine by aiding in decision-making and by optimizing combat medical operations. However, there are multiple barriers that are currently preventing the application of Artificial Intelligence in a military context. These barriers include lack of an effective data infrastructure and absence of advanced point of injury triage algorithms. Military health systems need to transition from industrial age data practices to ones where large amounts of high-quality data are passively collected and integrated with existing communication networks. Additionally, Artificial Intelligence triage algorithms are needed to assist in accurate point of injury decision-making and to ensure optimal medical resource allocation during future conflicts. Artificial Intelligence's potential to transform military medicine during future conflicts is substantial, but critical infrastructure needs to be developed before military health systems can optimally use this transformative technology.

Introduction

Artificial intelligence is poised to fundamentally change society, and military medicine needs to be postured to take full advantage of this innovative technology. In 2022, United States (US) Army Futures Command (AFC) published its Concept for Medical 2028(1). Central to Concept for Medical 2028 is the use of Artificial Intelligence (AI) to assist with decision making across a wide range of medical capabilities. While Concept for Medical 2028 is US centric document, the application and implementation a military medical AI capability is universally applicable. Barriers to the implementation of military medical AI capability include a lack of a high-quality data infrastructure and the absence of triage algorithms.  

Discussion

Artificial Intelligence relies on evaluating a large amount of quality data. This demands developing a comprehensive data infrastructure, which must automatically enter passively collected data relevant to the operational environment, eliminating manual input. It is critical that these data collection systems are fully nested within existing data and communication networks.  Additionally, point of injury triage will be central to medical resource decisions during future conflicts. To be effective, AI algorithms must be developed to triage patients quickly and accurately during mass casualty events.

Given the high lethality and precision of current munitions, it is expected that future conflicts will generate far more casualties than the medical system can treat. For example, current estimates predict that an US Army Corps (90,000 soldiers) will face approximately 50,000 casualties over eight days of fighting(2). For context, a US Corps has an organic Medical Brigade with about 350 hospital beds. Even when one considers additional Role I-II capacity (Battalion Aid Stations/Medical Company), and potential augmentation with reserve medical units, it is clear the demand for medical capability will vastly exceed capacity. As such, the central problem facing military medicine is scale; the number of casualties requiring treatment will vastly outstrip the capacity of available medical resources. 

This fundamental mismatch will lead to significant strategic risk. If operational units are inundated with casualties they cannot evacuate, there is a higher likelihood they will no longer be able to perform offensive operations. Additionally, the public has become conditioned to believe that every injured service member will receive high quality medical care. The strategic will of the public could be shaken by images of injured service members dying before receiving medical care. Taken together, the potential lack of quality battlefield care could lead to significant strategic risk to both the mission and the force.

To address these challenges, AFC published its Concept for Medical 2028, which lays out a vision of how US Army Medicine can address these challenges. Artificial Intelligence is a fundamental feature.  Army Futures Command envisions AI being used to solve a diverse group of challenges, including mass casualty triage, evacuation platform choices, geographic allocation of medical units, logistic resupply, and provider credentialing. While AI is not a panacea to solve all problems, it is a potentially powerful tool to optimize medical operations at scale.

Artificial Intelligence has been around for decades, but in the last five years AI’s ability to solve complex problems has improved markedly.   Recent advances in computing power, combined with massive increases in data collection, have facilitated rapid advances. One of AI's emerging strengths is its ability to decipher large complex problems that are difficult for humans to solve. For example, Google developed an AI algorithm using 128,000 retinal photographs that could not only diagnose diabetic retinopathy more accurately than ophthalmologists(3), but could also accurately predict the risk of cardiovascular events(4).

Artificial Intelligence is a generic term that encompasses many related, but different, techniques. At its basic level, all AI finds patterns and answers questions. It is designed to take input data, apply an algorithm, and then produce output data often predicting the probability of an event occurring.  At the one end of the AI spectrum are expert systems designed to mimic decisions a human expert would make. At the other end of the spectrum are the most sophisticated, powerful, and complex AI technologies include machine learning, deep learning, and neural networks. Because of this complexity, deep learning models require massive amounts of data and computing power to develop. Manual data entry systems typically cannot keep up. 

For example, many of the most common deep learning models of image classification use the ImageNet database, which contains 10 million labeled images and is 150 Gigabytes in size. Frequently, deep learning models use transfer learning, in which a model uses knowledge gained from training on a different but related task, to decrease the data required to learn a new task. Even with the use of transfer learning, however, deep learning algorithms demand Gigabytes of data. For context, the entire US Joint Trauma System Department of Defense Trauma Registry, including 15 years of patient data is only 0.017 Gigabytes(5). This is orders of magnitude less data than what is needed for the simplest deep learning algorithm developed using transfer learning. 

Currently, most of the military medical data we collect is entered by hand. For example, if a deployed service member were injured today, his/her clinical information would be input by hand into a paper form. Additional clinical information would be manually entered into the electronic medical record. Information to inform the medical common operating picture, such as medical logistics resupply requirement, hospital bed status and available units of blood, also would be manually entered into legacy products such as Excel or PowerPoint.  

This leaves two major gaps between our current state and the one envisioned by Medical Concept 2028. First, we need to transition from our industrial age manual data collection practices to those of a digital age, where data is passively and continuously collected. Second, we need to develop clinical algorithms that assist with point of injury mass casualty triage. Given the limited availability of medical assets in Large Scale Combat Operations (LSCO), it is imperative that we are able to accurately triage wounded service members both to maximize the impact of medical care and to aggressively return service members back to duty at the lowest echelons of care.      

To transition to passive collection of data we need to develop systems that are designed from the bottom up to not require human data entry. For example, instead of manually inputting medical supplies inventories, an AI image recognition algorithm could use video of medical supply storage to automatically update quantities of medical supplies on hand in the system of record.  Additionally, future electronic medical record systems could be designed to passively record heart rate, blood pressure, and oxygen saturation.  

A particularly important aspect of building this data infrastructure is the development of wearable technology. Wearables are small electronic devices that track physiologic parameters such as heart rate, sleep, and movement. The development of a wearable sensor that could continuously track and record data of vital signs and activity measurements could be a powerful facilitator of AI.

Establishing this medical data infrastructure will not be easy because of healthcare data privacy concerns, operational security requirements, and communication constraints. The operational force deals with similar challenges in other domains, and it is critical that military medicine’s data systems are fully nested within the overall data and communication infrastructure of the overarching force. Furthermore, these systems need to be low bandwidth, given that they will compete for bandwidth with other warfighting functions. The current data infrastructure cannot take advantage of AI. It needs to transition into something similar to Amazon, a system that passively collects large amounts of high-quality data.

In addition to developing a quality data infrastructure, the military medicine needs to invest in developing AI algorithms that focus on point of injury mass casualty triage. It is critical to rapidly stratify casualties to identify service members who can safely return to duty in order increase combat strength. Concurrently, we also need to accurately determine who needs to be evacuated and what level of care they require. The scale of this problem is likely to be so large that human decision-making is suboptimal. Artificial Intelligence’s strength is finding patterns in massive amounts of data can speed and sharpen this decision-making process.

We know that clinicians can make accurate triage decisions based on the appearance of a patient and other limited data(6). We also know that AI is particularly effective with image and voice data. Given this knowledge there is a high likelihood that an accurate triage algorithm could be developed using short audio/video recording combined with vital sign data from wearables. Optimally, this could be developed by civilian partners who frequently see a high volume of trauma patients. It could be collected with time dependent outcome data. This algorithmic development could occur concurrently with the development of the medical data infrastructure. One could envision a soldier at the point of injury quickly taking audio/video recording of multiple injured soldiers. An AI algorithm on his/her phone could then rapidly tell him/her who could return to duty, who needs to be evacuated, the priority of those who need to be evacuated, and to what location.

Conclusion

Artificial Intelligence has the potential to drastically improve military medicine’s ability to care for combat causalities during future conflicts. To take full advantage of this technology, military medicine needs to take a comprehensive approach to developing a data infrastructure where high quality, passively collected medical data is fully incorporated into the operational force’s communications and data systems. This would require the development of a cross functional team to oversee design and implementation. Finally, special emphasis should be placed on developing a high-quality mass casualty triage algorithm. Implementation of these elements will greatly increase the military medicine’s ability to take advantage of this game changing technology, and it will allow AFC Concept for Medical 2028 to come to fruition.

References:
(1.) Army Futures Command Concept for Medical 2028. Published online March 4, 2022.
(2.) Fandre M. Medical Changes Needed for Large-Scale Combat Operations. Mil Rev. Published online May 2020:36-45.
(3.) Gulshan V, Peng L, Coram M, et al. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. JAMA. 2016;316(22):2402-2410. doi:10.1001/jama.2016.17216
(4.) Poplin R, Varadarajan AV, Blumer K, et al. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat Biomed Eng. 2018;2(3):158-164. doi:10.1038/s41551-018-0195-0
(5.) Drakos ND. Personal Communication on the size of the Joint Trauma System Department of Defense Trauma Registry.
(6.) Wiswell J. “Sick” or “not-sick”: accuracy of System 1 diagnostic reasoning for the prediction of disposition and acuity in patients presenting to an academic ED. Am J Emerg Med. 2013;31:1448-1452.

Author Biography:
Colonel Benjamin Donham, M.D. is currently the Command Surgeon for the United States Army XVIII Airborne Corps. He is board certified by the American Board of Emergency Medicine, is a Fellow American Academy of Emergency Medicine, and is an Associate Professor of Military/Emergency Medicine at the Uniform Services University.

Date: 11/24/2023