The most data-rich medical diagnostic ever built is designed for the environment that has the least data of all.
Validated on 5,895 trauma patients across 8 medical centers, the system identifies casualties at 6.88× the baseline risk of hemorrhagic injury.
The U.S. Army is now licensing the invention for commercial field-medicine products. A direct pipeline from DOD lab to battlefield deployment.
APPRAISE-HRI validation cohort
543 hemorrhagic injury cases and 5,352 controls across emergency departments and prehospital transport. The study was published in NEJM AI. · NEJM AI, 2026
The Automated Processing of the Physiological Registry for Assessment of Injury Severity — Hemorrhage Risk Index (APPRAISE-HRI) is the first triage system ever cleared by the FDA for assessing hemorrhage risk in trauma patients. It is classified as a Class II medical device under 510(k) clearance K233249.
The software runs on an Android device. It receives heart rate and blood pressure data via Bluetooth from a standard vital-sign monitor. Within 10 minutes of first contact, it outputs one of three risk levels: I (low), II (average), or III (high).
Over 90% of combat casualties die at or near the point of injury before evacuation. The primary cause is uncontrolled bleeding. Medics in prolonged field care scenarios have no decision-support tools for hemorrhage triage. They rely on vital-sign thresholds that were never designed for combat physiology.
It fills that gap. The algorithm took two decades to build.
Dr. Jaques Reifman, director of the DOD's Biotechnology High Performance Computing Software Applications Institute, led the development. His team collected data from three clinical studies spanning roughly 2,000 patients, including recordings from moving ambulances, helicopter transports, and the Massachusetts General Hospital emergency department.
"One of our biggest challenges was ensuring all that data were reliable and consistent," Reifman told the Army's Medical Research and Development Command.
The independent validation added 5,895 patients across eight sites. Hemorrhagic patients were 6.88 times as likely as controls to be classified at level III. That is a strong signal that the algorithm correctly identifies those needing immediate evacuation. Level I patients were 0.18 times as likely to be hemorrhagic, suggesting the absence of bleeding with high confidence. The FDA classified it as Class II, the first regulatory green light for an AI triage tool of its kind.
How APPRAISE-HRI works under the hood
The system was designed for low echelons of care (Roles 1 and 2), the first medical contact a casualty receives and the forward surgical team. No lab results, no imaging, no history. Just vital signs and a risk score.
It is not an isolated case. The DOD's broader JOMIS system now includes BATDOK-J, a mobile app for point-of-injury care with wireless sensor integration, and the Operational Medicine Care Delivery Platform, a fit-for-purpose EHR built for disconnected environments. These systems share one constraint: they must work without network connectivity, without lab support, without a blood bank on site.
The same AI techniques that power it, ensemble classifiers trained on sparse field data, are being validated across other military medical problems. The Army Institute of Surgical Research is testing Compensatory Reserve Measurement, an AI model that flags vital-sign degradation earlier than conventional thresholds. DARPA's FSHARP program is building shelf-stable blood substitutes with a $46.4 million budget.
Battlefield medicine is becoming a proving ground for AI diagnostics that function under information scarcity. Those lessons will transfer to civilian mass-casualty events, rural emergency care, and any environment where clinical decision support must work without infrastructure.
Licensing deals: MRDC's Technology Transfer Office is actively seeking commercial partners. Two companies have already expressed interest.
Integration into battlefield monitoring systems: BATDOK-J and other JOMIS platforms could embed its algorithm directly.
Civilian trauma adoption: The same triage problem exists in civilian emergency medicine, especially in rural and mass-casualty settings.
What happens to battlefield triage in the next five years?
Probability: 65%. The DOD has already cleared the regulatory path, validated the clinical utility, and begun licensing. The remaining barrier is procurement integration with existing field monitoring systems.
✅ Arguments for
Existing battlefield IT infrastructure (BATDOK-J, OpMed CDP) creates a natural integration path. The algorithm requires only HR and BP, which are already collected.
Confirmation criteria: First procurement contract for APPRAISE-HRI integration within 12 months.
❌ Arguments against
The validation was retrospective. Prospective battlefield data may show different performance under actual combat stress.
Disconfirmation criteria: No procurement contract by mid-2027.
Development scenarios
🟢 Optimistic scenario (25%)
Implications: Combat casualty survival rates improve measurably. The DOD's two-decade investment becomes the standard of care.
🟡 Base-case scenario (50%)
Implications: The technology works but reaches the field slowly. This is a common pattern for DOD-to-commercial transitions.
🔴 Pessimistic scenario (25%)
Implications: Medics continue relying on manual vital-sign thresholds. The two-decade development effort produces a peer-reviewed paper but no operational impact.
