Industries / Government, education and community / Government / Military
Medical care
Caring for sick and wounded service members, from the point of injury through evacuation to hospital, and keeping the records that show what worked.
7
Milestones achieved, from 1862 to 2024.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1862
Ambulances under doctors
An order of 2 August 1862 for the Union's Army of the Potomac, drawn up by its medical director, Jonathan Letterman, took the ambulances away from the quartermasters and put them under a medical corps. The National Museum of Civil War Medicine says the system removed all the wounded from the battlefield at Antietam within 24 hours, and that in March 1864 Congress applied Letterman's changes to the whole US Army.
LessonA system proved in one army took two years to become the rule for all of them.
Credit
Jonathan Letterman
1917
Blood stored in advance
In 1917 Oswald H. Robertson, a US Army doctor in France, collected type O blood ahead of time and stored it, instead of drawing it from a donor when a patient needed it; 22 units resuscitated 20 wounded soldiers, and blood was kept for up to 26 days. The Stanford Blood Center, citing a 2017 paper in the journal Transfusion, calls this "the first blood bank"; the Canadian Museum of Health Care says he "is credited by some".
Credit
Oswald H. Robertson
1953
Surgery near the front
In the Korean War (1950 to 1953), mobile army surgical hospitals worked close to the front, and helicopters carried some of the worst wounded to them. A 2023 historical review gives the case fatality rate for US troops, the share of wounded who died, as 2.5% in Korea, down from 4.5%, but calls the helicopters' share "small and difficult to determine precisely"; a 2012 paper puts it at about 4% of hospital admissions. The US Air Force's 801st Medical Evacuation Squadron, flying fixed-wing aircraft, moved more than 311,000 sick and wounded out of Korea.
LessonThe helicopter became the symbol, but most wounded went by road, rail and plane, and deaths fell for several reasons at once.
Credit
US Army Medical Service; US Air Force aeromedical evacuation
1994
Plasma as a powder
Plasma, the liquid part of blood, can be freeze-dried to a powder for storage and use in the field. A 2010 paper by French military blood specialists says "the French army has used freeze-dried and secured plasma (FDSP) since 1994." The Israel Defense Forces' Medical Corps made plasma its first choice for fluid resuscitation in 2013 and gave freeze-dried plasma to 109 casualties from January 2013 to June 2016.
pmc.ncbi.nlm.nih.govcris.tau.ac.ilmdanderson.elsevierpure.com
Credit
Centre de transfusion sanguine des armées (French military blood transfusion center); Israel Defense Forces Medical Corps
2009
Tourniquets save lives
A tourniquet is a band tightened above a wound to stop bleeding from a limb. Kragh and colleagues studied casualties at a combat support hospital in Baghdad in 2006: 232 had 428 tourniquets applied on 309 injured limbs, survival was 87% among those who needed one and got one against 0% for the five who needed one and did not, mortality was 11% when tourniquets went on before hospital and 24% when applied in the emergency department, and "No limbs were lost due to tourniquet use". The study appeared in the Annals of Surgery in January 2009.
LessonBleeding is best stopped where the injury happens, before the hospital. /kragh-survival-with-emerg-tourniquets-ann-surg-2009.pdf).
Credit
John F. Kragh Jr. and colleagues
2015
The golden hour, measured
In 2009 US Defense Secretary Robert Gates required that wounded troops in Afghanistan reach a treatment facility within 60 minutes of the call for help. Russ Kotwal and colleagues of the US Army Institute of Surgical Research studied 21,089 US casualties from September 2001 to March 2014, using the DoD Trauma Registry, a database of injured patients' care that the services have kept since 2004: median transport time fell from 90 to 43 minutes, the share killed in action from 16.0% to 9.9%, and the case fatality rate from 13.7% to 7.6%, which they equate to 359 lives saved. Their paper appeared in JAMA Surgery in September 2015.
LessonEarlier work by Eastridge and colleagues found 87.3% of battlefield deaths happened before a treatment facility, so time to care was where lives could be saved.
Credit
Russ S. Kotwal and colleagues
2024
A health record, finished late
MHS GENESIS, the US military's new electronic health record, was first tested at four sites in Washington State, and a memo from the Pentagon's operational test office in April 2018, reported by Federal Times, found it neither operationally effective nor suitable "because of poor system usability, insufficient training". The last site went live in March 2024, at the Federal Health Care Center, but GAO found in April 2024 that users' satisfaction was "generally lower" than with the old systems; on one survey measure, patient-centered care, the rate was 39%, against 56% for users of the legacy systems.
LessonFielding a system and having it work for its users are separate finish lines.
Credit
GAO; the Pentagon's Director, Operational Test and Evaluation
Watch for
Donors on the spot
A walking blood bank collects fresh whole blood from screened service members on site when stored blood runs short. The US Army says an exercise in Pyeongtaek, South Korea, in August 2026 "marked the first time the Joint Trauma System validated a Role 3 field hospital", the 121st Field Hospital, for its walking blood bank, and that staff documented care offline with records sent automatically once the connection returned. The article gives no donor numbers or timings.
Achieved whena deployed hospital uses it and someone outside the Army publishes the time to the first unit, units collected and how many records reached the trauma registry.
Claimed by
US Army, November 2026
Casualty cards read by AI
The DD Form 1380 casualty card is the paper record medics fill in at the point of injury. A poster from the Uniformed Services University (USUHS) and IBM says their system can "accurately extract handwritten data from all 172 distinct form fields", and shows a full match on 60 de-identified cards from the Joint Trauma System's archive; an open-source project on GitHub, created in June 2026, turns card photos into structured records with several readers.
Achieved whenfield-by-field error rates on held-out cards appear in a peer-reviewed paper, or the Joint Trauma System uses the tool to fill its registry.
Claimed by
USUHS and IBM (undated poster, written after September 2025); the DD-1380 GitHub project, June 2026
Open
01
Made-up casualty cards, with answers
A public set of at least 300 made-up DD Form 1380 casualty cards, filled in by hand by volunteers from written scenarios, front and back, each with an answer key for every field, and a scoring script, so that any card reader, including the USUHS and IBM system, can be compared on the same cards.
The briefCredit · unclaimed02
Spoken handoffs in noise
A public English set of scripted medic handoff reports in the four-part MIST format (mechanism of injury, injuries, signs, treatment), read by volunteers and mixed with vehicle and aircraft noise at several levels, each with a transcript and the four parts as its answer key. Score speech-to-text systems on word errors and on the parts they recover.
The briefCredit · unclaimed
Open 01 · the brief
Made-up casualty cards, with answers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of at least 300 made-up DD Form 1380 casualty cards, filled in by hand by volunteers from written scenarios, front and back, each with an answer key for every field, and a scoring script, so that any card reader, including the USUHS and IBM system, can be compared on the same cards.
Why it is still open
The US Army's Telemedicine and Advanced Technology Research Center writes that the military health system "lacks real-world datasets for TCCC care at the point of injury" (TCCC is Tactical Combat Casualty Care) and that its own data "contains PII", personal information that cannot be shared without agreements. A 2011 study cited in 2014 found prehospital data in only 18.6% of records from Afghanistan and 25.4% from Iraq. The USUHS reader was checked on 60 real cards that cannot be published, and the GitHub project ships no dataset.
Built with what exists
The blank DD Form 1380; a script that writes scenarios and keys; docTR and TrOCR, open-source text readers, as baselines; AI models that read images, such as Claude and Gemini; a small vision model fine-tuned on one GPU to run without a network (see the Note "One GPU to fine-tune").
First step
Write 20 scenarios, have five volunteers fill printed cards, key them, and run docTR on them.
Done looks like
300 cards by at least 20 writers, published under an open licence with keys for all 172 fields; at least three readers scored on exact match for each field and on checkbox accuracy.
Open 02 · the brief
Spoken handoffs in noise
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public English set of scripted medic handoff reports in the four-part MIST format (mechanism of injury, injuries, signs, treatment), read by volunteers and mixed with vehicle and aircraft noise at several levels, each with a transcript and the four parts as its answer key. Score speech-to-text systems on word errors and on the parts they recover.
Why it is still open
A 2021 study of voice documentation for combat casualty care (Woo and colleagues) recorded 27 cases, 5.05 hours of audio, and reached a medical word error rate of 33.3%, but published no data. A public German set of 99 synthetic emergency dialogues with noise appeared in November 2025, and its authors wrote that "no public dataset captures the full range of noise constellations relevant to EMS practice". A 2023 study found the spoken handoff "much shorter and less informative" than the card.
Built with what exists
Whisper and Vosk, open-source speech recognizers; commercial systems such as ElevenLabs Scribe v3 (see the Note "Fewer word errors"); public noise recordings and the noise-mixing method of the German study; an AI model to fill the four parts from each transcript.
First step
Write 30 MIST scripts from public training examples and record three volunteers in a quiet room.
Done looks like
100 scripts read by at least 10 volunteers, each mixed at five noise levels; at least four systems scored on word error rate, medical word error rate and the accuracy of each MIST part.