Field Brief: Prehospital Intubation Was Never the Problem — Patient Selection Was
By John Gomez | Field Brief | ~11 min read
For most of the last twelve years, the received wisdom about prehospital intubation has sat on top of a small stack of high-profile trials, all of which pointed the same direction: don't. PART said supraglottic airways beat endotracheal tubes in out-of-hospital cardiac arrest. AIRWAYS-2 said the same in a different way in the UK. The HAIRPIN sub-work said paramedic RSI in trauma didn't move the needle. A generation of US medics was trained under the shadow of those papers. Somewhere between the classroom and the truck, 'the tube wasn't better than an SGA in a cluster-randomized OHCA trial' mutated into 'prehospital tubes kill people,' which then mutated into a scope-of-practice quiet retreat. Some services pulled RSI. Others let it wither by attrition. In too many rooms, the answer to 'should we tube this trauma patient?' became a shrug and a King LT.
The problem was never intubation. The problem was that we asked the wrong question of the wrong data.
Nelson et al., published February 12, 2026 in The Lancet Respiratory Medicine (DOI: 10.1016/S2213-2600(25)00370-4), took 6,467 patients admitted to a UK major trauma centre, threw the machinery of doubly-robust causal modelling and machine learning at them, and produced a number that ought to change every EMS airway conversation for the next five years: an absolute 10.3% reduction in 30-day mortality (95% CI 8.7% to 11.9%) when prehospital emergency anesthesia with intubation is delivered to the risk-stratified patients who actually benefit.
That is not a subgroup analysis. That is the conditional average treatment effect for the correctly selected patient.
The paper does not say tube everyone. It says the population-level 'no benefit' result you have been living under for a decade was a risk-mixed illusion — the sick patients who needed a tube got one and did poorly, the well patients who didn't need one got one and did poorly, and the mean scrubbed the signal off.
Condition on risk, and the signal comes back.
Where the Intubation Dogma Came From
It is worth naming the four pillars of the 'don't tube in the field' consensus, because Nelson doesn't dismiss any of them — it reframes them.
PART (Wang et al., JAMA 2018) randomized 3,004 US OHCA patients cluster-crossover to primary laryngeal tube (LT) vs. primary endotracheal intubation (ETI). LT beat ETI on 72-hour survival (18.3% vs 15.4%). PART was not a trauma trial and it was not a study of whether to tube, it was a study of what to try first when you can't ventilate someone in cardiac arrest. The right takeaway was 'when time-to-effective-ventilation matters more than the airway itself, an SGA is a legitimate first move.' The wrong takeaway — the one that ended up on training slides — was 'SGAs beat tubes.' Those are different statements.
AIRWAYS-2 (Benger et al., JAMA 2018) ran the same question in the UK across 9,296 OHCA patients and got, essentially, a wash on functional survival at 30 days (favorable outcome 6.4% i-gel vs 6.8% ETI). Same domain as PART. Same misreading in transfer.
HAIRPIN and the older US paramedic RSI work described high complication rates and no mortality benefit when field RSI was rolled out broadly to paramedic services with variable training and low reps. That was a training-and-throughput problem masquerading as an airway-device problem. Every honest ED physician who read those papers said the same thing at the time: this is not about tubes, this is about who is holding the tube.
CRASH-2 airway subanalyses and the trauma literature got dragged into the general 'prehospital advanced airway = bad' narrative even though CRASH-2's actual finding was about TXA. Bystanders in an argument they were never in.
None of those trials were wrong. They just answered narrow questions and got generalized past their evidence.
What Nelson Actually Did
Nelson's team drew from London's Air Ambulance and a linked UK major trauma centre dataset spanning February 23, 2012 to November 13, 2019. They split it into a training set (n = 3,882) and a temporally-held-out test set (n = 2,585). Every patient's prehospital record — vitals, GCS, mechanism, treatments, timings — was cleaned and structured. Then they built two machine-learning models.
Model 1 predicted whether a patient received early intubation. Its test-set AUROC was 0.943. Read: this model figures out with near-perfect discrimination who the on-scene clinicians decided to tube.
Model 2 predicted 30-day mortality. Test-set AUROC 0.867. Solid discrimination between high- and low-risk patients.
Because Model 1 predicted the treatment (intubation) and Model 2 predicted the outcome (death), you can plug them into a doubly-robust causal estimator. What that lets you do — and what a naive Kaplan-Meier or unadjusted logistic model doesn't — is estimate the counterfactual: if this specific high-risk patient had been intubated, what was the probability of death? If this specific patient had not been intubated, what was the probability of death? Subtract.
The output is the conditional average treatment effect: –0.103, 95% CI –0.119 to –0.087. Ten point three percent absolute reduction in 30-day mortality for correctly selected patients. Applied to the annual UK major trauma volume, the paper projects 170 lives saved per year if this risk-stratified approach were adopted nationally. The authors also note the finding meets the NICE cost-effectiveness threshold for HEMS as a national service on its own — an unusual and quietly important claim.
The commentary that ran alongside the paper ('From data to decision,' Lancet Respir Med 2026) called it the strongest causal evidence yet on prehospital intubation efficacy in trauma. That is not marketing. That is a peer-review commentary understating the shift.
The Eight Variables That Actually Matter
The most useful thing Nelson produced is a reduced-feature model the paper calls Intub-8, built with only eight routinely collected prehospital variables and still hitting AUROC 0.897 — enough for real-world dispatch or scene use.
The eight:
Glasgow Coma Scale (with pupil reactivity)
Pulse rate
Respiratory rate
Intravenous fluid administered
Age
Requirement for spinal immobilization
Tranexamic acid administration (proxy for suspected significant bleeding)
Incident category — specifically, low-height fall (<2m)
None of that is a surprise if you've worked a trauma call in the last twenty years. What is a surprise is that eight variables — every one available at the scene, none of them requiring an ultrasound, an ISS, or a pH — carry enough signal to distinguish the patient who benefits from prehospital anaesthesia from the patient who doesn't.
For the mortality prediction, the top drivers were age, total GCS, absence of any airway support at scene, 'normal' breathing status, and incident at home. Read that list twice. It is the same list a good field medic would have written on a napkin: how old is the patient, how obtunded are they, is anyone doing anything for their airway, does their breathing look wrong, and did they collapse at home instead of getting mangled on I-95. Nelson's contribution is not the list. It is the causal quantification that the eight-variable list, when it flips positive, means the patient benefits from a definitive airway before the doors of the trauma bay close.
How to Actually Pick the Patient
Walk five prehospital trauma cases through the eight variables. Pick the intervention you would deliver. The Decision Engine below is not a scoring calculator. It is not a substitute for judgement. It is a way to pressure-test the reflex you brought with you.
For working through it in the truck, three practical anchors:
GCS ≤ 8 in the presence of any of the top-priority mortality drivers (age >70, no airway support at scene, abnormal breathing pattern, obvious major mechanism) is where the causal benefit concentrates. That patient does not become less sick during transport — they become harder to intubate as swelling, aspiration, and hypotension march forward.
TXA activation is now a formal risk-marker. If your service protocol has triggered TXA — either by shock index or by physician review — you are already asserting significant hemorrhage. Nelson's model treats that assertion as one of the eight variables.
Low-height fall <2m in the elderly is now formally on the airway radar. Not because the mechanism is dramatic, but because the population is old, often anti-coagulated, often on the edge of decompensation before EMS arrives. The model spotted what the field medic already suspected.
For services still running King LTs or I-gels as the reflex first move on unconscious trauma patients: nothing in Nelson's paper says stop using SGAs. What it does say is that if the Intub-8 profile screams tube, and your system has the capacity to deliver a good tube — physician, retrieval consultant, HEMS DSI paramedic, or a critical care paramedic with genuine reps — then treating that patient with an SGA-then-transport strategy is now demonstrably tied to a mortality cost.
Why This Doesn't Mean Paramedic RSI Free-For-All
Here is where we need to draw the line, because the paper does not.
Nelson's dataset came from a physician-staffed HEMS system with retrieval consultants and doctor-paramedic pairings on scene, running RSI drug packs, video laryngoscopy, and post-intubation packages the average ground ALS truck does not run. The causal effect Nelson estimated is the effect of intubation as delivered by that team. It is not a general 'tubes save lives' result. It is a 'targeted intubation in a mature retrieval system saves lives' result.
That distinction matters because the ground EMS landscape is bimodal. Some services run genuine critical care paramedic teams with real airway reps, capnography discipline, and DSI protocols governed by an active medical director who audits every attempt. Other services stopped credentialing paramedic intubation altogether after the mid-2010s malpractice cycle. The population-level 'prehospital tubes kill' narrative served the second group's protocols. Nelson's paper does not automatically undo those protocol decisions.
What it does do is force the honest question in every EMS system: what is our system's Intub-8 profile response? If we identify the high-risk trauma patient — and the eight variables are all field-visible, no lab required — do we have a defensible pathway to a definitive airway before the trauma bay? A HEMS launch? A critical care intercept? A ground unit with the training and the physician oversight to deliver targeted DSI? Or is our answer a SGA and a fast transport?
The answer 'SGA and fast transport' was defensible when the population-level literature said intubation was a wash. That answer is now weaker. Weaker doesn't mean wrong — some geographies genuinely lack the resource depth — but it is now an answer that has to be defended, not assumed.
What Every Medical Director Should Do Monday Morning
Three things.
First, audit your current trauma airway pathway against the Intub-8 profile. Pull the last hundred trauma activations. Score them retrospectively on the eight variables. Ask what percentage of your positive-Intub-8 patients received a definitive airway before hospital, by whom, with what timing. If the number is under 40%, you now have a gap that Nelson's evidence base makes uncomfortable to leave open.
Second, name your system's targeted-airway resource. Is it a HEMS unit with physician-paramedic staffing? A critical care intercept truck? A paramedic DSI/RSI protocol with a specific, credentialed subset of medics? If the answer is 'we don't have one,' Nelson's paper is not evidence that you should immediately create one — it is evidence that you should stop pretending the SGA-and-run strategy is neutral for high-Intub-8 trauma patients. It has a cost, and now that cost has a number.
Third, review your paramedic intubation training reps. Nelson's benefit rides on a physician-staffed retrieval system's first-pass success and post-intubation stability. If your paramedic intubation first-pass success is under 90%, or if you don't measure it, your version of the intervention is not the version that produced Nelson's numbers.
There will be commentary and follow-on work — probably a US multicenter re-analysis, probably a set of protocol proposals from NAEMSP or ACEP — that fills in the transatlantic gap. Some of it will be better than Nelson. Some of it will be worse. But the era of citing PART or AIRWAYS-2 as evidence that prehospital tubes don't help is over. Those trials answered a different question. Nelson answered the one we should have been asking all along.
The tube was never the problem. The reflex was.
Sources
Nelson AP, et al. Survival effect of prehospital emergency anaesthesia with intubation in risk-stratified patients with major trauma: a causal modelling study. Lancet Respir Med. 2026. DOI: 10.1016/S2213-2600(25)00370-4. PubMed 41690329.
Editorial: From data to decision: the future of prehospital intubation in major trauma. Lancet Respir Med 2026. PIIS2213-2600(26)00002-0.
UCL press release, 12 Feb 2026: Breathing tube insertion before hospital admission for major trauma saves lives.
Healio Pulmonology, 9 Mar 2026: Mortality reduced with intubation before hospitalization in patients with major trauma.
Physicians Weekly: Prehospital Intubation Cuts 30-Day Mortality by 10% in High-Risk Trauma.
Wang HE, et al. PART trial. JAMA. 2018;320(8):769–778.
Benger JR, et al. AIRWAYS-2 trial. JAMA. 2018;320(8):779–791.
Intensive Care Medicine narrative review, 2026: Prehospital airway and ventilatory management.
FEMA/EMS.gov: Evidence-Based Guideline for Prehospital Airway Management (2023).
Comments