The study was performed in three UK HEMS with five operational bases: two are operated by East Anglian Air Ambulance (EAAA), two by Essex & Herts Air Ambulance (EHAAT), and one by Magpas Air Ambulance (Magpas). HEMS provide prehospital critical care on behalf of the statutory ambulance service in the East of England (The East of England Ambulance Service NHS Trust (EEAST)) to a population of over six million people over a geographic area of 20,000 km2 [19], dispatched by either rotary wing (H145 (EAAA), AW169 (EHAAT/Magpas), or MD902 (EHAAT)) or rapid response vehicle, depending on patient location, weather constraints, and time of day.

The core of each team consists of a physician and a critical care paramedic with at least three years’ post-registration experience. Physicians in these teams are predominantly emergency medicine (EM) or anaesthesia consultants or senior registrars (at least five years post-registration), with a minimum of six months training in hospital anaesthesia. Prior to independent practice, physicians undergo further specialist training in prehospital care, including a period of supervision and local formative assessment prior to independent practice [8].

These services deliver PHEA according to a shared guideline [8]. This includes a standardised drug regime: ketamine (1–2 mg kg−1), rocuronium (1 mg kg−1), ± fentanyl (1–3 mcg kg−1) at the discretion of the attending clinician to attenuate the hypertensive response to laryngoscopy; subjectively tailored to each patient, based on factors such as age and haemodynamics [7]. Intubation is typically performed using direct laryngoscopy. In 2017, the option (for use at the discretion of the attending clinician) of videolaryngoscopy was introduced at EAAA and Magpas (McGrath® videolaryngoscope, Aircraft Medical, Edinburgh, UK). All services use positive pressure ventilation targeting a tidal volume of 7 ml kg−1 (ideal body weight) with an initial PEEP of 5 cm H2O and a frequency set to achieve normocapnia. A pre-induction checklist attempts to identify and initiate correction of physiological derangement prior to administering anaesthesia. All services use HEMSbase (MedicOne Systems Ltd, UK) electronic medical record software.

In this retrospective observational study, a consecutive sample of trauma patients ≥ 16 years old, attended to by EAAA or EHAAT (1st January 2015 to 31st December 2020) or Magpas (1st November 2015 to 31st December 2020, owing to later HEMSbase implementation) and underwent PHEA were included.

Clinical records were reviewed by one of the study authors to identify exclusions: duplicate cases, unascertainable patient age, secondary transfer, intubated in arrest, intubation by a non-HEMS clinician, and mechanisms not meeting the definition of trauma (injury through the transfer of kinetic energy); including medical cases initially coded as ‘trauma’, overdose, hanging, asphyxiation, burns, drowning, electrocution. Records were also excluded if systolic blood pressure readings were not available pre- and post-PHEA.

Anonymised data were extracted from HEMSbase and collated into a password-protected data sheet (Microsoft® Excel for Mac, v16.45) stored on a secure server.

The following data items were retrieved: demographics (age, sex, estimated weight), trauma type (blunt or penetrating), mechanism of injury, Glasgow Coma Scale (GCS) score, injury pattern suspected by attending clinician, indication for PHEA, time interval from arrival of HEMS team to PHEA, and intravenous crystalloid administration by EEAST before HEMS arrival.

Physiological variables were captured from time-calibrated patient monitors (EAAA – X Series, ZOLL Medical Corporation, Runcorn, UK; EHAAT & Magpas – Tempus Pro, Philips Electronics UK Ltd, Farnborough, UK) and uploaded automatically to HEMSbase at two-minute (EAAA, EHAAT) or three-minute (Magpas) intervals. Using manual review of each case by the study authors, the following pre-PHEA physiological variables were recorded based on the closest time-point preceding the recorded time of PHEA: heart rate (HR), respiratory rate (RR), systolic blood pressure (SBP), diastolic blood pressure (DBP), and derived shock index (SI). Post-PHEA SBP readings were recorded at the time points closest to two, four, six, eight, and ten-minutes post-PHEA. Data were excluded if deemed explicitly erroneous. If data were equivocal, a decision to include was reached by consensus after independent review of all available case notes.

PHEA drug doses of fentanyl (mcg kg−1), ketamine (mg kg−1), and rocuronium (mg kg−1), were calculated using the recorded dose and estimated patient weight. These were rounded to the nearest integer and then summarised by the universally cited regimes of drug administration (fentanyl:ketamine:rocuronium) e.g., 3:2:1, 1:1:1 etc. Records of patients who had been administered a vasoactive drug (metaraminol, ephedrine, or adrenaline) were individually reviewed to record the time of vasopressor administration and coded as pre-RSI, post-RSI ≤ 10 min, post-RSI > 10 min, and post-RSI < time unknown > .

Hypotension was defined as a new SBP < 90 mmHg ≤ 10 min of induction, or a > 10% drop if SBP was < 90 mmHg pre-PHEA [20].

Data manipulation and statistical analyses were performed using the R statistical programming language (R Core Team [2018]; R: A language and environment for statistical computing [R Foundation for Statistical Computing, Vienna, Austria]). Statistical significance was pre-defined as p < 0.05. Characteristics of the sample were described as number (percentage) for categorical variables, and mean (± standard deviation (SD)) or median [interquartile range (IQR)] for continuous variables as appropriate.

A purposeful selection logistic regression model was used [21]. Each variable was first tested in turn to explore the unadjusted association with the outcome. Variables with a p-value < 0.25 and variables of known clinical importance were included in the multivariable analysis. Variables were then sequentially eliminated until only statistically significant variables remained and the model achieved the best fit based on likelihood ratio tests. The assumptions of logistic regression were tested, checking for linear relationships in the logit of the outcomes, unduly influential values and multicollinearity. Plausible interactions were tested, with likelihood ratio tests and McFadden’s pseudo R-squared used to determine the final best model.

Administration of vasoactive medication pre-PHEA was not defined a-priori as a determinant of hypotension. Therefore, vasopressor administration was not included in the purposeful model build. In recognition that post-PHEA vasopressor administration could confound the results or attenuate the outcome, the sensitivity of the final model to including vasopressor administration was evaluated.

For categorical variables, the group containing the largest number of cases was used as the reference group in the regression model. Patients were divided into four age groups. Pre-PHEA SBP was grouped as: Low (< 90 mmHg), Mid (90–140 mmHg), High (> 140 mmHg), heart rate was grouped as: Low (< 60 beats/min), Mid (60–100 beats/min), High (> 100 beats/min), and respiratory rate was grouped as: Low (< 10 breaths/min), Mid (10–25 breaths/min), High (> 25 breaths/min) [22]. For drug regimes, the 3:2:1 dose regime was used as the reference group, compared with 1:1:1, 0:1:1, and 0:0:1 (rocuronium only); alternative regimes were coded as ‘other’ [7].

Ethical approval for the study was granted by Anglia Ruskin University Research Ethics Panel (AH-SREP-20-047). The study was registered and approved by each participating organisation. The STROBE (Strengthening the Reporting of Observational studies in Epidemiology) reporting guideline was followed [23].