Risk factors and mortality associated with undertriage after major trauma in a physician-led prehospital system: a retrospective multicentre cohort study

To assess the incidence of undertriage in major trauma, its determinant, and association with mortality. A multicentre retrospective cohort study was conducted using data from a French regional trauma registry (2011–2017). All major trauma (Injury Severity Score ≥ 16) cases aged ≥ 18 years and managed by a physician-led mobile medical team were included. Those transported to a level-II/III trauma centre were considered as undertriaged. Multivariable logistic regression was used to identify factors associated with undertriage. A total of 7110 trauma patients were screened; 2591 had an ISS ≥ 16 and 320 (12.4%) of these were undertriaged. Older patients had higher risk for undertriage (51–65 years: OR = 1.60, 95% CI [1.11; 2.26], p = 0.01). Conversely, injury mechanism (fall from height: 0.62 [0.45; 0.86], p = 0.01; gunshot/stab injuries: 0.45 [0.22; 0.90], p = 0.02), on-scene time (> 60 min: 0.62 [0.40; 0.95], p = 0.03), prehospital endotracheal intubation (0.53 [0.39; 0.71], p < 0.001), and prehospital focussed assessment with sonography [FAST] (0.15 [0.08; 0.29], p < 0.001) were associated with a lower risk for undertriage. After adjusting for severity, undertriage was not associated with a higher risk of mortality (1.22 [0.80; 1.89], p = 0.36). In our physician-led prehospital EMS system, undertriage was higher than recommended. Advanced aged was identified as a risk factor highlighting the urgent need for tailored triage protocol in this population. Conversely, the potential benefit of prehospital FAST on triage performance should be furthered explored as it may reduce undertriage. Fall from height and penetrating trauma were associated with a lower risk for undertriage suggesting that healthcare providers should remain vigilant of the potential seriousness of trauma associated with low-energy mechanisms.


Background
Major trauma is a multifactorial worldwide health issue and prehospital triage is a critical component of trauma care. Indeed, accurate prehospital triage and direct transportation to trauma centres with a high level of trauma care designation are consensually considered optimal as they may be associated with a lower risk of mortality [1][2][3]. In contrast, secondary transfer to a trauma centre may increase mortality [4]. Emergency Medical Services (EMS) caregivers, therefore, aim to avoid undertriage, defined by the American College of Surgeons Committee on Trauma (ACSCOT) as the transportation of severely injured patients to a trauma centre with a lower level of trauma care designation or another acute care facility. The ACSCOT guidelines for field triage recommend a maximum 5% undertriage and 35% overtriage to limit the potentially life-threatening impacts of 1 3 undertriage, to optimise specialised trauma resource use and to minimise healthcare expenditure [5]. The exact undertriage rate among trauma patients is unknown but two systematic reviews have recently reported global rates that range from 1 to 71.9% [6] and from 1 to 68% [7], respectively. These wide ranges were primarily due to the heterogeneous definition across studies. The underlying reasons for this undertriage have been studied and some risk factors have been identified; they include but are not limited to, advanced age [8], female sex [9], blunt trauma and low-energy trauma [10], and minority race [1]. Although, undertriage may prevent longer prehospital transfer times, and interregional paramedic transportation, it carries a heavy burden and contributes to higher morbidity and mortality rates [1,3,11], but also delay in diagnosis and treatment, missed injuries, and decreased functional outcomes [12,13].
In France, prehospital EMS is physician-led, from call reception to on-scene intervention where care is provided by a mobile medical team (MMT). This model of EMS has been shown to be an effective way to optimise prehospital triage [14,15]. However, the factors associated with undertriage and its impact on patient-centred outcomes have yet to be investigated in such EMS systems.
The main objective of this study was to assess the incidence of undertriage. We also sought to determine its potential risk factors and its association with mortality.

Study design and setting
We conducted a retrospective multicentre cohort study using the Réseau des Urgences de la Vallée du Rhône (RESUVal) trauma registry. The corresponding geographical area has a population of 3 million inhabitants and includes three level-I, one level-II, and five level-III trauma centres. RESUVal prospectively collects pre-and in-hospital (trauma room, Emergency Department [ED], and Intensive Care Unit [ICU], if applicable) information on all trauma patients managed by a mobile medical team (MMT).
The prehospital EMS in France is a 24-h-physician-led system, and out-of-hospital trauma or medical life-threatening situations are managed by the Service d'Aide Médicale d'Urgence (SAMU) (16). Country-wide 24/7 access to the SAMU is provided by a single national telephone number (15); a few calls are also received through the universal European number for emergency assistance (112). A firefighter brigade and an MMT are systematically dispatched to the scene of suspected life-threatening trauma. The MMT is composed of an emergency physician or an anaesthesiologist-intensivist, a nurse, an ambulance driver/pilot, and a medical resident (in academic centres). It can either be transported in an ambulance or a helicopter and are distributed throughout the country at hospital-based locations named Service Mobile d'Urgence et de Réanimation. After clinical evaluation, the on-scene physician and the dispatching physician determine together the most adapted care facility for the patient(s). Clinical judgement combined with a regional triage protocol based on the Vittel criteria are used to make triage decisions (see Online Table S1).

Study population
All trauma patients aged 18 years or above, managed and transported by a prehospital MMT from January 2011 to December 2017 with an ISS ≥ 16 [17] were included.

Data collection
Prehospital, Emergency Department (ED), and Intensive Care Unit (ICU) data are collected by the physician in charge of the patient. Research technicians provide continuous monitoring of the completeness and correctness of the RESUVal registry. They also collect patient outcomes at hospital discharge. Data management was performed by a data manager statistician (CC) and a methodologist (LF). MMT physicians were asked to fill out a standardised case report form for any trauma patient with at least one of the Vittel criteria corresponding to confirmed or suspected seriously injured patients. A research technician reviewed patients' medical records if the case report form was incomplete.

Study data
The following prehospital variables were prospectively recorded: age, sex, type and cause of trauma, first physiological parameters measured by the MMT (systolic blood pressure, heart rate, peripheral oxygen saturation [SpO 2 ], respiratory rate, shock index, Glasgow coma scale (GCS) score and capillary haemoglobin concentration [HemoCue ® Hb 201 + system]). Data regarding prehospital procedures and management were also recorded. The Abbreviated Injury Scale (AIS) score based on the 1998 version and the ISS were calculated after anatomical and physiological assessments were completed. In-hospital mortality was also recorded.

Outcome measure
The primary outcome of this study, undertriage, was defined as the transportation of a severe trauma patient (ISS ≥ 16) to a level-II or III trauma centre. The secondary outcome, in-hospital mortality, was defined as a trauma patient who died in the hospital (all causes). The causes of death were determined by the physicians in charge of the patient.

Univariate analysis
Baseline characteristics were described by frequencies and percentages for categorical variables, medians, and interquartile range [IQR] for continuous variables. Groups were compared using the Pearson Chi-squared test for categorical variables and the non-parametric Wilcoxon rank-sum test for continuous variables.

Multiple imputation
Candidate covariates were identified based on a literature review and univariate analysis; age, sex, season, year of admission, type of transport, time on site, transportation time, GCS, trauma mechanism, type of accident (weapon, road collision, fall, other), shock index, endotracheal intubation, FAST. The candidate covariates were complete for 61% (n = 1583) patients included in the study cohort. We performed multiple imputation by generating 100 datasets over 20 iterations to gain statistical power and reduce bias. Regressions were calculated separately on each dataset, and then combined into a single result using Rubin's rule [18]. Auxiliary variables were added to optimise the imputation process [19]: centre level, cause of injury (gunshot/stab injuries, suicide, road collision), head trauma, abdominal trauma, thoracic trauma, limb trauma, pelvis trauma, spinal trauma, prehospital administration of catecholamines, prehospital endotracheal intubation, admission to the ICU, and AIS by body zone. We used the mice package in R software, and observed convergence for all covariates [20].

Sensitivity analysis
We performed two logistic regressions with undertriage as the outcome; (1) based on complete cases (CC), and (2) post-multiple imputation, and calculated the variation rate (%) obtained as (100 × (Odds Ratio [OR] MI -OR CC )/OR CC )). We observed the presence of a selection bias because the patients dropped from the CC approach were not distributed entirely at random; the effects of age (in particular 51-65 and > 81 years), season, gunshot circumstances, and onscene time were characterised by the most significant variation rate (see online Table S2). This approach validated the relevance of the multiple imputation approach. We evaluated the sensitivity of the multiple imputation to the missing at random (MAR) hypothesis by using the method proposed by Héraud-Bousquet et al. [21]. A variation rate is then computed under different deviations from the MAR hypothesis. We found that most of the covariates were robust to a deviation to the MAR hypothesis with variation rates under 5%. Only 2 variables were slightly sensitive from a deviation to the MAR hypothesis, the on-scene time and the FAST (see Online Fig. S1). For these 2 covariates, the variation rate ranged between 5 and 15%. Given the estimated OR, a variation rate of 15% would not change the direction of the effect or significance.

Multivariable logistic regressions
We performed two multivariable logistic regressions. The first one aimed to estimate the determinants of undertriage. The results were evaluated on each imputed dataset and pooled using Rubin's rule. The second logistic regression estimated the effect of undertriage on in-hospital mortality, and only included patients admitted or deceased before admission to an ICU. This logistic regression was performed on patients without missing data (covariates and outcome). The results are presented as ORs with 95% confidence intervals (CIs) and p-values. p-values < 0.05 were considered significant.

Ethics
All patients received written information and gave oral consent to their information being used for research without the need for research ethics board approval. Given that the study was retrospective in nature, and patient information was anonymised before the analysis, a written consent was waived according to French law and the project receive approval from the national data protection commission (Commission Nationale de l'Informatique et des Libertés, CNIL; DE-2012-059), and the advisory committee on the treatment of research information (Comité consultatif sur le traitement de l'information en matière de recherché, CCTIRS). The study was conducted in accordance with the declaration of Helsinki.

Characteristics of the population
During the study period, a total of 7110 trauma patients were screened, of which 2591 had an ISS ≥ 16 and were included (Fig. 1). Overall, 75% (n = 1934) were men and the median age [IQR] was 42 years . Most patients had sustained a blunt trauma (94.1%, n = 2406), the cause was mainly a road collision (57.4%, n = 1472). Patients were transported by helicopter in 21.2% cases (Table 1).

Factors associated with undertriage
The multivariate logistic regression analysis found that patients aged 51-65 years had a greater risk of undertriage were associated with a lower risk for undertriage (Fig. 2).  AIS Abbreviated Injury Scale, FAST Focused assessment with sonography for trauma, ICU Intensive Care Unit, ISS Injury Severity Score, MGAP score Mechanism, Glasgow coma score, Age, arterial Pressure score a Among the 542 falls described, 474 (87.4%) were from high height (> 5 m), others were from "low height" (< 5 m). Among those, 7 were "fall from standing" and were admitted to a level I trauma centre b The "other" category referred included cases such as animal horn, crushing by various objects, work accident with machines… c From mobile medical team arrival on-scene to departure

Association between undertriage and in-hospital mortality
The frequency of in-hospital mortality not significantly different between under-and correct-triage groups (

Discussion
The frequency of undertriage was more than twice as high as that recommended by the ACSCOT (< 5%) [ [24]. Due to the diverse definitions of undertriage in the literature, it is difficult to compare undertriage rates between studies. This has been specifically highlighted in a recent systematic review that reported four definitions of undertriage: based on ISS, the formula for mis-triage, the need for life-saving emergency intervention, and patients triaged to a non-trauma centre [6].
Herein, patients older than 30 years of age were at greater risk of undertriage, and this was particularly significant for those aged 51-65 years, although insufficient statistical power cannot exclude to explain why age greater than 65 years was not significantly associated with undertriage. Nakahara et al. reported similar results with a trend towards a greater risk of undertriage for older patients, and this was significant for the 45-54 years age group [25]. Previous studies also suggested that age was a predictive factor of undertriage and different thresholds were reported: ≥ 55 years [25], ≥ 65 years [1,2], ≥ 80 years [26]. This is a particularly critical issue considering that older adults referred to facilities with highest designation level for trauma service have lower mortality rates [27,28]. A specific triage protocol to identify high-risk injured geriatric patients could further be explored, as it was found to lead to lower mortality [29]. Interestingly, we found that patients who sustained gunshot/stab injuries had a lower risk for undertriage than those involved in a road collision. This is consistent with the findings of Schellenberg et al. who observed a higher risk for undertriage following blunt trauma, particularly in motor vehicle collisions [10]. It is not entirely clear why, however this may be related to the severity of a penetrating trauma being more apparent to prehospital caregivers (i.e. external haemorrhage, limb amputation) than blunt trauma. Furthermore, European physicians are less exposed to penetrating trauma compared to their counterparts in America [30][31][32].
An original finding of this study is that patients with a prehospital FAST were at lower risk of undertriage regardless of the severity of their injuries. Ultrasound-based clinical algorithms might improve the assessment of injured patients, but their effectiveness in the prehospital setting lacks evidence [33]. We assume that fluid detection influenced physicians to consider patients as requiring a higher level of care; another hypothesis is, however, that physicians performing FAST had more experience and better assessment capacities. The value of FAST in our field triage algorithm needs to be further and specifically explored. Another interesting finding is that fewer patients transported by helicopter were undertriaged. This might be explained by higher availability of helicopter emergency medical services (HEMS) in level-I centres. Nevertheless, there was no significant association between the type of transport and inhospital mortality, which is in accordance with that reported elsewhere [34,35].
It is also of note that in the present study, there was no significant association between undertriage and in-hospital mortality, conversely to that reported elsewhere [1,2,36]. This may be explained by large proportion of undertriaged patients being admitted to level-II centres that offer a wide range of 24-h in-house resources (intensive care, general and orthopaedic surgeons, emergency physicians, embolization, CT scan). Hence, a patient not requiring neuro-or cardiothoracic surgery could be appropriately managed in a level-II centre.
The study has some limitations. First, the exact location of the scene of trauma was not available in the database, and we could not therefore evaluate the distance to the nearest trauma centre, which may have played a significant role in the choice of destination. Furthermore, triage may have been influenced by other elements that were not collected (i.e. chronic conditions, trauma centre overcrowding, a decision to withdraw treatments, futility, or family/patient wishes). Hence, we could not identify the exact underlying cause of the undertriage in our prehospital EMS system. More generally, although the definition of undertriage used herein is validated by the ACSCOT [5], it may also be argued that ISS is not the most accurate tool for assessing whether a patient requires level-1 trauma care compared to a clinical definition based on the need for critical intervention (i.e. urgent surgery, massive blood transfusion, craniectomy, or secondary transfer)..

Conclusion
In our physician-led prehospital EMS system, undertriage was higher than recommended. Advanced aged was identified as a risk factor highlighting the urgent need for tailored triage protocol in this population. Conversely, the potential benefit of prehospital FAST on triage performance should be furthered explored as it may reduce undertriage. Fall from height and penetrating trauma were associated with a lower risk for undertriage suggesting that healthcare providers should remain vigilant of the potential seriousness of trauma associated with low-energy mechanisms.