Predictors of Undertriage after Major Trauma in a Physician-led Prehospital System: Insights of a French Registry

The and of to lever 1 centers with Hence, avoid undertriage in these The this study to assess the rate and predictors of undertriage in a prehospital system. We conducted an observational multicentric, region-wide, retrospective study based on the RESUVal Trauma-System registry, Rhône-Alpes region, France. All adults assessed by physician-led EMS units, from January 2011 to December 2017 with major trauma (Injury Severity Score (ISS) ≥ 16) were included. We dened the correct-triage group as major trauma patients admitted to a level I trauma center. We performed univariate then multivariate logistic regressions with undertriage as outcome. type of trauma, clinical data: systolic and diastolic arterial blood pressures, heart rate, peripheral oxygen saturation (SpO2), respiratory rate, shock index, Glasgow Coma Scale (GCS), hemoglobin test. Prehospital data care was recorded: endotracheal intubation, thoracostomy, chest tube insertion, pelvic immobilization, Focused Assessment with Sonography for Trauma (FAST). The following scores were calculated after anatomic and physiologic assessments had been completed: Abbreviated Injury Scale (AIS) (1998 version) and ISS score, M-GAP score including Mechanism, Glasgow coma scale and Age, arterial Pressure score, New Trauma and Injury Severity Score (NTRISS) and patient outcome (in-hospital mortality).


Abstract Background
The proper prehospital triage and transportation of patients suffering major trauma to lever 1 trauma centers is associated with better outcomes. Hence, emergency medical systems (EMS) aim is to avoid undertriage in these patients. The main objective of this study was to assess the rate and predictors of undertriage in a physician-led prehospital system.

Methods
We conducted an observational multicentric, region-wide, retrospective study based on the RESUVal Trauma-System registry, Rhône-Alpes region, France. All adults assessed by physician-led EMS units, from January 2011 to December 2017 with major trauma (Injury Severity Score (ISS) ≥ 16) were included. We de ned the correct-triage group as major trauma patients admitted to a level I trauma center. We performed univariate then multivariate logistic regressions with undertriage as outcome.

Results
A total of 7,110 patients were included in the registry, of whom 2,591 patients with an ISS≥ 16. Among these patients, 320 (12.35%) were undertriaged. Median ISS was 25. In-hospital mortality was 16

Conclusions
In our region-wide, physician-led prehospital system, undertriage of major trauma was not rare. The typical pro le of undertriaged patients was a mid-aged male suffering from a blunt trauma, without respiratory distress or neurologic impairment, not bene ting from prehospital ultrasound examination and located close to a non-trauma center hospital.

Background
Major trauma remains a worldwide major health issue despite efforts and strategies to reduce its global burden. The World Health Organization reported that Injuries claimed 4.9 million lives in 2016 and that total deaths from injuries increased by 2.3% between 2007 and 2017 (1). More than a quarter (29%) of these deaths were due to road tra c injuries.
The proper prehospital triage, and transportation, of patients suffering major trauma to level 1 trauma centers is associated with better outcomes, probably so as initial referral compared to inter-hospital transfer (2,3). Therefore, emergency medical system (EMS) aim is to avoid undertriage, which de ned as the transportation of patients suffering major injuries to non-level 1 trauma centers, or other acute care facilities (3).
Many de nitions of undertriage exist and one of the most common is based on the Injury severity score (ISS), which is calculated at hospital discharge (4). Using this score, patients are classically considered undertriage if presenting an ISS ≥ 16 and transported to non-level 1 trauma centers. At prehospital scene, international guidelines use algorithms or scores to predict the probability of an ISS ≥ 16. Adherence to these guidelines adherence could reduce the risk of undertiage (5,6). The American College of Surgeons Committee on Trauma consider an undertriage rate below 5% within an ideal organized trauma system (3). Therefore a recent systematic review reported within international studies an undertriage rates ranging from 1 to 71,9% (4). Few studies have investigated the characteristics of undertriaged patients, in a physician-led prehospital system. The French prehospital system is physician-led, both at call reception (for dispatch) and on scene (delivery of care). Physician-led prehospital triage system has showed its effectiveness in optimizing prehospital triage and in a regional trauma system in mountain region (7,8).
In a neighboring geographical area, the use of a grading system signi cantly decreased the risk of undertriage by 0.47 (95% CI 0.40 to 0.56) with the ACSCOT de nition of undertriage, and by 0.33 (95% CI 0.26 to 0.42) with the TRENAU de nition (7). Our regional trauma system was implemented in 2011 and comprised composite geographical area with urban and non-urban area. We tested the hypothesis that our trauma system was mature with an undertriage rate under 5%.
The main objective of this study was to therefore to assess the rate of undertriage in a physician-led prehospital system, and identify predictive factors leading to undertriage. Secondary objectives were to describe the patients who were managed for trauma by a physician-led mobile medical team (MMT) and evaluate the mortality.

Study design and settings
We conducted a multicentric retrospective study based on a trauma registry (RESUVal: Réseau des Urgences de la Vallée du Rhône) based in the Rhône Valley in France. Population area concerns 2 level I trauma centers, 1 level II and 6 level III covering a population of 3 million inhabitants (Fig. 1). The Trauma-system registry prospectively collects prehospital and in-hospital pathways of care of adult trauma patients. Since 2011, a total of 7500 patients managed for trauma by a MMT were enrolled using a structured and standardized case report form (CRF). The pathways of care are reported by emergency physicians in charge of the patient from the pre-hospital management, to the emergency department, and if applicable to the intensive care unit until hospital discharge.
The pre-hospital emergency medical system in France is a two-tiered, physician-led system, and out-ofhospital emergencies are managed by the Service d'Aide Médicale d'Urgence (SAMU). Emergency calls from bystanders are centralized to the telephone number "15" available 24/7. Then the dispatching physician can activate an MMT in critical cases (9). The MMT can either be a ground-or helicopter ambulance. MMTs are distributed throughout the region at hospital-based locations named Service Mobile d'Urgence et de Réanimation (SMUR). In suspected life-threatening injuries, both a re ghter team and an MMT are systematically dispatched to the scene. After clinical evaluation, the on-scene emergency physician and the dispatching physician determine together the most adapted care facility to refer the patient(s) according to a protocol based on Vittel criteria (7).

Patients selection
All adults managed from January 2011 to December 2017 with ISS ≥ 16 were included in this study. Then, we de ned the correct-triage group as major trauma patients admitted to one of the two regional level I trauma centers. Other patients were categorized in the undertriage group.
Patients included received written information about the objectives of the trauma-system registry in accordance with French legislation. The registry received approval from the National Commission for

Data collection
The registry was developed in 2011 to prospectively collect data related to the management of injured patients, from the scene to hospital admission (emergency department or intensive care unit) by the MMT physician. Research technicians provided continuous monitoring of the completeness and correctness of the database. They also collected patient outcome at hospital discharge.
The following prehospital items were prospectively recorded: age, sex, patient treatment (antiplatelet, anticoagulant), type of trauma, clinical data: systolic and diastolic arterial blood pressures, heart rate, peripheral oxygen saturation (SpO2), respiratory rate, shock index, Glasgow Coma Scale (GCS), hemoglobin test. Prehospital data care was recorded: endotracheal intubation, thoracostomy, chest tube insertion, pelvic immobilization, Focused Assessment with Sonography for Trauma (FAST). The following scores were calculated after anatomic and physiologic assessments had been completed: Abbreviated Injury Scale (AIS) (1998 version) and ISS score, M-GAP score including Mechanism, Glasgow coma scale and Age, arterial Pressure score, New Trauma and Injury Severity Score (NTRISS) and patient outcome (inhospital mortality).
Outcomes of the study Primary aim was to assess the rate of undertriage, and predictive factors among patient epidemiology and their injury's characteristics. Secondary aims were to evaluate the association of undertriage with mortality, and to determine the maturity of the departmental trauma network, as de ned by the evolution of undertriage rate from 2011 to 2017.

Statistical analysis
Univariate analysis Baseline characteristics were described by frequencies and percentages for categorical variables, medians and interquartile range (IQR) for continuous variables. For categorical variables, percentages were computed among known data and the denominator was reported in tables. Comparisons between groups were assessed using the Pearson Khi² test for categorical variables and the non-parametric Wilcoxon rank test for continuous variables. We tagged a "X" in tables when statistical power was limited (< 10 patients per group). We also provided a temporal trend using Khi² test for trend, referring to "ptrend".

Multiple imputations
To perform logistic regressions, we de ned candidate covariates based on the 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, fall, other), shock index, endotracheal intubation, FAST. Among the 2, 591 included patients, the candidate covariates were 61% complete cases (n = 1583). To gain statistical power and reduce bias, we performed multiple imputations (MI) by generating a total of 100 datasets over 20 iterations. Regressions were calculated separately on each dataset and then combined into a single result using Rubin's rule (10). Auxiliary variables were added to optimize the imputation process (11): center level, circumstances (aggression, suicide, accident), head trauma, abdominal trauma, thoracic trauma, limb trauma, pelvis trauma, spinal trauma, prehospital administration of amines, prehospital intubation, prehospital triage, hospital endotracheal intubation, blood transfusion at shocking room, triage evaluated at the shocking room, hospitalization in resuscitation unit, AIS by body zone and Glasgow Outcome Score (GOS) at discharge. In order to avoid circularity in the imputation model, a post processing has been added on the GOS so that it is not imputed for patients who are not hospitalized in intensive care. We used the mice package in R software (12), and observed convergence for all covariates.

Sensitivity analysis
We performed two logistic regressions with undertriage as outcome; 1) based on complete cases (CC) and 2) post-MI, and calculated the variation rate (VR, %) obtained as (100*(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 completely at random (MCAR); the effects of age (especially 51-65 and > 81 years old), the season, the gunshot circumstances and the on-scene time were characterized by the largest VR (see Additional le 1). This approach validated the relevance of the MI approach.
We evaluated the sensitivity of the MI to the missing at random (MAR) hypothesis by using the method proposed by Héraud-Bousquet et al. (13). This method upweights imputations which are more plausible under missing not at random (MNAR). A variation rate is then computed under different deviations from the MAR hypothesis. We showed that most of the covariates were robust to a deviation to the MAR hypothesis with VR under 5%. Only two variables were slightly sensitive to a deviation to the MAR hypothesis, the on scene-time and the FAST-Echo (see Additional le 2). For these two covariates, the VR was ranged between 5% and 15%. Given the estimated OR, VR of 15% would not change the effect direction or signi cance.

Multivariate logistic regressions
A rst logistic regression was used to estimate undertriage determinants as primary outcome based on 2,591 patients. The results were evaluated on each imputed datasets and pooled using the Rubin's rule.
We performed a second logistic regression based on the 1,724 patients admitted or deceased before admission in intensive care unit to estimate the effect of undertriage on mortality as secondary outcome, considering in-hospital death. This logistic regression was performed on complete cases (covariates and outcome) for 91% of the patients (1,570/1,724). We did not use the imputed datasets from previous regression because each imputed dataset would have had different number of patients hospitalized in intensive care unit, the Rubin's rule would be impracticable. The covariates were age, sex, trauma mechanism (blunt/penetrating), pre-hospital Glasgow score, pre-hospital triage, ISS and undertriage.

Overall Characteristics
From January 2011 to December 2017, 7,110 patients were screened for inclusion from the Trauma-System registry. In total, we included 2,591 patients with an ISS ≥ 16 for analysis. Among them, 320 (12.35%) were de ned as undertriaged (Fig. 2). The median age was 42 years old (interquartile range [IQR], 27-59) years. A total of 1,934 patients (74.64%) were men. There was no statistically signi cant difference in sex or age between groups (p = 0,815 and p = 0.0547). The majority of patients suffered from blunt trauma (n = 2406, 94.13%). Road tra c accident was more frequent in the undertriage group than in the correct-triage (66.56% vs 56.14%, p = 0.0006). Median on-scene time of pre-hospital management by MMTs was 36 minutes , and total prehospital time was 60 minutes [44-82]. It was shorter in the undertriage group compared to the correct-triage one (57.5 vs 60 min, p = 0.0131). More than three-quarters of patients (78.77%) were transported by ground ambulance but patients from the correct-triage group bene ted more from air transportation (22.37% vs 12,12%, p = 0.0002) ( Table 1).  Primary outcome In total, 320 patients (12.53%) were undertriaged. The results of univariate and multivariate analyses of global population characteristics are shown in Fig. 3

Discussion
In our study, undertriage rate was 12.35% among patients included in our regional database as they received rst care at non-level I centers. This result is above the recommendations of less than 5% as set by the ACSCOT (3). Many European and American studies reported similar or even higher undertriage rate (4,14,15,16). In a comparable health system also bene ting from physician on scene and medical dispatching, French authors reported higher undertriage rate about 18% (7). Voskens et al. reported a 21.6% (95% CI,18.0-25.7) rate of undertriage. In their study, allocation of trauma patients to the appropriate level of trauma care was guided by the Dutch Field Triage Protocol and their ambulance care system was nursebased (17).
Our ndings showed no association between undertriage and in-hospital mortality, contrary to other studies (6,18,19). Probably because a large amount of undertriaged patients (n = 202, 63.1%) were admitted in the center considered as level II because there is no neurosurgery department nor thoracic surgery department but yet offers large 24-hour in-house resources (intensive care, general and orthopedic surgeons, emergency physicians, radiology). Some of these patients were secondarily transferred to a level 1 center but no data were available about the reason of such transfer, or about vital status at secondary hospital discharge. Pickering et al. reported no difference in outcomes for direct referral to a trauma center versus initial triage to a local hospital in their systematic review (19).
In our study, penetrating trauma were rare, counting for only 4.56%. This is consistent with other French and European series. An interesting nding was that compared to road tra c accident, fall and gunshot/stab wounds were protective for undertriage. Schellenberg et al. also reported higher risk for undertriage following blunt trauma, especially motor vehicle collisions (21). When the gravity is not as obvious as these penetrating injuries, we should search for other predictors to guide eld triage. In our study, any age showed a tendency toward higher risk of undertriage, except age of 51 to 65 years that was a predictive factor of undertriage. Nakahara et al. found the same result in the group 45-54 years (22). On the opposite, most studies reported higher age cut-off for undertriage risk: ≥ 55 years (14), ≥ 65 years (6,18), ≥ 80 years (23). In our study, the eld triage criteria for level I were: respiratory failure (SpO2 < 90% despite oxygen supply), hypotension (MAP < 100 mmHg despite 1000 mL of uid resuscitation) GCS ≤ 8, use of vasopressor or blood transfusion. Hence, a possible explanation is that in our trauma system, neither age ,nor comorbidity excluded referral at level 1 centers. Indeed, previous studies have reported improved outcomes for elderly patients admitted in higher levels of trauma care (24)(25)(26) Some authors even suggested that treating severely injured elderly trauma patients in designated geriatric trauma centers or units may be associated with statistically signi cant gains in likelihood of survival (27,28). Bradburn et al. assessed a speci c protocol to identify high-risk geriatric injured patients which lead to signi cant reduced mortality (29) Several studies found that patients suffering from head injury were more affected by undertriage (14,15). We did not report that trend. On the opposite, GSC ≤ 8 was a protective factor for undertriage. So was the need for endotracheal intubation regardless of the GSC.
A new nding of this study is that patient who bene ted from a FAST were less undertriaged regardless of hemodynamic condition. Ultrasound-based clinical pathways might enhance the speed of injured patient assessment. But effectiveness of ultrasound-based clinical pathways to triage algorithms lacks evidence and its bene t remains unclear (30). In our study, FAST was not mandatory, not integrated in the eld triage algorithm and results were not available. We ignored if triage decision was or not based on FAST results. We hypothesized that when an effusion was detected, physicians would consider patients to require higher care (e.G. a level 1 center referral). Another hypothesis considered that physicians performing FAST were more experienced with better assessment capacities than other colleagues. In our cohort, ultrasonography was performed in 2.7% cases in 2011 while it was in 34.9% in 2017. It may contribute to lower undertriage rate in the next years. Further explorations are needed to integrate FAST in our eld triage algorithm and assess its performance.
Air transport also appeared to be protective against undertriage compared to ground transport regardless of severity (ISS) and transit time from scene to hospital. It might be explained by highest availability of helicopter emergency medical service (HEMS) in level I center. Therefore, we assumed that physicians using HEMS tended to facilitate the referral of patients to their level 1 center instead. Nevertheless, we showed no in uence of transport on in-hospital mortality. Other studies did no report strong evidence of any bene ce of HEMS on in-hospital mortality among injured patients ( This study has several limitations. First, exact locations of the on-scene trauma were not available in the database. Thus we could not evaluate the distance towards the nearest trauma center and de nitive destination. Whether the physician decision was based on the triage algorithm, their own clinical judgment or any other element was not reported. Some missing data could also have in uenced the destination such as chronic conditions, result of the FAST, trauma center overcrowding.

Conclusions
In our region-wide, physician-led prehospital system, undertriage of major trauma concerned 12.35% of major traumas. The typical pro le of undertriaged patients was a mid-aged male suffering from a blunt trauma, without respiratory distress or neurologic impairment, not bene ting from prehospital ultrasound examination and located close to a non-trauma center hospital.
Despite the creation of our regional trauma system, and the physician-led process of on-scene care delivery and adjusted referral of patients, the undertriage rate has not signi cantly decreased between 2011 and 2017. The need for triage tools that reinforce medical expertise is probably necessary to improve the quality of care. Availability of data and materials Determinants of undertriage