Traumatic brain injury (TBI) is among the major causes of mortality and disability worldwide [1]. In the United States, approximately 1.7 million patients with TBI were taken to the emergency department annually, and 52,000 patients died [2]. Moreover, 5.3 million patients have TBI-associated disability, which results in a socioeconomic burden [3-9].

Assessing injury severity and swift transport of the patient to the designated trauma center are essential to reduce the mortality in patients with severe trauma [10]. The mortality rate increased by two-fold when a patient with moderate to severe TBI was not transported to the designated trauma center according to prehospital field triage [11]. Therefore, measuring the injury severity and predicting the clinical outcome of TBI patients are critical to reduce secondary insult and improve outcome.

There are scoring systems to assess injury severity and predict outcome for TBI patients [12,13], but most include variables that are not commonly available in the routine prehospital setting, such as CT findings or laboratory results [12,14,15]. Recent studies reported that prehospital hypoxia and hypotension in patients with TBI increased mortality [16-19]; thus, systolic blood pressure (SBP) is a feasible indicator of mortality during the prehospital phase, and oxygen saturation measured via pulse oximeter is more objective and feasible than the respiratory rate (RR) included in the Revised Trauma Score (RTS) [20-24]. The Glasgow Coma Scale (GCS) is a practical method for assessing neurologic function, and the predictive power of the motor component of the GCS (motor GCS) is equivalent to that of the total GCS [25,26], which is used as a physiologic indicator in field triage [27,28].

This study aimed to develop a new trauma score optimized for prehospital care patients with TBI (NTS-TBI) to predict mortality and disability using motor GCS, hypotension, and hypoxia instead of total GCS, SBP, and RR of the RTS.

There were 13,508 cases enrolled in the EMS-ST-DB during the study period; of these, 3,642 cases were included for model development and internal validation according to the inclusion criteria (Fig. 1). A total of 426 (11.7%) of the 3,642 cases died, and 452 cases had severe disability (Table 1). Mortality was higher in urban areas, during daytime, and in level 1 emergency departments. Mortality was also higher in the shock group than that in the non-shock group (28.2% vs. 10.5%, P<0.01) (Table 2) and in the hypoxia group than that in the non-hypoxia group (36.0% vs. 9.3%, P<0.01). The incidence of severe disability was also higher in the shock and hypoxia groups.

We assessed the discriminative power of NTS-TBI models 1 and 2 by using the original data set (Table 3). The AUC value of NTS-TBI models 1 and 2 for in-hospital mortality was 0.833 (95% CI, 0.815 to 0.852) and 0.852 (95% CI, 0.835 to 0.869), respectively. The AUC of NTS-TBI models 1 and 2 for predicting severe disability was also higher than 0.75, indicating that the NTS-TBI model showed excellent performance in predicting outcome.

We also conducted an external validation by analyzing 4,497 of the 32,841 cases registered in the EMS-ST-DB from 17 provinces in 2015 (Table 4). The AUC value of NTS-TBI models 1 and 2 for in-hospital mortality was 0.779 (95% CI, 0.762 to 0.800) and 0.802 (95% CI, 0.786 to 0.828), respectively. The AUC for NTS-TBI models 1 and 2 to predict severe disability was also higher than 0.7, indicating that the NTS-TBI model showed good performance in predicting outcome.

We compared the predictive power of the NTS-TBI and RTS for mortality and severe disability (Table 5). The discriminative power of the NTS-TBI models 1 and 2 for mortality were 0.833 and 0.852, respectively, and they were significantly higher than that of the RTS (P<0.01). The AUC of the NTS-TBI models 1 and 2 for predicting severe disability was 0.772 and 0.784, respectively. NTS-TBI model 2 showed significantly better performance than that of RTS (AUC, 0.761) (P<0.01).

We developed a new scoring system to predict the clinical outcome for TBI patients and optimized it to be implemented in the prehospital care. We used motor GCS scores and oxygen saturation values obtained via pulse oximeter instead of total GCS score and RR used in the RTS. Motor GCS and oxygen saturation measured by pulse oximeter could be easily and objectively integrated into prehospital care. We developed two versions of the NTS-TBI model using the nationwide trauma registry operated by the KCDC. Both NTS-TBI models showed excellent performance as a predictor for mortality and good power for predicting severe disability in TBI patients. The discriminative power of NTS-TBI was significantly higher than that of RTS.

RTS has been used to predict outcomes for trauma patients, and the components of RTS are included in the first step of assessment in the field triage scheme [27,28,33]. However, some variables in the RTS are not easily available in the field. Thus, to improve ease of use and discriminative power, we developed an NTS-TBI. First, we replaced the GCS with the motor GCS, which has been shown to have a predictive power equivalent to that of total GCS in previous research. Moreover, the logistic model showed a more linear relationship with the outcome [25,26]. Second, hypoxia or hypotension during prehospital transport was included in the models as they have been shown to cause worse clinical outcomes such as mortality in recent studies [16-19]. Hypotension before emergency department arrival has a linear correlation with mortality [16]. Prehospital hypotension in TBI patients causes a 2.5-fold increase in mortality, and hypoxia results in a 3.0-fold increase. If combined, hypoxia and hypotension increased the risk of mortality by 6.1-fold [17]. Thus, assessing hypoxia and hypotension is essential during prehospital care to predict outcome. Third, RR has been used as a physiologic indicator to assess the severity of trauma. Given that it is difficult to measure the RR accurately in the prehospital stage [34,35], the pulse oximeter has been used as an objective tool to measure hypoxic status in diverse clinical settings [22-24], oxygen saturation reflects the patient’s respiratory function more objectively and efficiently [20,21]. The NTS-TBI included oxygen saturation instead of RR so that it can obtain more objective data and be used easily in the prehospital setting.

We reviewed diverse parameters for NTS-TBI. Age, GCS, RR, pupillary reactivity, hypotension, hypoxia, Injury Severity Score, injury mechanism, computed tomography images, and laboratory findings have been used to develop a trauma scoring system for TBI patients [12,13,15,36-39]. However, the Injury Severity Score, computed tomography images, and laboratory findings are not routinely available in the prehospital phase; thus, we developed a model that included variables that can be easily but objectively assessed during this phase.

This study has several limitations. First, the generalizability of the results is limited. Because we used data only from the EMS-ST-DB in Korea, and different trauma care systems vary among countries and could affect the performance of NTS-TBI. Second, there are inherent biases that may not have been eliminated due to the retrospective nature of the study. We retrospectively reviewed the medical records of each hospital and EMS run-sheet. Although we performed regular quality assessment, the quality of the raw medical record depends on each institution. Third, the variables in the NTS-TBI were limited to those in the EMS-ST-DB. For example, pupillary reactivity was not included and tested for model development.

In conclusion, the new trauma scoring system we developed to predict outcomes for TBI patients showed excellent performance to predict mortality. The NTS-TBI included variables that can be easily and objectively assessed in the pre-hospital setting, namely, motor GCS, hypoxia assessed using a pulse oximeter, shock, and age.