Prognostic models for the early care of trauma patients: a systematic review

Trauma is a major global contributor to premature death and disability. The burden of injuries is especially notable in low and middle-income countries and is expected to rise during the coming decades [1, 2]. Harm from major trauma may be minimized through early access to pre-hospital [2] and in-hospital trauma care [3]. A majority of trauma related deaths occur during the pre-hospital period or in the initial hours after injury. Emergency medical service (EMS) providers must therefore rapidly assess trauma severity in order to identify patients that require prompt referral to an appropriate hospital [2, 3] and to ensure that necessary diagnostic and therapeutic interventions are initiated upon admission. However, early recognition of major trauma remains a challenge due to occult injuries, unpredictable evolution of symptoms, and the complexities of evaluating patients in the early hours after injury.

If patients only suffering minor injuries bypass the local clinic (overtriage; false-positives), the regional hospital will be overwhelmed and create a strain on scarce financial and human resources. However, if major trauma victims are treated at the local clinic rather than being stabilized and rapidly transported to a facility providing higher level of trauma care (undertriage; false-negatives), avoidable deaths may occur. Sensitivity and specificity are often negatively correlated making optimal prognostic model performance a balance between patient safety and optimal resource utilisation. American College of Surgeons-Committee on Trauma (ACS-COT) therefore describes 5% undertriage as acceptable and associated with an overtriage rate of 25% - 50% [4].

At hospital admission, delay to high resource resuscitation can result in unfavourable outcome [5, 6]. Traditionally, these early decisions have been informed by the patient's injury severity. In this context, severity has been defined by the patient's risk or prognosis. Although commonly used interchangeably, risk and prognosis differ in their meaning. Prognosis can be defined as "the probable course and outcome of a health condition over time" [7]. Risk is sometimes used as a synonym of probability, but it can also used as a synonym for hazard [8]. We believe the term prognosis is more appropriate in this context and will use this term throughout this manuscript.

Assessment of injury severity traditionally includes clinical findings pertaining to physiological derangement, obvious anatomical injury, mechanism of injury, and pre-injury health status. These individual variables have been useful to predict a patient's prognosis in trauma (i.e. predictors), but have showed limitations when used as isolated parameters [9].

To overcome the limitation of individual characteristics, different predictors can be combined into scores or models to estimate patient's prognosis and guide EMS providers in their early evaluations of these patients. Prognostic models in the context of trauma are also referred to as risk models, prognostic scores, triage scores or risk scores. The abundances of prognostic models in the trauma setting indicate not only the need for early objective quantification of prognosis, but also the difficulties of addressing all requirements to be valid, precise and practical.

The ideal prognostic model for trauma should be developed following methodological guidelines, it should be clinically sensible, well calibrated and with good discriminative ability [10, 11]. Further, it is cost-effective, externally validated, field-friendly and it provides useful information to EMS providers that improve triage decision-making and patient outcome [12‐15]. We aim to conduct a systematic review that identifies existing prognostic models aimed at improving early trauma care, appraise their quality and describe their characteristics and performance in order to inform clinical practice and future research.

This systematic review located 5 prognostic models for the early care of trauma patients. The majority of models were developed in cohorts of trauma patients from the 80's. All except one of the models were developed in populations from high-income countries. The number of predictors included in the models ranged from three to five, and SBP was the only predictor included in all models. GCS has proven to predict the need for trauma centre admittance [36], but have been criticized for being difficult to score correctly [37, 38]. Reflecting this, variously defined predictors depicting consciousness were included in all models. All the prognostic models evaluated survival as an outcome, although the timing was defined differently for all the models. Further, we revealed heterogeneity in outcomes other than survival highlighting the consensus among researchers regarding a common definition of "major trauma" is needed (see additional file 4; Characteristics of investigated outcomes").

All the models, except PSS, were validated in external populations. The T-RTS was the most frequently validated (7 studies). The performance of the prognostic models showed a large variation between different validation studies (see table 2), although the majority of studies were conducted on populations from USA and Canada. The reason for these differences can be related to methodological issues, such as different variable definitions or alternatively it could be related to the difficulty of transporting prognostic model to different settings. Factors that may affect the transportability of prognostic factors could be related with injury characteristics (e.g. penetrating injuries), patient's characteristics (e.g. age), or medical services characteristics (e.g. pre-hospital transportation distances or level of EMS personnel competence).

Importantly, although 80% of trauma deaths occur in low and middle-income countries where many of these characteristics are likely to be different from developed countries, we did not find any model that was developed and validated for this setting [1]. Trauma care providers in low and middle-income countries should have access to prognostic models derived in cohorts including patients from these populations.

Although prognostic models for trauma should be developed following methodological guidelines, the quality appraisal revealed several areas of improvement for most models. We found methodological limitations pertaining to issues such as inadequate methods to develop the prognostic models, handling of continuous variables and dealing with missing data. The MGAP was the one that fulfilled most of the recommended methodological quality items.

For a prognostic model to be used it should be well accepted by EMS providers. However, none of the studies evaluated the "acceptability" and "practicality" of the prognostic model. For a model to be effective it should improve patients' outcomes when tested in a randomised clinical trial, nevertheless the impact was not evaluated for any of the models. All models successfully discussed the rationale to include the predictors and included clearly defined predictors and valid outcomes.

We acknowledge that his systematic review has limitations. Some relevant studies may not have been located during our database search. Our literature review was only conducted in MEDLINE, although several other databases exist. The search strategy used in MEDLINE performed with high sensitivity (4 939 records retrieved) and low specificity (14 included studies). We identified three of the included studies through alternative sources (bibliographies); however, all 14 studies are included in MEDLINE. Closer examination of the included studies indicated inconsistent indexing of articles on prognostic scoring in adult trauma on MEDLINE. In the future, more homogenous reporting of studies pertaining to prognostic trauma models may reduce these limitations. Further, our exclusion of non-English language has contributed to the risk of missing relevant studies. However, we identified all the models included in a recently published triage guideline [39].

We only identified 9 validation studies indicating a need for further evaluation of performance transportability. In order to be able to evaluate the validity of future prognostic models we recommend to report the items included in our quality appraisal list (see additional file 3) as well as other relevant standards for reporting [40, 41].

Our review should be incentives to further evolve the accuracy of prognostic models for the early care of trauma patients.

This systematic review located and appraised the quality of five prognostic models for the early care of trauma patients. The prognostic models reported various outcomes pertaining to major trauma, but all models evaluated survival as an outcome. The general impression is that all models predict survival adequately. The MGAP fulfilled most of the suggested methodological quality items and is recommendable for routine use. However, there are many areas for improvement, including model development, analysis of continuous measures, handling of missing data, practicality and impact analysis.