Our study has shown that significant predictors for mortality of chest trauma patients were low GCS, hypotension on presentation and increased severity of injury. Old age was reported as a predictor of mortality for chest trauma [2, 4, 5, 11]. Liman et al. have shown that an age of more than 60 years significantly increased mortality . Patients’ demographics (age, gender and nationality) did not predict mortality in our logistic regression model. The UAE population is a very young population composed mainly of male workers [12, 13]. Elderly expatriate patients prefer to retire at their homeland. Having small number of elderly patients in our study may have resulted in the non significant effect of age on mortality.
Majority of chest trauma in our community is blunt in nature. There is less than 1% penetrating trauma. The strong law enforcement in our community contributes to the reduction of domestic violence because aggressors may be deported to their homeland if they cause physical injury to others.
Univariate analysis has shown that the mechanism of injury, mechanical ventilation, and admission to the ICU, head and chest AIS, ISS, GCS and systolic blood pressure on arrival to hospital were significantly different between those who died and those who survived. The multivariate logistic regression model excluded all these factors except ISS, GCS and systolic blood pressure as significant predictors of mortality.
Our cut off values for prediction were systolic blood pressure of less than 103 mmHg, GCS lower than 10.5 and an ISS score higher than 11.5. In contrast, Emircan et al. found the same significant factors with different cut off values. These were hypotension (in general), GCS lower than 13 and an ISS higher than 22 . Patients having a lower ISS, compared with Emircan et al., died in our setting. We think that this occurred because our prehospital care is relatively not well developed. Lack of pre-hospital care increases mortality  while improved pre-hospital care reduced mortality in chest trauma patients .
RTC is the most common mechanism of chest trauma . Abbas et al. have described the biomechanics of chest trauma in RTC . They defined three mechanisms of injury occurring in front impact collisions of unrestrained drivers. This includes the impact of the chest to steering wheel, and the head injury to the wind shield. A recent prospective study has shown that more than 80% of RTC patients in the UAE were unrestrained . This finding explains the high frequency and severity of head injury in our study. Seatbelt usage reduces severity of head injury . Emircan et al. highlighted that the mechanism of injury is a significant predictor of mortality. The most common associated extra-thoracic injury in our study was to the head; which reflects the low usage of seatbelts .
Similar to others, our study has shown that chest injuries associated with extra-thoracic injuries were more common than isolated chest injuries [2, 5]. In contrast, Demirhan et al. reported the opposite . This can be explained by selection bias. Presence of associated extrathoracic injuries, especially abdominal injuries increases the mortality [2, 11]. Splenic and hepatic injuries increase the risk of death by threefold . In contrast, the type of intra-thoracic injury per se was not a predictor for mortality in chest trauma patients .
Limitations of the study
We have to note that our study includes only patients who were admitted in the hospital for more than 24 hours or those who died in the Emergency room. It does not include patients with more severe injuries who died before arriving to the hospital, those with minor injuries who were treated in the Emergency Department, and those who were hospitalized for less than 24 hours. Pre-hospital care details are not available in our data. This is a major limitation of our study because pre-hospital care can be a major predictor of mortality.
Furthermore, our data represents the period before 2006 which may not exactly reflect what happens at present. Establishing the Trauma Registry of Al-Ain Hospital was a specific limited research project supported by the UAE University. Nevertheless, we, think that the factors affecting mortality of chest trauma did not change since then. Despite the fact that pre-hospital care has improved, the observed severity of head injury in our setting is still high because of the very low compliance of seatbelt usage .
Furthermore, other important predictors for chest trauma mortality which were reported in the literature are missing in our study. This includes the number of fractured ribs, the presence of co-morbidities, and the use of seatbelts. A strong correlation between the number of fractured ribs and mortality was demonstrated by several authors [5, 19, 20]. Presence of co-morbidities was also a significant predictor of chest trauma mortality . Data on co-morbidities were not available in our study. Nevertheless, we anticipate that this factor was not important in our young population.
There are two cautions that have to be addressed in this analysis, the missing data and the linearity of the model. We assumed that missing data of a variable were random and were not replaced by the average of that variable. Since there is a risk that some variables were eliminated by the backward logistic regression model because of missing data, a direct logistic regression model for the significant factors was redone. The significant factors were the same supporting our assumption that missing data were random.
To address the linearity concern, we have entered ISS2 and (systolic blood pressure)2 to the direct logistic model and these new variables turned out to be non significant. Furthermore, we categorized GCS into three categories; severe head injury (GCS=3-8), moderate head injury (GCS=9-13), and mild head injury (GCS=14-15) and redone the direct logistic regression model. The odds ratio of death of severe head injury was 61.4 (95% CI, 6–630) compared with mild head injury. The odds ratio of death of moderate head injury was 19.6 (95% CI, 1.92-200) compared with mild head injury. We could have used a more robust test for linearity like the fractional polynomial analysis instead of the methods used. Nevertheless, it would have been difficult to prove the exact linearity of the model because of the small sample size. We think that the above findings justify the assumption of approximate linearity in our logistic regression model given that the results are interpreted cautiously.
Finally, it is recommended to have ten events per each variable entered in the logistic regression model . We have entered eight variables having 34 events (deaths) into the model. This ideally needs 80 events for proper statistical analysis. Nevertheless, other methodologists  have recently argued that “the rule of ten” should be relaxed. We think that our results are still valid if interpreted cautiously .