Despite the progress made in care over the last few years, trauma continues to be the most common cause of death in people under 45 in most countries. An important percentage of these deaths can be prevented.

Monitoring quality of care and outcomes is essential to improve results. Major differences have been reported in the quality of care during the management of severe trauma among different countries and hospitals.

Of the 3 components of the Donabedian's healthcare quality model (structure, process and, outcomes), the assessment of outcomes is probably the ideal method.1

A global measure of the outcomes in severe trauma care is mortality. It depends on the circumstances of trauma and care provided, which justifies the growing interest in the mortality prediction systems that combine multiple variables such as quality indicators. Although over the last 30 years multiple tools have appeared there is no unanimity as to what system is the best predictor of mortality.

Of the different models available the Trauma and Injury Severity Score (TRISS) is the most commonly used. It was developed from the Major Trauma Outcome Study2 in 1983 by the American College of Surgeons Committee on Trauma to develop a reference standard to assess the quality of care in the management of trauma. This tool consists of a multivariate logistic regression system that estimates a probability of survival (PS) from the mechanism, anatomical injuries (assessed by the Injury Severity Score [ISS]), physiological repercussion they produce (assessed by the Revised Trauma Score [RTS]) and the subject's functional reserve (assessed according to his age).

The coefficients applied to estimate PS have been updated from the National Trauma Data Bank that Schluter et al. database built back in 2009.3

Back in 2004, the Trauma Audit & Research Network developed a logistic regression model based on their own registry including age, sex, injury severity score (ISS) and score obtained on the Glasgow scale. In 2014 this model was updated, the coefficients changed, and comorbidity was added to the model (quantified by the modified Charlson Comorbidity Index [mCCI]). The details of the estimates of both models appear in the methodology.

Other European trauma registries that we should mention here are the German (German National Trauma Registry) and the Scandinavian ones (Scandinavian Networking Group for Trauma and Emergency Management). Since 2008 the foundations are being laid to create a new common European registry capable of unifying all existing registries (European Trauma Registry Network).

The main objective is to confirm the utility of TRISS models (the original one and the one with updated coefficients) and PS14 at a trauma ICU of a tertiary care center.

The secondary objectives are to assess the validity of the survival prediction models mentioned before by calculating their discrimination and calibration, and studying the different profiles of patients (age, sex, type of trauma, and severity of anatomical injuries).

During the period of study 1817 trauma patients were admitted to the unit (Fig. 1). Of these patients, 1240 (68.24%) met the inclusion criteria. All the patients’ personal information and demographic data are shown in the main registry. However, based on the information of their past medical history, the reports from other centers and pre-hospital services, it is not possible to estimate the ISS in 25 patients, the RTS in 48 patients, the TRISS in 77 patients, and the PS14 in 35 patients, which in relative terms means losses <10%.

Figure 1.

Flow diagram of the study.

(0.11MB).

The patients included in the study are mostly males (80%) with an mean age of 40 years (29.7–55.0) (Table 1).

Regarding their origin, 969 of them (78%) were brought by the pre-hospital services, 149 (12%) were transferred from the ER, and 121 patients (10%) were secondary transfers from other centers.

A clear prevalence of closed traumas can be seen (92%) as opposed to penetrating traumas (8%). The most common injury mechanism in this cohort is traffic accidents (42%) followed by autolysis, and aggression (12% and 10%, respectively).

Before being admitted to the hospital, low blood pressure (SBP<90mmHg) is confirmed in 357 patients (29%), Glasgow scale scores≤8 in 414 patients (34%), and an RTS score<12 points in 834 patients (67%). The mean ISS during admission of the overall sample was 26.02. During pre-hospital care, 585 patients (47%) required secure airway management. During their ICU stay a total of 190 patients died (15.32%). Of the remaining 1050 patients who were discharged, 976 (79%) were transferred to the hospital floor, and 19 of them eventually died (1.95%).

The W-statistic values for the TRISS, updated TRISS, and PS14 models applied to our global population with their 95% confidence interval were 6.72 (5.03–8.41), 1.47 (−0.19 to 3.15), and 2.74 (1.1–4.39), respectively (Fig. 2).

Figure 2.

Graphic representation of the observed and predicted mortality and W-statistic value for the TRISS, updated TRISS, and PS14 models applied to our overall population.

(0.12MB).

To study the ability to discriminate of each prediction model, the ROC curves were obtained (Fig. 3). The values of the area under the ROC curve (AUC) for the TRISS, updated TRISS, and PS14 models with their corresponding 95% confidence intervals were 0.91 (0.89–0.93), 0.91 (0.89–0.93), and 0.92 (0.90–0.94), respectively. After using the DeLong test to compare the AUC values of the different models, no significant differences were found between the TRISS model and the classic and updated coefficients (P=.44), between the classic TRISS model and the PS 14 (P=.75) or between the TRISS model with updated coefficients and the PS 14 (P=.67).

Figure 3.

ROC curves of the 3 predictive models (from left to right, TRISS, updated TRISS, and PS14).

(0.11MB).

The Hosmer–Lemeshow statistical values for the TRISS, updated TRISS, and PS14 models were 29.9, 56.2, and 49.9, respectively; all of them with P values <.05.

The GiViTI calibration belt for each model (with representation of the 80% and 95% confidence interval) is shown in Fig. 4.

Figure 4.

GiViTI calibration belt of the TRISS (left), updated TRISS (center), and PS14 (right) models in the overall sample. The area shown in light gray represents a 80%CI and the one shown in dark gray represents a 95%CI.

(0.15MB).

Data of the subgroup analysis are shown in Table 1.

The 3 models share high discriminating capacity and low calibration. This pattern had already been reported. The AUROC values are consistent with those from other studies conducted in our field.6 The low calibration capacity of the TRISS model had already been reported, and it is common to other models applied in the intensive medicine setting.7,8

Regarding the TRISS 2010 model, other authors have seen that mortality rate is underestimated when it is <60% and overestimated when it is >60%.6 The calibration belt obtained in our study for this model is consistent with this tendency seen. Underestimating the severity of the patients’ severe TBI patients and that of old patients is another limitation reported by the authors. This limitation can be responsible for the unfavorable difference seem between the survival observed and predicted in patients >70 years, which was not seen when model PS14 was used. Probably the latter model is a better predictor of severity in these patients. On one hand, it is not based on a dichotomic division of age (it rather assigns different coefficients to 8 ranges of age). On the other hand, it includes comorbidity as one of the variables, which is more common in these patients. In this sense, treating age as a continuous variable would be associated with better calibration values. However, the NORMIT model built from the Norwegian registry where age is included as a cubic (continuous) function also shows poor calibration values.7

Another aspect that can be associated with low the calibration values of both models is the use of the ISS scale to quantify the severity of the injuries (that overlooks the presence of several serious injuries in the same anatomical region). A possible alternative to this scale would be the use of the NISS scale in the prediction model (as it is the case of the NORMIT scale developed from the Norwegian model).7 Other simpler survival prediction models such as the BIG (that don’t use the ISS calculation) show lower discriminating capabilities.9

The Glasgow scale score stands out among the multiple elements involved in PS calculation in both models. However, it has been criticized due to its complexity, subjectivity, and low reproducibility.10 In the case of trauma patients, this scale is often affected by artifacts resulting from alcohol or drug abuse.7

We should mention here that none of the models studied included analytical values as it occurs with the German model (RISC) that incorporates base excess and prothrombin activity and shows discriminating capabilities that can be compared to the TRISS system.11

The disparity in the differences seen when applying the classic TRISS methodology and TRISS 2010 methodology can be explained by the 30-year difference that separates both scales, during which trauma care has improved substantially.

When comparing our results to those of a contemporary British population using the PS14 model, we saw a favorable statistically significant difference.

The group of old patients is especially important given that they it is a growing group in our center. This group is characterized by having higher mortality rates compared to the overall mortality rate. In our series, the mortality rate of patients >70 years was 45%, similar to that reported by other Spanish groups. An unfavorable survival difference was seen in our center when the TRISS methodology was used that was not confirmed when the PS14 model was used.12 This inconsistency is probably associated with the calibration of the models for this population subgroup as we have already mentioned.

Regarding the female population, it is possible to see higher mortality rates compared to males (RR, 1.47, 95%CI, 1.05–2.07) despite the fact that there are no statistically significant differences in age, ISS, GCS or RTS. These results may be due to differences in the injury mechanisms or patterns or to sex related different responses to trauma. In this sense, mortality and side effects have been postulated in women with TBI compared to males. This may be associated with hormonal, metabolic (that condition the response to treatment), and functional factors of the central nervous system.13

We saw an unfavorable difference in the mortality rate associated with penetrating trauma, which does not reach statistical significance in the PS14 prediction model. This difference can be due to several factors. On the one hand, it may be due to the scarce percentage of patients it represents with respect to the overall number of patients, which speaks of much less experience in the healthcare provider. On the other hand, it can be linked with a calibration error of the predictive model, something that has already been described by other authors.14

In patients with severe TBI (depending on the injuries, defined as MAIS>3) in our center we obtain mortality results that are consistent with the American ones and slightly better than Anglo-Saxon results. The overall mortality of these patients (34%) is acceptable according to the trauma quality markers proposed by SEMICYUC.1

Another subgroup that stands out is that of patients with severe pelvic trauma (defined as MAIS in such region >3), where survival is strikingly higher compared to the one expected for both models. We believe that these results can be associated with our multidisciplinary protocol agreed with prehospital, interventional radiology, trauma and surgery units for a fast, systematic, effective, and comprehensive care of these patients.

This study has some limitations. It was a single-center study whose results probably cannot be generalized to the healthcare provided in other units.

Because this is a retrospective study across time, we are facing problems such as data and patient loss. The variable that is more often incorrectly recorded is respiratory rate, a crucial parameter for the TRISS model. This associated a small but tolerable loss of the patients included initially (3.7%), but it could jeopardize its internal validity. Arterial blood pressure is another variable that was incorrectly recorded partly due to its changing character during initial care.

Lastly, we should not forget that the methodologies used calculate the probability of survival based on the results seen in US (TRISS) and UK hospitals (PS14). The use of such methodologies to assess the ICU of a Madrid tertiary care center has some limitations given the international differences among centers. The resources, experience, type of trauma and patients, and healthcare protocols can vary significantly from one country to the next, which can be a confounding factor if we use it in our setting.

In conclusion, with the objective of assessing the quality of the chain of healthcare provided for the management of patients with severe trauma at our hospital, we tried to compare our own center to the international standards. For the lack of a single international model, and given the limitations found in our study, we believe that the best thing to do would be to create a methodology for evaluation purposes from the data obtained from Spanish hospital ICUs and then compare the results more reliably. A trauma registry would be of special interest in the development of such methodology, and its utility has already been recognized by the WHO.12 Several initiatives have been published in our country being the RETRAUCI project (developed by SEMICYUC1 Trauma and Neurointensive Care Working Group) the most important of all.

None reported.