This analysis is an ancillary study of the FIRST epidemiological study which was conceived in order to prospectively gather pre-hospital and hospital data about patients with severe blunt trauma 8. According to French law (law 88-1138, pertaining to biomedical research, 20 December 1988, modified on 9 August 2004 9), this non-interventional study did not require approval by an ethics committee or written informed consent from patients. The study was presented to and approved by the National Commission for Data Processing and Civil Liberties (authorization number 05-1059, confirmed on 24 February 2005). However, in accordance with French law, the intensive care unit (ICU) physician informed all patients or their families about the study. It involved ICUs and emergency departments from 14 university hospitals located throughout France (three centers in Paris, two in Lyon, and one each in Marseille, Nantes, Lille, Grenoble, Besançon, Nimes, Poitiers, Limoges, and Dijon).

As previously described, consecutive patients were recruited between December 2004 and March 2007 if they were at least 18 years old and had a severe blunt trauma defined as trauma requiring admission into an ICU within 72 hours after injury or, in the case of early death before ICU admission, trauma managed by a mobile ICU (MICU). Exclusion criteria were (a) penetrating traumas and (b) deaths occurring before the implementation of any advanced life-sustaining treatment. A total of 3,205 patients were eligible for inclusion in the FIRST study. Patients with incomplete or poor-quality data regarding hospital of first admission, ISS, pre-hospital management, or vital status were secondarily excluded (n = 502), leading to a FIRST study sample of 2,703 patients.

For the purpose of the present analysis, further exclusion criteria were retained (Figure 1). First, 651 patients initially admitted in non-university hospitals before their admission in a university hospital ICU were discarded because information about CT could not be reliably collected in these patients. To guarantee that all patients had had a chance to undergo either whole-body or selective CT of one or more body regions, 21 patients who died within the first three hours after the accident were excluded (five patients with whole-body CT, three patients with selective CT, and 13 patients without any CT). Furthermore, 81 patients who did not receive any CT were excluded in order to limit indication bias. Thus, 1,950 patients were retained in the present observational analysis and were divided into two groups: those who benefited from a whole-body CT and those who benefited from a selective CT according to the diagnosis strategy defined by each trauma team.

ICU physicians collected data from the medical records of MICUs, emergency departments, and ICUs. In each center, ICU physicians aided by local research assistants entered data into the FIRST database that is hosted by the Clinical Investigation Center in Dijon. The Clinical Investigation Center was responsible for logistic coordination of the study, data quality control, and statistical analysis.

ICU physicians collected (a) patients' characteristics; (b) data about accident circumstances; (c) hospital units involved in the early care of patients before admission to the ICU; (d) clinical and biological data on the pre-hospital phase if available, upon hospital admission, and 24 and 72 hours after the trauma; and (e) a summary of clinical variables at patient discharge or death.

During the pre-hospital phase, the following data were recorded: initial physiological variables (arterial pressure, respiratory rate, and oxygen saturation as measured by pulse oximetry), pupil status, Glasgow Coma Scale [GCS] score, and life-sustaining treatments (venous line, fluid loading and catecholamine administration, tracheal intubation, ventilation, blood products, and chest tube).

Information on physiological variables and life-sustaining treatments was also collected upon arrival at the first hospital and 24 and 72 hours after the accident. The first available measurement, either at the pre-hospital phase or upon hospital admission, was used to describe the initial physiological status of the patient. At patient discharge from the ICU or death (within 30 days), anatomic injury diagnoses with corresponding Abbreviated Injury Scale (AIS) codes and the ISS were recorded from medical records. The AIS was coded according to the 1998 updated classification 10 by local research assistants using medical, radiological, and surgical reports. Local ICU physicians reviewed all problematic cases.

Several variables regarding the accident circumstances or the initial medical trauma assessment of the patient were important to take into consideration because they could influence imaging strategy. Two variables dealing with either the severity of accident or the suspected severity of trauma were constructed 1112. The accident was considered potentially severe if, in the case of a road traffic accident, at least one of the following points was recorded: pedestrian, no safety equipment (air bag, seat belt, crash helmet, and so on), excessive speed, victim ejected/crushed/burned/cut free from the vehicle, death of other victims in the vehicle, and vehicle fall of more than 6 meters. For the other accidents, severity was defined as a fall of more than 6 meters or crushing by lifting or agricultural equipment. Trauma was defined as potentially serious if, at the initial medical examination, there was suspicion of fractured skull, fractured pelvis, flail chest, or spinal injury or the presence of limb amputation, severe burns, or smoke inhalation. Because pre-hospital and hospital management may depend on the accident time during on-call periods, we defined two variables related to accident time: daytime (from 8:30 a.m. until 6:30 p.m.) versus night-time and weekend (from Saturday 1 p.m. until Monday 8 a.m.) versus other days.

All surgical procedures received by patients until ICU discharge were recorded and coded by physicians at the coordination center. Hemostatic procedures included arterial embolization and hemostatic thoracotomy or abdominal laparotomy. Orthopedic procedures included all types of bone fixation of upper and lower limbs.

Whole-body CT was an unenhanced CT of the head followed by contrast-enhanced CT of the chest, abdomen, pelvis, and complete spine. Information about whole-body and selective CT to one or more of these body regions was recorded in the first clinical department where the patient was admitted (emergency, surgical, or radiology unit or ICU) and, if needed, in subsequent departments that received the patient until his or her admission to an ICU. All other imaging procedures were recorded in a similar way. The main outcome measurement was the vital status at 30 days or at ICU discharge if discharge occurred within the first 30 days.

Comparisons of patients given a whole-body CT with those given a selective CT were performed by using chi-squared tests or, if needed, Fisher exact tests. To address selection and confounding biases that could not be totally controlled by the exclusion of patients who did not have any kind of CT and to assess the mortality reduction risk by using the initial whole-body CT, we constructed a propensity score. This approach is based on the idea that the probability of undergoing either whole-body or selective CT may depend on the patient's age, sex, study center, accident circumstances, initial medical assessment, and physiological status as well as on the administration of life-sustaining treatments during the pre-hospital phase or at hospital admission. We computed a non-parsimonious logistic regression model that included 24 potentially relevant covariates regarding the use of either whole-body or selective CT (variables listed in Table 1) 13. The predicted probability that was derived from the logistic equation defined the propensity score for each patient. The discriminative power of the propensity score was quantified by the c statistic corresponding to the area under the receiver operating characteristic (ROC) curve. The quality of the propensity score was confirmed by checking the balance of covariates among patients with whole-body CT and among those with selective CT after adjustment for the propensity score.

The impact of whole-body CT on mortality was assessed by using several multivariable logistic models. First, we used a classic model in which the CT variable and all covariates (baseline characteristics and post-CT characteristics related to medical treatment in the first 24 hours) associated with mortality at a significance level of less than 0.20 in bivariate analysis were introduced and selected through a backward procedure as described by Hosmer and Lemeshow 14. Second, the propensity score was used in two different ways, either for regression adjustment or for matching 1516. The propensity score (either as a continuous variable or categorized according to quintiles) replaced baseline characteristics in the logistic regression. The same set of variables that related to post-CT treatments and that was used in the first multivariate logistic model was used as covariates. We also used propensity-based matching to produce adjusted estimates of the effect of whole-body CT on mortality. We performed a five-digit case control match on propensity score by using SAS™ version 9.1 (SAS Institute Inc., Cary, NC, USA). Each patient who had a whole-body CT was matched to one sole patient who had only a selective CT on five digits, then on four digits, and (if needed) on three, two, and one digit of the propensity score (the matching became rougher and rougher). The quality of matching was assessed by comparing baseline characteristics between both CT groups by using the chi-squared test or, if needed, the Fisher exact test. A logistic regression model adjusted for covariates related to post-CT treatments was also used for assessing the impact of whole-body scan on mortality. The goodness-of-fit of the various logistic regression models was assessed according to Akaike information criteria and the Hosmer-Lemeshow test.

To compare our results with those obtained in a previous study 5, we also used a TRISS-adjusted approach. The TRISS method is used to predict the probability of survival at discharge 17. There were large differences in severity mix between our observed data set and the US prediction data set as reflected by the M statistic (M = 0.42 in the whole-body CT group and 0.51 in the selective CT group). Thus, to compare the observed survival with the TRISS-predicted survival, we calculated only the Ws score and corresponding Zs statistic 18. These scores are produced by a method of direct standardization of the difference between the observed number of survivors and the TRISS-predicted number of survivors according to the case mix of injury severity of the US prediction database. A positive value of Ws associated with a Zs of greater than 1.96 indicates a significantly better survival than that defined by the prediction database.

Data were expressed as mean with standard deviation, median with interquartile (25th to 75th) range, or percentage. We performed the statistical analyses by using SAS™ version 9.1 (SAS Institute Inc.) and STATA version 11 (StataCorp LP, College Station, TX, USA) software. A P value of less than 0.05 was considered significant, and all P values were two-tailed.

Among the 1,950 patients who had severe blunt trauma and who had a CT examination, 1,696 (87%) had a whole-body CT and 254 (13%) had a selective CT. Among patients with selective CT, the body regions were head in 202 patients (80%), abdomen/pelvis in 105 (41%), cervical spine in 89 (35%), and thorax in 59 (23%). The proportions of patients who received Focussed Assessment Sonography for Trauma (FAST) imaging (abdominal ultrasonography and chest radiography) were not significantly different in patients with selective CT and in those with whole-body CT (19.2% versus 23.4%, respectively; P = 0.15). There were significant differences in baseline characteristics between the two groups (Table 1). Patients with whole-body CT were significantly younger and were victims of more potentially severe accidents and more specifically of road traffic accidents. These patients were more often managed by physicians from MICUs in the pre-hospital phase and were more rapidly admitted to university hospital trauma centers. Although the initial physiological status (except for heart rate) and GCS score were similar in the two groups, patients with whole-body CT had benefited from a more aggressive management (fluid loading, intubation, and catecholamine administration) in the pre-hospital phase or at admission. The use of whole-body versus selective CT also significantly depended on the study center (Table 1).

The propensity score constructed from 24 baseline characteristics fitted well with data as indicated by the good area under the ROC curve (c index of 0.83). Furthermore, all baseline characteristics that were significantly related to the use of whole-body versus selective CT in the univariate analysis were no longer significant after adjustment for the propensity score (Table 1).

At day 30, 277 patients (16%) in the whole-body CT group and 56 (22%) in the selective CT group were deceased (absolute decrease of 6%, 95% confidence interval (CI) of 1% to 11%; P = 0.02). There was no significant difference between the two groups in the mortality rate at 24 hours (6.0% in the former group versus 8.3% in the latter). The adjusted impact of whole-body CT on 30-day mortality according to the various adjustment methods is summarized in Figure 2. Among the 1,607 patients for whom all variables were available, all methods (adjustment for pre-CT + post-CT covariates, propensity score + post-CT covariates, and propensity-based matching + post-CT covariates) led to a significant reduction of 30-day mortality in the whole-body CT group (Figure 2). The best model was the classic logistic regression model based on adjustment on pre- and post-CT covariates, as indicated by the area under the ROC curve (c statistics = 0.89) and the Hosmer-Lemeshow test (P = 0.54).

Cerebral death was the main cause of death and was significantly less frequent in the whole-body CT group than in the selective CT group (62% versus 79%, P = 0.016). The other main causes of death - hemorrhagic shock (14% versus 13%) and multivisceral organ failure (8% versus 2%) - were similar in the two groups.

The TRISS method was applied to 1,864 patients with an available revised trauma score. The TRISS-predicted mortality rates were 30% for the 1,635 patients who received whole-body CT and 22.7% for the 229 patients who received only selective CT. After standardization according to the case mix of injury severity of the US prediction database, the Ws and Zs scores were, respectively, 3.3 (95% CI 1.9 to 4.6) and 4.81 (P < 0.0001) among whole-body CT patients, indicating a significantly better survival in these patients than that predicted by the TRISS method. Corresponding Ws and Zs scores among selective CT patients were, respectively, 0.3 (95% CI -2.9 to 3.5) and 0.17 (P = 0.44), indicating the lack of significant difference between the observed and the TRISS-predicted survival in the latter group.

Compared with patients with selective CT, the proportion of whole-body CT patients with an AIS score of at least 4 was significantly higher for the thorax (P < 0.001) and spinal areas (P = 0.05) as well as for the lower limbs (P < 0.001) (Table 2). Similarly, the proportion of patients with an ISS of at least 35 was significantly higher for patients given whole-body CT (25.8%) than for those given selective CT (7.5%). The introduction of ISS as covariate in regression models shown in Figure 2 led to an increase in the mortality benefit associated with the use of whole-body CT. After adjustment for ISS, the odds ratio (95% CI) decreased from 0.58 (0.34 to 0.99) to 0.45 (0.26 to 0.77) in the classic logistic regression model and from 0.55 (0.35 to 0.86) to 0.42 (0.27 to 0.67) in the propensity logistic regression model.

As shown in Table 3, the proportion of patients undergoing surgery was significantly higher among the whole-body CT group than in the selective CT group both within the first 24 hours (67% versus 59%, P < 0.02) and until ICU discharge (73% versus 64%, P < 0.001). The main differences between groups in surgical procedures used within the first 24 hours were the higher percentage of patients with hemostatic or orthopedic surgery in the whole-body CT group than in the selective CT group (P = 0.03 and P < 0.001, respectively). In contrast, more patients had undergone intracranial surgery at 24 hours in the selective CT group than in the whole-body CT group (P < 0.001). Furthermore, early medical management, as reflected by the use of intubation, blood transfusion, and catecholamines, was significantly more aggressive in patients given whole-body CT. A similar pattern was observed in surgical procedures until discharge, with significantly more hemostatic (P = 0.02) and orthopedic (P < 0.001) surgery and less intracranial surgery (P < 0.001) for patients given whole-body CT than in those given selective CT. Furthermore, spinal surgery (P = 0.01) and the use of a thoracic drain (P < 0.001) were significantly more frequent in the whole-body CT group.