Pelvic fractures are associated with substantial morbidity and mortality1-14, and the optimal initial treatment remains controversial2,5,15-19. Although mortality rates among patients with pelvic fractures have been reported1-14, the relative risk associated with a pelvic ring disruption compared with other injuries has been studied less extensively3,6. Patient age, Glasgow Coma Scale score, and hemodynamic shock are known predictors of mortality following trauma5,7-10,14,19-25. It is unclear whether the impact of a pelvic fracture varies when combined with these risk factors.
The purposes of this study were (1) to compare the mortality risk posed by a pelvic fracture with the risk conferred by other injuries, and (2) to determine if the association of a pelvic fracture with mortality varies when combined with other known risk factors for mortality.
Trauma registry records from two level-I trauma centers were retrospectively examined. The institutional review boards at both centers approved this study. Information on 67,826 patients was screened. Patients who had a pelvic fracture were identified with use of the International Classification of Diseases, Ninth Revision (ICD-9) codes 806.6, 806.7, 808.2, 808.3, 808.4, and 808.5. This includes all open and closed fractures of the sacrum, ischium, ilium, pubis, and pelvic ring. We did not attempt to classify the pelvic fractures, nor did we screen our data to include only severe injuries. Acetabular fractures were not included unless the patient had an associated pelvic fracture. Patients with a penetrating mechanism of injury were excluded.
Center A provided records for 41,270 patients collected from October 1995 through January 2006. The majority of patients were greater than thirteen years of age (>99% of patients). Center B provided information on 21,763 patients collected from January 1993 through December 2003. Center B contained relatively more pediatric patients although the majority were still greater than thirteen years of age (>95% of patients). Both trauma centers are regional referral centers for patients with pelvic fractures. The average Injury Severity Score for patients from Centers A and B was 8 and 10, respectively.
Logistic regression analysis was done on 63,033 patients with complete data. A total of 3296 (5%) had pelvic fractures. The analysis was to determine the odds ratio of mortality during initial hospitalization after trauma for pelvic fracture compared with other variables, thereby giving an estimate of relative risk (Table I). A multivariate logistic regression analysis was initially performed with use of all possible independent variables. The variables not considered significantly predictive (p < 0.05) were dropped from the model, and only significant variables were retained. Variables included in the final model were pelvic fracture, an age of sixty years or more, systolic blood pressure (on admission) of =90 mm Hg, a Glasgow Coma Scale score26 of =8, and an Abbreviated Injury Scale (AIS) score27 of =1 for the head and neck, =1 for the chest, =1 for the abdomen, and =1 for the extremity. At each institution, experienced trauma registry nurses did the coding. Because of minor differences in coding practices at the two centers, the two data sets underwent identical side-by-side analyses instead of being combined into one larger data sample. These differences included patterns of coding for severity of face injury (face AIS) and skin injury (skin AIS), resulting in incomplete data for these two variables.
Additional analyses were done to determine whether any association of pelvic fracture with mortality varied when combined with other variables known to be strongly predictive of mortality: age, Glasgow Coma Scale score, and systolic blood pressure. Patients were categorized as elderly (an age of sixty years or more) or not; in shock on arrival (systolic blood pressure of =90 mm Hg) or not; and severely head injured (Glasgow Coma Scale score of =8) or not. This created eight patient groups, matched with regard to age, blood pressure, and Glasgow Coma Scale score. These groups were subdivided according to the presence or absence of a pelvic fracture (Tables II and III). A univariate chi-square statistical analysis was then performed to compare the mortality rate of patients with pelvic fractures with the mortality rate in matched patients without pelvic fractures. Chi-square p values were used in all cases except when any cell had an expected count of less than five. In those instances, the Fisher exact test p value was used.
Management of pelvic fractures at both centers was similar. Both centers used treatment algorithms that emphasized early recognition of markers for mortality, such as hypotension, transfusion requirements, age, and fracture pattern. Both centers focused on aggressive resuscitation with blood and fresh-frozen plasma, both employed early provisional stabilization with pelvic binders or sheets, and both used early pelvic arteriography. One difference in management was the use of pelvic packing at Center B19,28,29, a strategy that had evolved over the previous five years. This difference affected a small group of patients who were otherwise being treated in a similar manner.
Source of Funding
This research was supported by the Suzanne and Aaron A. Hoffman, MD, Orthopaedic Research Fund. The authors received nothing of value in exchange for their work on this project. The funding source played no role in this investigation.
Centers A and B had 2102 and 1194 pelvic fracture patients, respectively. The mean Injury Severity Score30 for patients with pelvic fractures at Centers A and B was 17 and 21 points, respectively. At Center A, the mortality rate was 8.4% for patients with pelvic fractures compared with 2.4% for those without pelvic fractures. At Center B, the values were 13.6% and 4.5%, respectively. At both centers, when the trauma patient sample was analyzed as a whole, the presence of a pelvic fracture was significantly associated with mortality (p < 0.001).
The odds ratio for mortality associated with a pelvic fracture was 2.4 at Center A and 2.0 at Center B. The odds ratios of the other variables are shown in Table I and are illustrated graphically in Figure 1. At Center A, shock was most predictive of mortality (odds ratio, 9.2), followed by head and neck AIS score (odds ratio, 8.0), severe head injury (odds ratio, 7.7), and an age of sixty years or more (odds ratio, 4.5). At Center B, severe head injury (odds ratio, 24.4) was found to be most strongly associated with mortality, followed by shock (odds ratio, 16.3), an age of sixty years or more (odds ratio, 8.8), and head and neck AIS score (odds ratio, 3.0). At both centers, the odds ratio of mortality associated with a pelvic fracture (approximately 2) was roughly equivalent to that posed by an abdominal injury. All variables except extremity AIS score were associated with varying degrees of an increased risk of death.
A receiver-operating characteristic curve of our model, with use of the variables included for predicting mortality, was 0.92 for Center A and 0.97 for Center B (Fig. 2), demonstrating a good fit between our model and the data.
Examination of the effect of pelvic fracture on mortality when combined with other variables showed that pelvic fracture was significantly associated with mortality in each of the eight groups tested, except in patients who had severe head injury combined with shock on arrival. In this subset, for both young and old patients at both centers, a pelvic fracture was not associated with mortality. These data are presented in Table II and Table III. The single difference between the two centers occurred in the group who were elderly, and in shock, but did not have a severe head injury. In this group, pelvic fracture was not associated with increased mortality at Center A, but it was at Center B.
The 4727 patients with missing data were excluded from the analysis because of missing information on age (459 patients) and systolic blood pressure (4268 patients). The average mortality rate for this group overall was 17% for patients from Center A and 8% for patients from Center B.
Numerous studies have examined mortality after pelvic fracture, and the mortality rates have ranged from 7.6% to 19%1-14. In our study, the mortality rates of 8.4% and 13.6% at Centers A and B, respectively, fall within this range. The orthopaedic literature contains numerous reports that describe strategies aimed at decreasing the risk of death after pelvic fracture2,5,15-19. We have conducted these sorts of investigations ourselves14,19,28,29,31,32.
Strategies that decrease the risk of death after pelvic fracture are difficult to evaluate. Pelvic fracture is one variable among many that carry a risk of death, and it is difficult to prove definitively that treatment aimed at pelvic injury is truly effective at saving lives. The interplay of important variables, such as age, hemodynamic state, and head injury, confound investigators' ability to set up comparable study groups.
Studies have compared the mortality risk posed by pelvic fracture with risks carried by other injuries3,6. Chong et al.3 performed a retrospective review of 343 patients with pelvic fractures and reported a mortality rate of 10.5%. In comparison, a cohort of patients without pelvic fractures treated at the same center from the same time period had a lower mortality rate. A 1.4-fold increase was found in the pelvic fracture group. Similarly, in a large review of trauma patients in the United Kingdom, Giannoudis et al.6 reported a mortality rate of 14.2% among 1586 patients with pelvic injuries compared with a rate of 5.6% in 7465 patients without pelvic fractures. However, to our knowledge, no previous study has stratified the risk of death posed by multiple variables, and no study has assessed the way that the mortality risk associated with a pelvic fracture might be altered in the presence of other risk factors. We explored the effect of multiple variables, as we thought such information would be useful to clinicians who manage multiply injured trauma patients and to those planning clinical research.
This study is a retrospective analysis of data from two level-I trauma centers, and it is subject to the limitations of all retrospective studies. A certain degree of error is inherent any time a large database is used. We had 4720 patients with missing data who were excluded from the analysis; inclusion of these patients might have altered our findings. The data set utilized did not include fracture pattern33, cause of death, or treatment received. Although we showed that a pelvic fracture is associated with mortality, we were not able to clarify the physiologic mechanisms involved.
Our initial analysis of the pelvic fracture and mortality included age, systolic blood pressure, Glasgow Coma Scale score, and the AIS score for the head and neck, chest, abdomen, and extremities. We selected these variables because other studies had led us to expect that they would influence mortality. One possible conclusion from our study is that pelvic fracture is simply a marker for another component of injury that is truly responsible for increased mortality in a certain portion of the patients, but which was not considered as a variable in our logistic regression model. The putative variable was excluded either because we failed to recognize it in developing our model or because the data set was not sufficiently robust. We attempted to lessen this possibility through testing how well the variables we chose were associated with mortality through the use of receiver-operating characteristic curves. Receiver-operating characteristic curves are graphical plots of the sensitivity compared with 1.0 minus the specificity for a binary classifier system as its discrimination threshold is varied. They are commonly used in medicine to assess the ability of a diagnostic test to predict a disease state. Using receiver-operating characteristic curves, researchers can assess how well some predictor—such as a blood test or questionnaire—is associated with an outcome. Receiver-operating characteristic curve analyses yield results ranging from 0 to 1.0. A result of 1.0 indicates perfect ability of the test to predict the disease state, while a result of 0.5 indicates the test is no better than 50-50, essentially the same as flipping a coin.
In our study, we wanted to see how well the variables we selected for inclusion in our logistic regression were associated with mortality. We constructed a model using the variables of pelvic fracture, an age of sixty years or more, systolic blood pressure of =90 mm Hg, Glasgow Coma Scale score of =8, head and neck AIS score of =1, chest AIS score of =1, abdomen AIS score of =1, and extremity AIS score of =1, and we tested that model as a predictor of death using receiver-operating characteristic curves. The results of the receiver-operating characteristic curve analyses were 0.92 for center A and 0.97 for Center B, indicating that the variables we selected are good predictors of mortality. The addition of some other variable might incrementally improve the ability of our model to predict death, but substantial improvement is unlikely.
The literature is conflicted regarding the etiology of death among patients with pelvic fractures. Many studies have suggested that pelvic fractures are infrequently the direct cause of death, with patients more often dying from associated injuries and associated complications4,11,12,34,35. Others have implicated hemorrhage from the pelvic injury as the direct cause of death13,19,36,37. Some studies have suggested that fracture pattern is predictive of mortality2,3,17,20,31,37,38, while others have been unable to demonstrate such an association7,9-11,14,19,39.
It is perhaps naïve to believe that any database analysis can identify the true source of lethal hemorrhage in a population of multiply injured patients, some of whom have an associated pelvic fracture. Nonetheless, the value in studying the association of pelvic fracture with mortality lies in improving the ability of clinicians to direct intervention in such a way as to mitigate the risk of death. As the extant literature shows, such studies often produce conflicting results. The goal of our study was to address the potential for selection bias in the study of pelvic fracture and death.
One potential weakness of the study is that alterations in mortality rates over time might have made earlier or later years in the study less comparable and negatively influenced the outcome of our analysis. However, when we reviewed our yearly data, we found that mortality varied little from year to year, with no apparent trends, both for patients with and without pelvic fractures.
Another potential weakness of this study is the loss of several thousand patients who had incomplete data. We carried out a sensitivity analysis to determine the degree to which our data may have been affected by the missing information, had it been present. In this technique, missing values from the each data set were imputed with use of the SAS Multiple Imputation procedure (SAS Institute, Cary, North Carolina). Tables IV and V show the odds ratio estimates for the original analysis model with missing values and for the model with imputed values. In the Center A data set, the odds ratio estimates are of similar relative magnitude to those found in the initial analysis but are somewhat different. In the Center B data set, the odds ratio estimates are all very close to those found in the initial analysis. In no instance did we see a change in direction of the odds ratio, from odds of >1.0 to odds of <1.0, which would indicate a major problem with missing values.
One strength of our study is that our data sample was large and included many patients with pelvic fractures. The results of the analyses from the two trauma centers were remarkably similar. Logistic regression analysis showed that shock, a Glasgow Coma Scale score of =8, a head-neck AIS of =1, and an age of sixty years or more were the top four variables associated with mortality at each center. The precise odds ratios differed, but the persistent surfacing of these parameters underscores their importance. It is also noteworthy that our analysis of how the mortality risk of pelvic fractures changed when combined with other risk factors yielded similar results at each institution. This persistent finding, from two large patient samples, seems to strengthen the soundness of the conclusion. The one area of disagreement in this analysis arose in testing elderly patients who were in shock but did not have a head injury. Center A showed no increased risk with a pelvic fracture, whereas Center B showed a strong association (odds ratio, 6.8). It seems likely that Center B represents the true situation in this case because the small number of expected counts for this particular group in Center A may indicate that the sample was too small to get an accurate odds ratio for this variable. However, this result must be interpreted cautiously.
Our usage of a second, univariate chi-square analysis, after the logistic regression, requires some discussion. We carried out this second analysis because we hoped to get more information from our data than odds ratios could provide. Odds ratios are valuable in that they provide a measure of association between the variables under study. In our case, we were able to stratify the variables under study, and we determined which had the strongest association with mortality. However, as clinicians, what we needed most was an assessment of the effect of pelvic fracture when combined with other injuries. For this reason, we did a secondary analysis using some of the same variables—age, Glasgow Coma Scale score, and systolic blood pressure. For the most part, the two analyses were in agreement: pelvic fracture was associated with mortality. The area of disagreement in the setting of severe head injury and shock revealed an interesting finding that we hope will spur future research.
In conclusion, pelvic fracture is significantly associated with mortality. However, pelvic fracture is one variable among many that affect the risk of death and must be considered in relation to other such variables. Pelvic fracture appears to carry a mortality risk similar to that posed by abdominal injury. The apparent association between pelvic fracture and mortality lessens in the setting of severe head injury and shock. Presumably, for patients with severe head injury and shock, the mortality risk is so high that the pelvic fracture adds little to the overall risk. This does not mean pelvic fractures can be ignored in such patients. However, in our opinion, it seems advisable to exclude patients with severe head injury and shock from analyses of the effectiveness of treatment strategies that seek to lessen mortality after pelvic fracture, since it appears that pelvic fracture has little impact on the risk of death for such patients. 