A histogram of pelvic hemorrhage volumes measured from computed tomographic scans, with mean volume and the number and percentage of subjects with pelvic hemorrhage volumes more than 200 mL and more than 500 mL.
A receiver operator characteristic curve of the volume of pelvic-fracture–related hemorrhage as a predictor of arterial injury demonstrated by angiography. The area under the receiver operator characteristic curve is 0.804.
Blackmore CC, Jurkovich GJ, Linnau KF, Cummings P, Hoffer EK, Rivara FP. Assessment of Volume of Hemorrhage and Outcome From Pelvic Fracture. Arch Surg. 2003;138(5):504-509. doi:10.1001/archsurg.138.5.504
Measurement of pelvic hemorrhage on computed tomographic (CT) scans can estimate the pelvic fracture component of total patient blood loss and predict the need for angiography.
Retrospective cohort study.
Large level 1 trauma center.
We examined data from 759 consecutive, nonreferral blunt trauma patients who sustained pelvic fracture.
Main Outcome Measures
Pelvic-fracture–specific outcomes included estimation of extraperitoneal pelvic hemorrhage volume from emergency department CT scans and determination of arterial injury from angiograms. General patient outcomes determined from medical record review included transfusion requirement, estimated blood loss, and mortality. Subanalysis was performed on subjects with only pelvic fracture as a source of major hemorrhage (derived from discharge International Classification of Diseases, Ninth Revision, Clinical Modification codes).
Overall mortality was 96 (13%) of 759 patients. Blood transfusion was given to 418 (55%) patients, and 258 (34%) received 6 or more units in the first 72 hours. Pelvic-fracture–related hemorrhage averaged 149 mL (range, 0-1423 mL). Angiography was performed on 163 patients, of whom 113 had arterial injury. Higher pelvic hemorrhage volumes on CT scans were seen in subjects with pelvic arterial injury demonstrated on angiograms (P<.001). In subjects without another source of major hemorrhage, pelvic CT hemorrhage volumes were strongly associated with transfusion requirement (P<.001). Subjects with large pelvic hemorrhage volumes (>500 mL) were more likely to have pelvic arterial injury (risk ratio, 4.8; 95% confidence interval, 3.0-7.8; P<.001) and require large-volume (≥6 U) transfusions (risk ratio, 4.7; 95% confidence interval, 1.8-12.3; P<.001) than patients with smaller pelvic hemorrhage volumes.
Pelvic hemorrhage volumes derived from pelvic CT scans were predictors of the need for pelvic arteriography and transfusions.
HEMORRHAGE IN patients with traumatic pelvic fracture remains a diagnostic and treatment challenge. Up to 52% of patients with pelvic fracture will develop hemorrhagic shock in the emergency department1 and 8% to 30% will die in the hospital.1- 5 Several investigations have tried to define clinical predictors to determine which patients with pelvic fracture are at high risk for major hemorrhage and thus might benefit from early intervention, such as angiography and embolization.2,6- 8 However, studies have been limited by the fact that mortality and blood loss may not be directly or exclusively related to the pelvic fracture.1,5,9 Patients who sustain pelvic fractures are generally victims of high-energy trauma and have multiple injuries involving different organ systems. To guide interventions such as angiography and embolization, predictors of hemorrhage in patients with pelvic fracture should specifically predict hemorrhage related to the pelvic vasculature. We hypothesized that computed tomographic (CT) evaluation of pelvic hematoma in patients with pelvic fracture could help predict fracture-related blood loss and the need for angiography and embolization.
The objectives of this study were to estimate the volume of pelvic hemorrhage that occurred in patients with pelvic fracture, and to determine the relationship between pelvic-fracture–related hemorrhage and patient outcomes, including death, transfusion requirement, and pelvic arterial injury.
The methods and procedures followed in this study were approved by the University of Washington, Seattle, Human Subjects Review Board.
We performed a retrospective cohort study at a single large level 1 urban trauma center. The series consisted of all blunt trauma patients who sustained pelvic fracture between January 1, 1997, and April 30, 2001. Subjects who were evaluated at another institution and transferred were excluded. Also excluded were patients who died prior to imaging with either radiography or CT scans.
For each patient, 1 of 3 members of the research team reviewed the medical record to determine the mechanism of injury and physiologic status while in the emergency department, including lowest blood pressure level and highest pulse rate. A different investigator determined clinical outcome, including mortality and transfusion requirement. For 10% of medical records, an additional reviewer repeated the chart abstraction to allow assessment of interobserver agreement. All chart abstracters were blinded to hemorrhage volumes. Interobserver and intraobserver agreement among the 3 reviewers was assessed using the κ statistic. All data were double-key entered.
For each patient who underwent CT scanning of the pelvis within 48 hours of presentation, a single board-certified emergency radiologist (C.C.B.) measured the volume of extraperitoneal blood in the pelvis. An electronic cursor was used to trace areas of hemorrhage on CT scans, and hemorrhage areas were calculated using ImageJ software (National Institutes of Health, Bethesda, Md). Image areas were multiplied by the CT scan slice thickness to determine hemorrhage volumes in each CT slice, and slice volumes were summed to determine the total volume of the pelvic-fracture–related hemorrhage. Subjects with extraperitoneal bladder ruptures were excluded from CT hemorrhage volume measurement. The reliability of CT volume measurements was determined through double measurement in 10% of subjects and calculation of the intraclass correlation coefficient.
A single board-certified radiologist (E.K.H.) with a certificate of added qualification in interventional radiology reviewed all the angiograms. The interventional radiologist was blinded to clinical presentation, other outcomes, and hemorrhage volumes. Angiograms were considered abnormal if arterial extravasation, occlusion, or pseudoaneurysm was identified.
For analysis, subjects were divided into 2 groups. Group A included subjects whose only known source of major hemorrhage was the pelvic fracture. Group B included subjects who had pelvic fractures as well as other sources of major hemorrhage. Group determination was made through review of all discharge International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes using the Harborview Assessment for Risk of Mortality score.10 Anatomic diagnoses in the Harborview Assessment for Risk of Mortality system associated with major hemorrhage were selected by consensus of the investigators. Subjects who had any non-pelvic–fracture anatomic diagnoses determined a priori by the investigators to result in major hemorrhage were included in group B. All subjects initially thought to be eligible for group A but who required 6 or more units of blood transfusion underwent a second medical record review to confirm group A eligibility.
Primary data analysis consisted of comparing pelvic-fracture–related hemorrhage volumes in subjects with identified arterial injury with volumes in subjects with no identified arterial injury. This primary analysis included all subjects in both groups A and B. Additional analysis in group A subjects included comparison of pelvic-fracture–related hemorrhage volumes with blood transfusion requirements and mortality. Linear regression was used for the continuous outcomes of hemorrhage volume and transfusion requirement, and logistic regression was used for dichotomous outcomes. Receiver operator characteristic curves were estimated from logistic regression analyses. We also performed cross-tabulation using the χ2 statistic to determine association between hemorrhage volume and clinical outcomes. Analyses were performed using Stata statistical software, version 6.0 (Stata Corp, College Station, Tex).
Medical record review for emergency department presentation and outcome was completed for all 759 eligible patients. Interobserver agreement on chart-abstracted emergency department presentation and clinical outcome data was excellent (mean κ = 0.90).11 The CT scans were performed within the first 48 hours, and hemorrhage volumes were measured in 592 (78%) of 759 subjects. Seventeen subjects were excluded because of extraperitoneal bladder rupture, which compromised the measurement technique. Intraobserver agreement for the CT measurements was good, with an intraclass coefficient of 0.90 and a mean intraobserver difference of 34 mL.
The subjects in this study were predominantly men and had a mean age of 37 years. The most common mechanism of injury was motor vehicle crash, followed by pedestrians being hit by automobiles and falls (Table 1). Approximately 30% of patients demonstrated evidence of hemodynamic instability or shock, as defined by a systolic blood pressure level of less than 90 mm Hg (203 [27%] of 759) or a pulse rate of more than 130/min (238 [31%] of 759) at some point during treatment in the emergency department. In addition, 152 (20%) of 759 patients received 4 or more units of blood transfusion during their emergency department care.
The overall mortality rate was 13% (96/759). Fifteen percent (113/759) had arterial injury in the pelvis demonstrated on angiograms, and 34% (258/759) received 6 or more units of blood transfusion in the first 72 hours (Table 2).
The volume of pelvic-fracture–related hemorrhage measured from CT scans varied from 0 to 1423 mL (mean, 149 mL). The distribution was highly skewed; 74% (441/592) of subjects had less than 200 mL of pelvic-fracture– related hemorrhage, and 5% (29/592) had more than 500 mL of pelvic-fracture–related hemorrhage (Figure 1).
There were 294 patients with only pelvic fracture as a major source of hemorrhage (group A) and 465 patients subjects with multiple potential sources of major hemorrhage (group B). Significant differences were noted in all major outcomes between groups A and B, with increased mortality, frequency of pelvic arterial injury, and blood transfusion requirement in group B (P<.001). Patients in group B were more severely injured (mean Injury Severity Scale scores: group A, 11 vs group B, 25) (P<.001). Group B subjects also had substantially higher Abbreviated Injury Scale (AIS) scores for the abdomen and head. However, AIS extremity scores were only slightly higher in group B, reflecting the influence of pelvic fracture on the AIS extremity score (Table 2).
For the primary outcome of pelvic arterial bleeding by angiogram, all patients (group A and B) were considered. The risk ratio for pelvic arterial injury was 4.8 in subjects with more than 500 mL of pelvic-fracture–related hemorrhage compared with subjects with less than 500 mL of pelvic-fracture–related hemorrhage (Table 3). Positive and negative predictive values (Table 4) provide an indication of the probability of arterial injury for various pelvic volume thresholds. For example, at a pelvic hemorrhage volume of more than 500 mL, the probability of pelvic arterial injury was 0.45. At a pelvic hemorrhage volume of less than 200 mL, the probability of pelvic arterial injury was 0.05. A receiver operator characteristic curve demonstrates the sensitivity and specificity of differing hemorrhage volumes for the prediction of arterial injury (Figure 2). For example, a sensitivity of 0.86 for pelvic arterial injury was achieved at a relatively low threshold of 100 mL. Specificity for pelvic arterial injury reached 0.99 at the high-volume threshold of 600 mL.
For blood transfusion requirement as an outcome, and for the combined analysis of any adverse outcome (death, transfusion requirement, or arterial injury on angiogram), only subjects from group A were considered because the transfusion requirement and mortality in group B subjects might not have been primarily related to the pelvic fracture. High hemorrhage volume (>500 mL) was associated with high transfusion requirement (risk ratio, 4.7) and with any adverse outcome (risk ratio, 7.0) (Table 3). Table 4 shows the probability of any adverse outcome at different pelvic hemorrhage volume thresholds.
In subjects without other source of major hemorrhage, we used linear regression analysis to estimate that there was an increase of 2.9 mL of blood transfused per single milliliter of additional hemorrhage identified in the extraperitoneal pelvis (95% confidence interval, 2.2-3.6) (P<.001). This relationship was not confounded by age, sex, mechanism of injury, or time to CT scan.
In this article, we report the strong association between pelvic hemorrhage volume and the clinically important outcomes of presence of pelvic arterial injury, blood transfusion requirement, and death. Most patients with pelvic fracture will undergo abdominopelvic CT scanning in the emergency department as part of the evaluation of blunt torso injury. The presence of high volumes of extraperitoneal blood in the pelvis on these CT scans should alert the trauma surgeon to the likelihood of arterial hemorrhage arising from pelvic vessels and guide the appropriate use of pelvic arteriography. A simple estimation of whether the amount of pelvic hemorrhage is less than 200 mL or more than 500 mL allows the trauma surgeon to differentiate between subjects who have 5% vs 45% probability of pelvic arterial injury.
Previous investigations have shown that the identification of arterial extravasation on CT scans is a strong predictor of arterial bleeding at subsequent angiography.12- 15 However, not all patients with arterial bleeding demonstrate active extravasation on CT. Demonstration of arterial extravasation on CT requires multislice CT scanning with an optimized technique in the arterial phase, which may not be possible in all trauma patients. In addition, arterial bleeding may be intermittent and may not occur in the time sequence of a CT scan. Our study demonstrates that estimation of the volume of pelvic hemorrhage is an additional tool for predicting arterial injury. In the absence of substantial extraperitoneal hematoma on CT scans, the probability of significant arterial hemorrhage was less than 5%, and evaluation of other potential sources of hemorrhage should be given priority. On the other hand, a high volume of pelvic hemorrhage carries high specificity and positive predictive value for pelvic arterial injury and may aid in selection of subjects for angiography.
In this study, by demonstrating the strong association between the volume of pelvic hemorrhage and eventual patient outcome, we also facilitate future research into the investigation of predictors of arterial hemorrhage. Past investigations that examined pelvic fracture patterns and attempted to assess whether pattern predicted outcome were limited by relying on outcomes such as blood transfusion requirement and mortality.2,4,6,7 However, because these outcomes may be related to other injuries sustained by major trauma patients,1,3,5 they may obscure important associations between fracture pattern and outcome. In this article, by showing the strong association between pelvic hemorrhage volume and clinical outcome, we enable the use of pelvic hemorrhage volume as an intermediate outcome for the study of predictors of pelvic arterial injury. The Evaluating Pelvic Hemorrhage Study currently underway will use pelvic hemorrhage volume as a surrogate outcome to develop and validate a clinical prediction rule that will rely on immediately clinically apparent factors, such as fracture pattern and emergency department presentation, to define clinical predictors that can be used to identify subjects at high risk of arterial bleeding.
This study is a retrospective single-institution investigation and carries the expected limitations. However, although patients were identified retrospectively, CT volume measurements and medical record review were performed in a blinded fashion. In addition, to limit referral bias, we excluded transfer patients. The subjects in our study are similar to other large published series of patients with pelvic fractures in terms of age, sex, mechanisms of injury, and mortality,2,4,5,16,17 suggesting that our patient population is representative of other trauma centers. Validation of hemorrhage volume as a predictor of arterial injury should be performed on a separate group of patients to confirm the relative risk estimates.18 In addition, negative and positive predictive values may be influenced by the prevalence of arterial injury in the study population and may differ in other populations.
A limitation of the study is the potential inaccuracy of the pelvic hemorrhage measurements. The CT measurement technique has been validated on solid abdominal organs.19,20 However, solid abdominal organs tend to be of a uniform shape, whereas clots may be irregular in contour and distort normal anatomic structures. Without a criterion standard, we cannot estimate the degree of inaccuracy in the CT measurements. However, the intraobserver, intraclass correlation was high. In addition, absolute accuracy of the measurement is not required. Measurements are only necessary to place patients within broad categories of hemorrhage volume to allow estimation of risk of arterial injury.
An additional limitation of this study is that the CT measurement technique is complex and can be time consuming. It is not appropriate to expect that such measurements will be made in the acute trauma setting. However, a crude estimation of whether the amount of pelvic hemorrhage is less than 200 mL or more than 500 mL may allow the trauma surgeon to broadly characterize the probability of pelvic arterial injury. Future investigations will consist of identifying whether there are specific locations within the pelvis that are more predictive of arterial vs venous injury.21 Because the courses of the relevant arteries are known, it is intuitive that clotting along the artery would be more predictive of arterial injury than generalized hemorrhage. This hypothesis is currently under investigation.
We found that there is an increase of 2.9 mL of blood transfused per additional milliliter of pelvic hemorrhage identified. This is not the one-to-one relationship that might be expected. Possible explanations include errors in the CT measurements, diffusion of hemorrhage in the interstices of the muscles and soft tissues, clot retraction, and failure of transfused blood to remain wholly within the intravascular space, indicating that transfusion requirements exceed blood loss. We cannot differentiate between these potential explanations. However, the stability of this estimate in multiple regression indicates that it is potentially useful as a way of estimating transfusion requirements in patients with pelvic-fracture–related hemorrhage.
In conclusion, the volume of extraperitoneal pelvic hemorrhage is a potentially important marker for the presence of pelvic arterial injury. In our study, subjects with low pelvic hemorrhage volumes (<200 mL) had only a 5% probability of pelvic arterial injury, whereas subjects with high volumes of pelvic hemorrhage (>500 mL) had a 45% probability of having angiographic evidence of pelvic arterial injury. Trauma surgeons can use pelvic hemorrhage volume as an additional tool to guide the often-challenging decision to select who should undergo pelvic arteriography.
Corresponding author and reprints: C. Craig Blackmore, MD, MPH, Box 359728, Harborview Medical Center, 325 Ninth Ave, Seattle, WA 98104 (e-mail: firstname.lastname@example.org).
Accepted for publication January 18, 2003.
This study was supported by grant K08-HS11291 from the Agency for Healthcare Research and Quality, Rockville, Md.
This study was presented at the 110th Scientific Session of the Western Surgical Association, Vancouver, British Columbia, November 20, 2002, and is published after peer review and revision. The discussions that follow this article are based on the originally submitted manuscript and not the revised manuscript.
Christine S. Cocanour, MD, Houston, Tex: Dr Blackmore and his colleagues have described a method for estimating the volume of pelvic hemorrhage from pelvic fracture and, most importantly, have correlated it with the need for transfusion as well as the need for arterial embolization. Pelvic fractures remain a significant problem for the trauma surgeon. They are often associated with other injuries, as exemplified by Dr Blackmore's group B, and with pelvic fractures, there is also the dilemma of whether the shock is due to the pelvic fracture hemorrhage or hemorrhage from another source. This can lead to not only delays in obtaining pelvic arteriogram but even unnecessary celiotomy. Most previous methods for quantifying hemorrhage from pelvic fractures have been based upon fracture pattern analysis; however, this does not always seem to hold true, especially in elderly patients.
The introduction of helical CT scanners has led to an increase in the use of CT for trauma. Because it can quickly evaluate the patient from head to toe, it has become the diagnostic tool of choice in many trauma centers, sometimes to the detriment of patients who would be better off in the operating room undergoing splenectomy. Dr Blackmore has given us another way of using the CT's information. I would like to ask several questions. Although in your manuscript you noted that the presence of contrast extravasation, or "blush" as we commonly know it, is not a reliable indicator for arterial hemorrhage, did its presence, in combination with an increased volume of pelvic hemorrhage, increase your accuracy in predicting the need for pelvic arteriography? Since the software that estimates the volume of pelvic hemorrhage takes more time to perform than what can realistically be expected to help in the real-time care of the trauma patient, are there other specific landmarks? Are there specific things to look for on the CT scan that the trauma surgeon can use to eyeball whether this is less than 200 mL or if it is greater than 500 mL of hemorrhage. Was there any correlation of the pelvic fracture pattern and the volume of pelvic hemorrhage? Was there any difference in the volume of pelvic hemorrhage with the fracture pattern in elderly patients, ie, would an older patient be more likely to have an increased amount of hemorrhage from the same fracture when compared with a younger patient?
Finally, did you notice any difference in the volume of pelvic hemorrhage when compared to the time of CT from injury, ie, if the CT scan was obtained 24 hours after injury or 12 hours after injury, was there a difference from someone who had had the CT scan obtained within an hour or two of injury?
Dr Jurkovich: Thank you for the opportunity to close this discussion. I want to begin by formally acknowledging the wonderful working relationship the trauma service has with the radiology department at Harborview. This is just one example of that cooperation. I am struck by the observation that during this meeting, and in particular during this morning's discussions, the important role of diagnostic and interventional radiology in our practice of medicine has been repeatedly emphasized. I am pleased to have a radiologist present at this meeting, particularly because there is such overlap in our efforts to care for the patient.
Dr Cocanour asked 5 specific questions that I will address, but I wanted to first say that the concept here, not to miss this, is that most patients are getting CT scans as part of their diagnostic workup. Most of the time that is done with IV contrast. What we are proposing is that an objective measure of the volume of pelvic hemorrhage can be seen in that CT scan, and that can help you predict if a patient is likely to have a positive angiogram. You can use that information in a couple of different ways. One is to triage the patient to the angiography suite as part of their initial diagnostic evaluation, or alternatively you can triage the patient to a higher level of trauma center care. Or you can estimate the volume of blood lost in the pelvis and predict transfusion needs. This technique is an effort to try to provide some objective measurement to this arena. This study is the first part of a 5-year, federally funded grant that Dr Blackmore is the principal investigator on, and we expect and hope to have more to report in later years.
First, regarding the question about extravasation and presence of a blush. Extravasation of contrast is likely to be a useful predictor, but unfortunately the specific hardware and software to really determine this wasn't available to us until about mid-1999, therefore excluding a large percentage of these patients from this analysis. We simply didn't have enough patients yet in the later years of the study to determine whether the presence of blush was an accurate predictor of hemorrhage.
Second, landmarks to estimate volume are of course an important kind of quick analysis and that will be our next part of the study—to look for specific landmarks or hematoma in anatomically specific areas on CT scan that correlate well with these volumes. It's an insightful comment, and it's our planned next stage of this study.
Third, regarding the issue of fracture pattern and the volume of hemorrhage, there have been many attempts to correlate fracture pattern with pelvic arterial injury. Although some fracture patterns more likely have a positive angiogram, there clearly are outliers, and the incidence of a positive angiogram with the most innocuous-appearing fractures still persists in about 5% of the patients. So, it won't be perfect, but we hope to give more information on fracture pattern and volume of hemorrhage as well.
Likewise, I cannot address the usefulness of this technique in elderly patients. We simply have not looked at that subgroup to analyze the difference in their patterns of fracture or their volume of hemorrhage.
Finally, your insightful question regarding the timing of CT scan is one that we have discussed as well. There is some correlation between positive CT, that is, representing more than 500 mL of blood, and a positive angiogram, and this correlation gets stronger the earlier one gets a CT scan. However, this improved correlation is not statistically significant, and, hence, Dr Blackmore did not present it. This makes a bit of sense if you consider that an earlier CT that has a significant volume of hemorrhage probably implies more active, aggressive bleeding, hence, a positive angiogram for an arterial injury. Although not a statistically significant correlation, it appears that the earlier you get the CT, and it has more than 500 mL of blood, the more likely an angiogram is to be positive.