| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1
Department of Radiology, University of Washington-Harborview Medical Center,
Box 359728, 325 Ninth Ave., Seattle, WA 98104-2499
2
Department of Radiology, University of North Carolina-Chapel Hill, School of
Medicine, CB 7510, Chapel Hill, NC 27599-7510
3
Nortwest Asthma and Allergy Clinic, 4500 Sand Point Way N.E., Ste. 222,
Seattle, WA 98105
Received October 20, 1998;
accepted after revision July 19, 1999.
Address correspondence to C. C. Blackmore
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. A 2-year, single-institution retrospective case-control study was conducted of 31 cases of traumatic aortic injury and 171 random major trauma control subjects. The presence of potential injury predictors was determined from chart review. Logistic regression was used to determine injury predictors, and clinically similar predictors were combined into composite predictors. The composite predictors were used to develop a seven-point injury index clinical prediction rule using multivariate logistic regression. Injury probabilities were determined through Bayes' theorem. Bootstrap validation was performed.
RESULTS. Predictors of aortic injury included head injury (odds ratio, 18.3; 95% confidence interval [CI], 7.3-46), pelvic fracture (odds ratio, 27.3; 95% CI, 8.8-85), pneumothorax (odds ratio, 27.3; 95% CI, 8.8-85), and lack of seat belt use (odds ratio, 6.8; 95% CI, 2.6-17). The seven composite predictors of age, unrestrained vehicle occupant, hypotension, thoracic injury, abdominopelvic injury, extremity fracture, and head injury, were combined into the seven-point injury index. In the injury index, each composite predictor had an adjusted odds ratio of 7.1 (95% CI, 3.7-13.5), and the odds ratios were additive. The injury index prediction rule had an area under the receiver operating characteristic curve of 0.97. All injured patients had at least one composite predictor.
CONCLUSION. The probability of traumatic aortic injury can be estimated from the injury index prediction rule. Because cost-effectiveness of various imaging strategies depends on probability of injury, the prediction rule can guide imaging selection.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Unfortunately, there are usually no clinical signs or symptoms that cause suspicion for aortic injury [1]. Accordingly, all patients with a mechanism of injury that might lead to acute traumatic aortic rupture undergo imaging evaluation to screen for aortic injury. Although a large number of individuals sustain trauma of sufficient mechanism to place them at risk for acute traumatic aortic rupture, few patients with the injury survive to hospital presentation; therefore, the yield from screening is extremely low. In a recent report by Fabian et al. [3], only 71 of 8000 patients screened at a single major trauma center sustained acute traumatic aortic rupture. Similarly, Fenner et al. [4] reported seven cases of acute traumatic aortic rupture among 1031 trauma patients. Overall, despite the fact that acute traumatic aortic rupture is rare, excluding acute traumatic aortic rupture is a frequent and time- and resource-intensive process in trauma care.
The traditional method of screening for acute traumatic aortic rupture is with chest radiography. However, controversy persists regarding the criteria for calling the mediastinum abnormal (or "widened") on chest radiography. Further, no consensus exists regarding the definition of abnormal or normal findings on chest radiographs [5, 6, 7, 8, 9], and experts do not agree on further imaging strategies for patients with abnormal findings. Some authors advocate using the appearance of the mediastinum on chest radiographs to triage for CT [10], whereas others advocate using the appearance of the mediastinum to select patients for angiography [11, 12]. Other researchers claim that even patients with normal mediastinal appearance on chest radiographs should undergo further imaging [13, 14]. Finally, several recent studies recommend using CT instead of chest radiography to screen high-risk patients [3, 10, 15, 16, 17, 18, 19].
In 1995, Hunink and Bos [20] reported that the most cost-effective imaging strategy for evaluating the thoracic aorta will depend on the probability of injury. Therefore, development of a method for determining clinical probability of acute traumatic aortic rupture before imaging can potentially improve the efficiency of screening for acute traumatic aortic rupture and aid in selection of the most cost-effective imaging algorithm. Several previous studies have examined clinical factors, including sternal fracture, diaphragm injury, and first rib fracture, that may be associated with aortic trauma [21, 22, 23]. Although associations have been shown between aortic trauma and other injuries, no single factor or combination of factors has been identified that is sufficient to stratify patients into differing probabilities of aortic injury. We reasoned that such a stratification scheme could be used to select the appropriate cost-effective aortic screening study for each patient. The use of clinical factors and nonmediastinal injuries to determine probability of acute traumatic aortic rupture can allow better determination of what imaging is appropriate and how the results of such imaging can be interpreted.
The goal of this study was to develop and validate a simple prediction rule, based on readily apparent clinical and nonmediastinal imaging information, that determines an individual patient's probability of having sustained blunt acute traumatic aortic rupture.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
To approximate five control subjects per patient, we randomly selected 200 control subjects from patients who presented to our emergency-trauma center with major trauma during the same study period. Of these, 180 met final eligibility criteria. Major trauma is a triage classification that is usually made in the field by emergency medical support personnel. Patients who are victims of trauma of sufficient energy to put them at risk for important injury and are therefore transported to a level 1 trauma center (either by surface or air transport) are classified as major trauma victims. This determination is made before imaging evaluation.
To avoid referral bias, we excluded from both groups any patients who were transferred from another institution. Because transfer patients are evaluated at another institution, they may receive mediastinal imaging before arrival at our trauma center and therefore may be transferred on the basis of imaging findings. Because our goal was to influence imaging decision making, it was critical that we exclude patients who might already have been imaged. Patients who were victims of penetrating trauma were also excluded, as were control patients and patients for whom charts could not be located.
For each patient and control subject one of the authors performed retrospective chart review to determine the presence or absence of 15 clinical factors that were potential clinical predictors of aortic injury. These were all clinical factors that would be readily apparent to the examining physicians in the course of the initial trauma center evaluation, including initial nonmediastinal imaging. Because our concern was with information from the initial examination of the trauma patient, we used only imaging and clinical information from the patient's initial emergency-trauma center medical records and the record from the transport to the hospital. Any inpatient medical records or imaging after admission was not included in the medical chart review. The clinical factors included the presence of other injuries, hypotension (any systolic pressure <90 mm Hg), and lack of use of air bag or occupant restraint devices (unrestrained). The chart reviewer used the information in the emergency department notes, emergency department nursing flow sheets, and transport notes from the ambulance or helicopter crews. The nursing sheets included a summary of skin integrity, and the emergency department charts included physical findings and results of initial radiographic evaluation. Because nearly all patients were victims of motor vehicle accidents, and because mechanism of injury was a criterion for inclusion in the category of major trauma from which the patients and control subjects were derived, we were unable to evaluate mechanism directly as a predictor of injury.
Data analysis consisted of two components, determination of simple predictors and development of the clinical prediction rule. In the initial phase, whether each clinical factor was a predictor of aortic injury was determined through bivariate logistic regression. Missing data were excluded from the analysis (an analyzable cohort strategy). Results are expressed as odds ratios for injury.
The second phase of the data analysis was developing a clinical prediction rule that could be used in the clinical trauma setting to determine the probability of aortic injury from clinical and nonmediastinal imaging. To serve in such a role, the clinical prediction rule needed to be simple to apply, based on easily assessed factors, and highly predictive. We determined that to be applicable in the trauma setting, the prediction rule would have to be free of interaction terms or other complex calculations. We also decided that to be simple enough, the prediction rule would have to contain no more than five to eight factors. Because significant statistical interactions between injuries to adjacent anatomic structures often occur, we combined those injuries occurring within anatomic regions to provide composite predictors. These combinations were made on the basis of anatomic relationships, not predictive ability or odds ratios. For example, we combined all thoracic injuries, including pneumothorax, rib fracture, chest abrasion, and pulmonary contusion, into the thoracic injury composite predictor.
Factors that did not correspond to anatomic regions (age, unrestrained, hypotension) were also included as separate composite predictors. The factor "diagnostic peritoneal lavage positive" was a special case that was not included in the final model. Positive findings on diagnostic peritoneal lavage that did not result in immediate abdominal surgery had a high likelihood of representing a false-positive test. Furthermore, the injury risk from diagnostic peritoneal lavage was encompassed and better represented by the variable abdominal surgery. Therefore, diagnostic peritoneal lavage was excluded from the final model. The development of the composite predictors is illustrated in Table 1. The composite predictors were assessed with stepwise multiple regression analysis that allowed determination of whether each composite factor was an independent predictor of acute traumatic aortic rupture.
|
We hypothesized that the probability of acute traumatic aortic rupture would increase as the number of composite predictors increased. Therefore, as the next step in the data analysis, an injury index prediction rule was developed by measuring the total number of composite predictors that were present for each patient. The injury index represents a simple count of the number of composite predictors that were present in each patient. To determine the probability of acute traumatic aortic rupture for patients having zero, one, two, three, or four or more of the composite predictors, we first calculated likelihood ratios from the clinical prediction rule. Next, using the likelihood ratio form of Bayes' theorem and the base prevalence of traumatic aortic injury in the major trauma population at our institution, we calculated the probability of acute traumatic aortic rupture for patients with various numbers of the composite predictors (STATA; STATA, College Station, TX). The base prevalence of injury is the number of patients who sustained acute traumatic aortic rupture of the total number of eligible major trauma patients seen at our institution in the relevant time.
Validation of the injury index prediction rule was accomplished using the bootstrap technique. Bootstrap validation is a method of randomly resampling from a given experimental sample to simulate the effect of drawing multiple samples from the same population. It is used to gauge the generalizability of a clinical prediction rule by ensuring that the rule is not overfit to the data from which it is derived [24]. We also calculated 95% confidence intervals (CIs) for the injury probabilities by performing simultaneous bootstrap resampling of the base prevalence of injury and the likelihood ratio for each level of the injury index.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Multiple factors were predictors of aortic injury from the bivariate logistic regression (Table 2). In fact, the presence of essentially any other injury increases the probability of an aortic injury. In addition, all patients with blunt traumatic aortic injury had at least one other injury. The composite predictors were defined on the basis of age, use of vehicle occupant restraint, hypotension, and anatomic region (thoracic injury, abdominopelvic injury, extremity fracture, and head injury). The seven composite predictors shown in Table 3 are all significant predictors in the multiple regression analysis, which means that each is a predictor after accounting for the effects of the other six. Only one study patient had six of the seven possible composite predictors (and had sustained aortic injury), and none had all seven composite predictors.
|
The probability of fracture was directly related to the number of composite predictors present, as shown on the injury index scale (Table 4). A patient with no predictors had a probability of aortic rupture of 0.0 because no patients with acute traumatic aortic rupture had zero predictors. The 95% CI intervals for zero predictors were 0-0.13%. Presence of one of the composite predictors connoted a probability of injury of 0.19% (95% CI, 0.04-0.54%), with increasing probability for higher numbers of composite predictors. A patient with four or more of the predictors had the highest probability of injury (30%; 95% CI, 11-83%). The area under the receiver operating characteristic curve for the injury index was 0.97, and the injury index explained 64% of the variance in acute traumatic aortic rupture probability (R2 = 0.64).
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Although other authors have examined clinical factors that may be associated with aortic injury, this study is different from previous work in a number of important ways. First, we use a case-control study design, thereby greatly increasing the power of the study for the relatively rare blunt thoracic aortic injury [25]. Second, in addition to measuring the clinical predictors separately, we used multiple logistic regression to determine whether each factor is a predictor, even accounting for the effects of all the other predictors. Third, to our knowledge we have for the first time developed and validated a methodologically rigorous yet relatively simple injury index composed of clinical predictors.
Several factors that we studied in this analysis have been previously examined. Kram et al. [21] noted an association between aortic injury and hypotension, pelvic fracture, myocardial contusion, and intraabdominal injury. They were unable to identify an association between aortic injury and several of the other factors we studied, including rib fracture, spine fracture, and extremity fracture. However, their series had only 10 patients with injury, limiting the power to detect association. Von Oppell et al. [23], Kieny and Charpentier [26], and Kodali et al. [27] also reported a high prevalence of extremity, cerebral, pelvic, and abdominal injury with aortic trauma. However, the case series of all these researchers included no control subjects. Thus, risk factors could not be assessed.
Our results are concordant with those of these earlier studies [21, 23, 26, 27], but we expand the results into a prediction rule that can stratify patients on the basis of injury probability. This prediction rule can aid radiologists, emergency physicians, and trauma surgeons in the development of efficient, cost-effective protocols for the evaluation of the major trauma patient. One application that demonstrates the usefulness of the study is to combine our results with the cost-effectiveness analysis of Hunink and Bos [20]. In that study, the researchers compared the relative cost-effectiveness of six strategies for screening the aorta in blunt trauma patients. The researchers used a decision analysis model from the societal perspective and showed that the cost-effectiveness of aortic screening depends on the probability of aortic injury. According to Hunink and Bos, for patients who are not to undergo CT of other organs, CT of the chest for evaluation of the aorta becomes most cost-effective if the risk of injury is between 0.5% and 5%, corresponding with having two or three of the risk factors from the clinical prediction rule. In patients who are at greater than 5% risk of injury (four or more of the predictors), aortography will be the most cost-effective initial screening method; and if the risk is less than 0.5% (less than two of the predictors), then chest radiography is the optimal initial screening tool.
Our study does have limitations. Motor vehicle accidents are the most common cause of aortic injury. However, because all but four of our patients were victims of motor vehicle accidents, we were unable to evaluate whether being a victim of a motor vehicle accident was a predictor of aortic injury. We did test the model with a variable for motor vehicle accident as the cause of injury. Adding this variable did not improve the predictive ability of the model, and motor vehicle accident itself was not found to be a predictor of injury. However, we can draw no conclusions from this result because the power for evaluating motor vehicle accident was very low. Another potential limitation of this study is the relatively small sample size, with only 31 cases of aortic injury. Use of the case-control study design, with an excess of control subjects, does help to increase the power of the study [25]. Further, the statistical significance of the results indicates sufficient power even with relatively small numbers. Finally, because this is a retrospective chart review, we relied on factors that are recorded in the medical record. We were unable to study the patients' subjective complaints, or the fine details from the scene of the accident, because these are not reported reliably in the medical record.
Critical to any case-control study is the selection of appropriate case and control subjects. In this paper, we define our study population as those patients who meet the triage definition of major trauma. This determination is made in the field, before transport and arrival at the trauma center. Patients are placed in the major trauma classification before any mediastinal or other imaging that might determine the presence of aortic injury and therefore determine whether a patient would be a "case" or a control subject. This practice ensures that the cases and controls are derived from the same population and limits potential for selection bias. It is possible for a patient initially not considered to have major trauma to be reclassified as major trauma in the trauma center if more severe injuries are discovered (including aortic injury). However, such changes in classification would actually add a conservative bias to the prediction rule. In other words, such bias would make the prediction rule appear less effective than it actually is. The prediction rule we present here has strong discriminating ability despite this potential bias.
An additional potential problem with our selection of the study population is that of generalizability. Trauma centers have a working definition of major trauma that is made either in the field or on arrival at the trauma center. However, definitions may vary between centers. We believe that our prediction rule will be equally effective in other major trauma centers, but like any prediction rule, validation in other populations is essential to ensure generalizability [28, 29]. The results of this study can potentially serve as the impetus for a future multicenter prospective trial.
Our goal was to produce a clinical prediction rule that was simple enough to allow bedside use. Accordingly, we combined several of the simple predictors into composite predictors on the basis of anatomy. This approach expands the usefulness of the prediction rule but results in some loss of predictive ability. For example, using the composite predictors, the injury index cannot differentiate between injuries within a body site (i.e., between rib fracture and pulmonary contusion). Nonetheless, we believe that the prediction rule presented is both simple and accurate enough for use in the emergency or trauma center.
The objective of this study was to develop a clinical prediction rule to determine the probability of acute traumatic aortic rupture before screening [29]. Accordingly, we did not consider the chest radiographic appearance of the mediastinum in our analysis. The effectiveness of screening with chest radiography has been vigorously debated elsewhere [3, 6, 7, 9, 10, 13, 17, 19, 20, 30] and is beyond the scope of this analysis.
In this paper we report a methodology to stratify patients into differing levels of probability of aortic injury. Using readily apparent factors, we found that we could stratify patients according to aortic injury probabilities that ranged from 0% to 30%. By determining the probability of injury, we found it was then possible to select the most cost-effective imaging strategy for a given patient and to develop evidence-based imaging guidelines for blunt trauma patients in general.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  S. D. Steenburg, J. G. Ravenel, J. S. Ikonomidis, C. Schonholz, and S. Reeves Acute Traumatic Aortic Injury: Imaging Evaluation and Management Radiology, September 1, 2008; 248(3): 748 - 762. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Brink, J. Deunk, H. M. Dekker, D. R. Kool, M. J. R. Edwards, A. B. van Vugt, and J. G. Blickman Added Value of Routine Chest MDCT After Blunt Trauma: Evaluation of Additional Findings and Impact on Patient Management Am. J. Roentgenol., June 1, 2008; 190(6): 1591 - 1598. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. FURLOW Aortic Emergencies Radiol. Technol., September 1, 2005; 77(1): 41 - 49. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
