| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Emergency Radiology |
1 From the Department of Radiology, CB 7510, School of Medicine, University of North Carolina–Chapel Hill, Chapel Hill, NC 27599–7510 (C.C.B.); and the Department of Radiology, Harborview Medical Center (C.C.B., F.A.M.), the Departments of Biostatistics (S.S.E.) and Epidemiology (T.D.K.), and the Robert Wood Johnson Clinical Scholars Program (C.C.B., T.D.K.), University of Washington, Seattle. From the 1997 RSNA scientific assembly. Received October 28, 1997; revision requested January 20, 1998; revision received September 1; accepted December 15. C.C.B. supported by the Seattle Veterans Affairs–Robert Wood Johnson Clinical Scholars Program. Address reprint requests to C.C.B.
 |
Abstract |
|---|
 TOP Abstract IntroductionMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
MATERIALS AND METHODS: Records from 472 patients with trauma (168 with fractures, 304 control patients) who visited the emergency department in 1994 and 1995 were reviewed for 20 potential predictors of cervical spine fracture in this retrospective case-control study. Simple logistic regression was used to determine predictors of cervical spine fracture. Prediction rules were formulated by using multiple logistic regression and recursive partitioning with bootstrap validation. Posttest fracture probabilities were calculated from base prevalence and likelihood ratios derived for predictors by using Bayes theorem.
RESULTS: Predictors of cervical spine fracture included severe head injury (adjusted odds ratio [OR] = 8.5, 95% CI: 4.0, 17.0), high-energy cause (OR = 11.6, 95% CI: 5.4, 25.0), and focal neurologic deficit (OR = 58, 95% CI: 12, 283). The prediction rule was used to stratify patients into groups with fracture probabilities of 0.04%–19.70%. After adjusting for overfitting, the area under the receiver operating characteristic curve was 0.87.
CONCLUSION: Clinically apparent factors, including cause of injury, associated injuries, and age, can be used to determine the probability of cervical spine fracture. Development of evidence-based imaging guidelines should incorporate knowledge of fracture probability.
Index terms: Receiver operating characteristic curve (ROC) • Spine, fractures, 31.11, 31.12, 31.41 • Spine, injuries, 31.11, 31.12, 31.41 • Trauma, 31.41
|
Introduction |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
Screening radiography of the cervical spine is expensive, costing society as much as $140 million annually (7), and usually has a low yield, with only 1%–5% of screening studies showing a fracture (8–10). In addition, depending on the clinical situation, from 4% to 28% of such screening radiographic examinations may lead to further imaging, without a fracture being present (11). Accordingly, considerable attention has been focused on developing optimal guidelines for cervical spine imaging. However, the same imaging strategy may not be appropriate for all patients. Patients with a very low probability of fracture may not need any imaging (12,13), whereas those with a modest probability of injury may require radiography. In addition, several authors (14–16) have suggested that patients with a high probability of fracture may benefit from screening computed tomography (CT). The key to determining who should undergo screening and to selecting the ideal imaging modality is the probability of fracture.
In efforts to predict fracture probability, several studies (13,17–20) have attempted to define clinical and radiologic factors that are associated with cervical spine fracture. These studies have examined factors such as cause of injury and coexistent injuries, including head injury. However, these studies have revealed few predictors of cervical spine fracture. As an example, although head injuries may result from causes similar to those of cervical spine injury, findings of several clinical studies (17,20,21) have failed to demonstrate an association between head injury and cervical spine injury. In addition, several investigators (18,22,23) disagree whether cause of injury, such as a high-speed motor vehicle accident, is associated with cervical spine fracture. However, many of these studies rely on trauma registries or on other inpatient databases that may not be representative of all patients with trauma (24). Further, imaging triage in many emergency departments is based on factors, such as accident cause and associated injuries, in contradiction to the reports that suggest such factors do not predict injury.
The objective of this study was to develop a method to determine the clinical probability of cervical spine fracture to guide selection of optimal imaging strategies. We defined the clinical predictors of cervical spine fracture and quantified the degree to which the presence of each factor influences the probability of fracture. In addition, we determined the absolute probability of cervical spine fracture for several clinical scenarios and developed clinical prediction rules to stratify patients into groups with different fracture probabilities.
|
MATERIALS AND METHODS |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
Patient Selection
Patients with cervical spine fractures were retrospectively identified from the inpatient trauma registry at a major urban trauma center (Harborview Medical Center) and consisted of 168 patients who were evaluated in 1994 and 1995, who were at least 18 years of age, who were not transferred from another institution, and who had nonpenetrating trauma. At our institution, Harborview Medical Center, Seattle, Washington, all patients with potentially unstable cervical spine fractures are admitted to the hospital, so the inpatient trauma registry was considered appropriate for the selection of patients with fractures.
As control subjects, 304 patients without cervical spine fractures were randomly selected from a database that lists all 20,617 patients with trauma who were evaluated in the emergency department. This database differs from traditional trauma registries in that it includes patients who were discharged from the emergency department, as well as patients who were admitted to the hospital. Control patients were selected by means of random sampling within the two strata, as defined by admission status, by using identical sampling fractions in each stratum. Thus, the control patients were selected from all patients with trauma, not only those who were admitted to the hospital. As with the selection of patients with fractures, patients with penetrating trauma, those who were transferred from another institution, and those younger than 18 years were excluded.
Data Acquisition
Clinical factors that may be predictors of cervical spine fracture were selected from sources of information that would have been available to the trauma team prior to their deciding on cervical spine imaging. Clinical predictors selected are listed in Table 1 and include findings from the patient history (eg, cause of injury), the initial screening physical examination (eg, facial laceration), and the initial diagnostic tests (eg, intracranial hematoma at CT). We assumed that head CT had a greater priority than cervical spine imaging in the evaluation of the patient with trauma and that preliminary information from head CT would have been available before cervical spine imaging. By using all available information from emergency department notes, ambulance reports, and preliminary diagnostic studies (those performed prior to complete cervical spine evaluation), one author (C.C.B.) and a research assistant trained to abstract data from medical charts determined the presence of each potential clinical predictor for both control and case populations. Abstracted data were entered into a computer database by using double-key entry and verification. Double review of 5% (26/472) of the charts was performed to assess interobserver agreement by using the 
statistic.
|
In the second phase of data analysis, predictors that were clinically similar, strongly associated with each other in the study data, and connoted similar probabilities of cervical spine fracture were combined into composite predictors.
In the third phase of data analysis, these composite predictors were used to produce the clinical prediction rules, which are models for predicting the probability of cervical spine fracture. Clinical prediction rules were derived from models used in both multiple logistic regression and recursive partitioning analysis. Multiple logistic regression is a method used to determine whether a factor is a clinical predictor of fracture after adjusting for other known predictors. Recursive partitioning is a tree–based method of using the clinical predictors to split the whole sample into successively smaller groups with differing probabilities of fracture.
Stepwise formation of models was undertaken, with area under the receiver operating characteristic (ROC) curve used for comparison of each successive model (STATA, College Station, Tex). The area under the ROC curve was calculated by using the trapezoid method (STATA Reference Manual, 5th ed, Vol 2, Stata, College Station, Tex, 1997, page 366). Bootstrap validation (26), a method for testing the prediction rule with new samples that are randomly selected from the original sample, was used to estimate the degree of overfitting that resulted from our evaluating the performance of the rule by using the data set that was used to derive it. Final estimates of the area under the ROC curve were corrected for this overfitting.
In the last phase of data analysis, the grouping scheme from the simplified recursive partitioning model was used to identify combinations of clinical factors, or clinical scenarios, that had similar risks of fracture (eg, severe head injury from a high-energy cause). The probability of fracture for each of these clinical scenarios was then determined by multiplying the likelihood ratios (27) from the prediction rule by the base prevalence (in the form of pretest odds) of cervical spine fracture in the entire population; this process is an application of Bayes theorem (28). The base prevalence of fracture was calculated by dividing the total number of fractures in the eligible group by the total number of eligible patients examined at our institution in the study time frame. CIs for the fracture probability estimates were calculated by using the bootstrap technique (26).
In the bootstrap technique, the numbers of fractures and nonfractures were each assumed to follow a Poisson distribution, with rate parameters estimated from the observed data. For each bootstrapped sample, individual cases were resampled with replacement from the observed data to obtain the simulated sample size. Recursive partitioning of the bootstrap sample was performed by using the same criteria for model selection as was used for the observed data. The bootstrapped recursive partitioning analysis was then used to estimate the probability of fracture for each pattern of predictors, and those estimates were combined to obtain estimates of the probability of fracture for the clinical scenarios defined by the final model. The results from 1,000 such bootstrapped samples were used to compute the bootstrapped CIs (26).
Further explanation of the method used to derive the clinical prediction rule is provided in the Appendix.
|
RESULTS |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
Statistical significance for each factor as a predictor of cervical spine fracture is indicated by P values. Significant predictors of cervical spine fracture include age, cause of injury (eg, motor vehicle accident at a high speed, fall), injury to the head or face (eg, skull fracture, facial fracture), and focal neurologic deficit. Seat belt use was associated with a decreased probability of fracture. Agreement between chart abstracters, as determined by means of double chart review, was excellent, with an average of 0.80 (29).
Table 2 describes the composite predictors, which represent clinically similar predictors that suggest similar probabilities of fracture and that, therefore, have been pooled. Table 2 also lists the adjusted ORs for each of the composite predictors derived from the stepwise multiple logistic regression. The adjusted ORs represent the odds of fracture in a patient with a given composite predictor, adjusted for all of the other predictors shown. Severe and mild head injury, high- and moderate-energy cause, age over 50 years, and focal neurologic deficit are all significant independent predictors of cervical spine fracture.
|
|
|
|
DISCUSSION |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
Cervical spine fractures can result from several biomechanical mechanisms, including flexion, compression, axial loading, and rotation. Often, these forces are transmitted to the cervical spine as a result of trauma to the head. It is not surprising, therefore, that injuries to the head or face are found in over half of all patients with cervical spine injuries (30) and that cervical spine injuries occur in 5%–21% of all patients with head injury (17,31,32). Our results demonstrate a prevalence of cervical spine fracture of 7.2% (Fig 2) in patients with severe head injuries, which is similar to with the findings of these previous studies. However, previous clinical articles have been published that could not demonstrate an association between head injury and cervical spine fracture (17,20,21). Our study findings may help to resolve this controversy and show that head injury is a strong predictor of cervical spine fracture.
There are two reasons why our study design is superior to that of other studies: selection of appropriate control subjects and use of a case-control study design. Unlike previous investigators (l7,19–21,33,34) who used inpatient trauma registries, we selected control subjects from among all patients with trauma who were evaluated in the emergency department. Trauma registries, because they include only patients who are admitted to the hospital, are not representative of all patients with trauma who need care. Trauma registry patients tend to be more severely injured, be older, and have more coexistent morbidities than do patients who are not admitted (24). Our control population, in contrast, is representative of all patients with trauma seen at the study hospital.
An additional advantage of our control population is that our control patients were selected from a group identified before any imaging decisions were made. Decisions to admit patients and, therefore, to enroll them in the trauma registry can be made, in part, on the basis of findings from cervical spine imaging studies. Since our goal is to guide the decision to use imaging, it is important to use only the information that is available before imaging occurs. Most decisions to use cervical spine imaging are made prior to admission, so the relevant control population for a study designed to influence imaging decisions is patients with trauma who are evaluated in the emergency department, not only patients who are admitted.
In addition, our case-control study design has the further advantage of increasing the power of the study beyond that of other study types that involve the same number of patients (25). The power of a study is the ability to detect an association when it exists. For a rare condition, such as cervical spine fracture, the power of a study will generally depend on the number of cases included. Since we were able to use data from 168 patients with cervical spine fractures, the power of our study may be higher than the power of other studies with fewer cases of fracture, even if the total number of patients in the other studies is higher. Other studies that investigated clinical predictors of cervical spine fracture have included many fewer cases of fracture and, therefore, have lower power (8–10,13,18,21–23). A true association might have been present, but it was not identified by the research findings. On the basis of power calculations, our study had an approximate 98% probability of identifying any factor that was present in 10% of the cases and any factor that was associated with a twofold higher probability of fracture (25).
We acknowledge several limitations of this study. Because not all of our predictors are objective, there is a potential for misclassification. For example, the distinction between high- and low-speed motor vehicle accidents is arbitrarily set at 30 mph (48 km/h) and is based on the reports of victims and ambulance crews. It is possible that the speed traveled is exaggerated when extensive damage is present or when severe injuries are sustained. However, we used the same information that would have been available at the time of imaging decision making. Therefore, the prediction rule should achieve similar accuracy in the clinical setting. In addition, we performed bootstrap validation of the prediction rule. This method helps ensure that the accuracy of the prediction rule is not exaggerated by our evaluating it with the same data that were also used to derive it. However, external validation by using data from a different trauma center would be useful to determine whether the results can be generalized to other settings (35).
For this study, we defined severe head injury as skull fracture, intracranial hematoma, brain parenchymal contusion, or unconsciousness at the time of emergency department evaluation. Because the presence of intracranial hematoma is a life-threatening, though potentially treatable, injury, at our institution and at many others, head CT in patients with trauma is performed prior to definitive evaluation of the cervical spine. Therefore, in our base analysis, we included information from head CT in the prediction rule. However, we recognize that information from head CT might not be available at all institutions before cervical spine imaging decisions are made. The prediction rule can also be used when severe head injury is defined as only unconsciousness at the time of initial evaluation, with only a slight decrease in accuracy; the area under the ROC curve is 0.83. Patients who are unconscious at the time of initial evaluation have a probability of cervical spine fracture of 8.6% under this new assumption.
Because we relied on retrospective chart review, we were limited to factors that were recorded in the medical record. As an example, we were unable to obtain estimates of speed for motorcycle accidents. Therefore, all motorcycle accidents were grouped together as accidents with unknown speed. It is intuitive that high-speed motorcycle accidents have a higher risk, similar to that of high-speed motor vehicle accidents. However, we were unable to explore this possibility with this set of data. In addition, we were not able to study factors such as neck pain or focal tenderness over the spinous processes because they were not reliably recorded in the medical record.
Despite this, however, the success of the prediction rule is demonstrated by its ability to stratify patients into groups with probabilities of cervical fracture that ranged from 0.04%–19.70%. Clinical judgment is obviously vital to determining which patients should undergo imaging. Nevertheless, we believe that the study findings can provide a valuable framework for decision making about who should undergo imaging and can guide the further prospective research that is needed.
In conclusion, the cause of injury, the patient's age, and the presence of severe head injury or focal neurologic deficit are important predictors of cervical spine fracture. By using these simple factors, patients can be assigned to groups with a broad range of probabilities of fracture. The probability of fracture can then be used as the foundation for cost-effectiveness analysis and for the development of evidence-based guidelines to help optimize the use of cervical spine imaging.
|
APPENDIX |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
The second goal of the analysis, developing the clinical prediction rule, was more complex. To be clinically useful, the number of clinical predictors would, by necessity, have to be small. In addition, many of the predictors identified in the initial analysis were strongly associated with each other and were clinically similar (eg, facial fracture and facial laceration). Accordingly, we sought to combine similar predictors into composite predictors.
It was obvious from the initial exploration of the data that there were several clinical factors with similar ORs for cervical spine fracture. For example, many types of head injury, including skull fracture and brain parenchymal contusion, suggested similar levels of risk of cervical spine fracture. However, post hoc combining of clinical factors on the basis of their ORs for injury may lead to overfitting of the prediction rule to the data from which it is derived, thereby limiting its generalizability.
Accordingly, we sought to combine factors into composite predictors on the basis of clinical grounds, by using the unadjusted ORs secondarily to guide our determination of the number of categories. As an example, examination of the range of head and facial injuries revealed rough distinctions among factors with ORs around 2, among those with ORs around 5, and among those with ORs in the 8–10 range. Accordingly, we believed it was appropriate to stratify individual head injury predictors and facial injury predictors into several composite predictors.
However, we combined the clinical predictors into composite predictors on the basis of clinical grounds rather than on the ORs. For example, using clinical grounds, we were unable to discriminate among many factors with ORs around 2 (eg, facial laceration) and among those with ORs around 5 (eg, scalp laceration). However, there is a clear clinical distinction between skull fracture or intracranial hematoma (with ORs around 8–10) and facial or scalp laceration (with ORs around 2–5). Accordingly, we developed only two composite predictors, severe head injury and mild head injury, with allocation of factors to each of these groups based on clinical considerations.
Several of the simple predictors (eg, sex, use of vehicular occupant restraint device, focal neurologic deficit) had no clinically similar correlates and were, therefore, considered individually as potential composite predictors. The simple predictor of age was a continuous variable that we converted to a categorical variable of age over 50 years to simplify the model. We selected 50 years of age because, qualitatively, this threshold seemed to be the best discriminator of patients with fracture and patients without fracture.
The next step in the analysis was to use the composite predictors to develop a clinical prediction rule. This was accomplished through the use of both stepwise multiple logistic regression and recursive partitioning. Initially, stepwise logistic regression was used with a P value cutoff of .05 for inclusion in the model. From this initial analysis, the composite predictors of age over 50 years, high-energy cause, moderate-energy cause, severe head injury, mild head injury, and focal neurologic deficit were all significant predictors and therefore were included in the model. Use of seat belts, sex, and intoxication were not significant predictors and were therefore excluded. We believed that this model was too complex for bedside use and, therefore, excluded the predictor of mild head injury from the final model. The simplified stepwise logistic model had an area under the ROC curve of 0.88. The area under the ROC curve was calculated by using the STATA statistics package, which uses the trapezoid method (STATA Reference Manual, 5th ed, Vol 2, Stata, College Station, Tex, 1997, page 366).
The problem with the logistic regression model was that bedside use might be limited by the mathematics involved, and there was still interaction among some of the composite predictors. In an effort to further simplify the model, we used recursive partitioning. The composite predictors derived from the stepwise logistic regression model were evaluated through recursive partitioning by using P values as the criteria for stratification. The ROC curve for the recursive partitioning model was derived by converting the final recursive partitioning end groups to indicator variables, followed by logistic regression. The area under the ROC curve was then calculated, as before. The recursive partitioning model was also simplified for bedside use. The final area under the ROC curve for the simplified recursive partitioning model (Fig 1) was 0.88, which, when corrected for overfitting by using bootstrap validation, decreased only slightly to 0.87. The final groups from the simplified recursive partitioning model defined the clinical scenarios used for calculation of fracture probability.
|
Footnotes |
|---|
Abbreviations: OR = odds ratio ROC = receiver operating characteristic
Author contributions: Guarantors of integrity of entire study, C.C.B.; study concepts, C.C.B., F.A.M., T.D.K.; study design, C.C.B., S.S.E., F.A.M., T.D.K.; definition of intellectual content, C.C.B., S.S.E., F.A.M., T.D.K.; literature research, C.C.B.; data acquisition, C.C.B.; data analysis, C.C.B., S.S.E., T.D.K.; statistical analysis, C.C.B., S.S.E., T.D.K.; manuscript preparation, C.C.B., S.S.E.; manuscript editing and review, C.C.B., S.S.E., F.A.M., T.D.K.
|
References |
|---|
|
TOPAbstractIntroductionMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  J. Blackham and J. Benger `Clearing' the cervical spine in conscious trauma patients Trauma, April 1, 2009; 11(2): 93 - 109. [Abstract] [PDF]
|
|
|
|
 |
|
 W. C. Black, E. A. Krupinski, A. Relyea-Chew, and F. S. Chew Methodology and Application of Clinical Trials in Radiology: Self-Assessment Module Am. J. Roentgenol., March 1, 2008; 190(3_Supplement): S23 - S28. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. D Ross and J. M Cheeks Undetected Hangman's Fracture in a Patient Referred for Physical Therapy for the Treatment of Neck Pain Following Trauma Physical Therapy, January 1, 2008; 88(1): 98 - 104. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. C. Blackmore Clinical Prediction Rules in Trauma Imaging: Who, How, and Why? Radiology, May 1, 2005; 235(2): 371 - 374. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. G. M. Hunink Decision Making in the Face of Uncertainty and Resource Constraints: Examples from Trauma Imaging Radiology, May 1, 2005; 235(2): 375 - 383. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. D. Bub, C. C. Blackmore, F. A. Mann, and F. M. Lomoschitz Cervical Spine Fractures in Patients 65 Years and Older: A Clinical Prediction Rule for Blunt Trauma Radiology, January 1, 2005; 234(1): 143 - 149. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. C. Blackmore and P. Cummings Observational Studies in Radiology Am. J. Roentgenol., November 1, 2004; 183(5): 1203 - 1208. [Full Text] [PDF]
|
|
|
|
 |
|
 C G Morris, E P McCoy, and G G Lavery Spinal immobilisation for unconscious patients with multiple injuries BMJ, August 28, 2004; 329(7464): 495 - 499. [Full Text] [PDF]
|
|
|
|
|
|
I. F. Gareen and C. Gatsonis Primer on Multiple Regression Models for Diagnostic Imaging Research Radiology, November 1, 2003; 229(2): 305 - 310. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Berlin CT Versus Radiography for Initial Evaluation of Cervical Spine Trauma: What Is the Standard of Care? Am. J. Roentgenol., April 1, 2003; 180(4): 911 - 915. [Full Text] [PDF]
|
|
|
|
|
|
F. M. Lomoschitz, C. C. Blackmore, S. K. Mirza, and F. A. Mann Cervical Spine Injuries in Patients 65 Years Old and Older: Epidemiologic Analysis Regarding the Effects of Age and Injury Mechanism on Distribution, Type, and Stability of Injuries Am. J. Roentgenol., March 1, 2002; 178(3): 573 - 577. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. H. Daffner Identifying Patients at Low Risk for Cervical Spine Injury: The Canadian C-Spine Rule for Radiography JAMA, October 17, 2001; 286(15): 1893 - 1894. [Full Text] [PDF]
|
|
|
|
|
|
R. H. Daffner Helical CT of the Cervical Spine for Trauma Patients: A Time Study Am. J. Roentgenol., September 1, 2001; 177(3): 677 - 679. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. C. Blackmore, W. N. Zelman, and N. D. Glick Resource Cost Analysis of Cervical Spine Trauma Radiography Radiology, September 1, 2001; 220(3): 581 - 587. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. R. Perry, E. J. Stern, F. A. Mann, and A. B. Baxter Lateral Radiography of the Cervical Spine in the Trauma Patient: Looking Beyond the Spine Am. J. Roentgenol., February 1, 2001; 176(2): 381 - 386. [Full Text] [PDF]
|
|
|
|
|
|
R. H. Daffner Cervical Radiography for Trauma Patients: A Time-Effective Technique? Am. J. Roentgenol., November 1, 2000; 175(5): 1309 - 1311. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Zipser Invited Commentary RadioGraphics, October 1, 2000; 20(90001): 261S - 262. [Full Text]
|
|
|
|
|
|
J. A. Hanson, C. C. Blackmore, F. A. Mann, and A. J. Wilson Cervical Spine Injury: A Clinical Decision Rule to Identify High-Risk Patients for Helical CT Screening Am. J. Roentgenol., March 1, 2000; 174(3): 713 - 717. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. C. Blackmore, S. D. Ramsey, F. A. Mann, and R. A. Deyo Cervical Spine Screening with CT in Trauma Patients: A Cost-effectiveness Analysis Radiology, July 1, 1999; 212(1): 117 - 125. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
