| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Evidence-based Practice |
1 From the Department of Radiology, Harborview Medical Center, University of Washington, Box 357115, 1959 NE Pacific St, RR 215, Seattle, WA 98195-7115 (L.D.B., C.C.B., F.A.M.); Harborview Injury Prevention and Research Center, Seattle, Wash (C.C.B.); and the Department of Radiology, University of Vienna, Austria (F.M.L.). From the 2003 RSNA Scientific Assembly. Received October 19, 2003; revision requested January 12, 2004; revision received March 5; accepted April 8. C.C.B. supported in part by Agency for Healthcare Research and Quality grant K08 HS11291. Address correspondence to L.D.B. (e-mail: lbub@u.washington.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Institutional review board approval was received with waiver of informed consent. A retrospective case-control study was performed on blunt trauma patients 65 years and older with cervical spine fractures and on randomly selected control subjects without fracture. Potential predictors of fracture were evaluated through simple and multivariate logistic regression. Simple predictors were grouped into clinically similar composite variables and were analyzed with multivariate logistic regression and recursive partitioning. A clinical prediction rule was generated. The receiver operating characteristic curve was calculated and adjusted through bootstrap validation. Absolute cervical spine fracture probabilities were calculated by using Bayes theorem for all elderly patients and for patients who underwent computed tomography. Results were compared with a previous prediction rule for all adults.
RESULTS: Composite predictors of fracture in the elderly included focal neurologic deficit (adjusted odds ratio, 17.7; 95% confidence interval [CI]: 3.8, 83.4), severe head injury (odds ratio, 3.2; 95% CI: 1.5, 7.1), high-energy mechanism (odds ratio 6.7; 95% CI: 3.1, 14.8), and moderate-energy mechanism (odds ratio 3.3; 95% CI: 1.3, 8.3). The prediction rule stratified patients into risk groups with fracture probabilities ranging from 0.4% (95% CI: 0.1%, 1.3%) to 24.2% (95% CI: 5.7%, 100%).
CONCLUSION: Clinical factors can be used to stratify patients 65 years and older into risk groups with a wide range of probabilities of cervical spine fracture. Knowledge of cervical fracture risk can help guide appropriate imaging in high-risk patients.
© RSNA, 2004
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Findings from several studies have demonstrated that, compared with patients younger than 65 years, elderly patients require lower-energy mechanisms to fracture their cervical spines, and these fractures are more frequently overlooked (5–7). Also, the biomechanic response of the cervical spine to blunt trauma in elderly patients is different from the response in younger patients. Senescent degenerative changes tend to occur in the mid- and lower cervical spine, conferring a greater degree of mobility onto the C0 through C2 motion segments, which is where fractures most often occur in the elderly (8,9). The relatively lower force that is required to fracture the aging cervical spine is likely a combination of this altered biomechanic response and osteopenia. Findings from a recent analysis revealed that, in elderly patients, 15%–40% of cervical spine fractures are overlooked in the initial survey, compared with 4% of fractures overlooked in patients younger than 65 years (5). Missed injuries were attributed to difficulty visualizing the minimally displaced fractures in osteoporotic bone, altered bony anatomy resulting from spondylosis and spondylolisthesis, and "satisfaction of search" due to the presence of multiple, less substantial findings. The resulting delay or failure of diagnosis may lead to otherwise preventable spinal cord injuries (10,11).
The standard method for evaluating the cervical spine for trauma is the conventional radiograph series, which usually consists of anterior, lateral, and odontoid views. While this series remains the primary screening tool for most patients, 5%–8% of patients with cervical spine fractures will have normal radiographs (12–14). Also, obtaining adequate views may be difficult in the acute trauma setting and may prolong emergency department stays. Patients with abnormal, equivocal, or inadequate findings will then require further imaging. Accordingly, several authors have advocated computed tomographic (CT) screening for cervical spine fracture in high-risk patients, and CT screening is now routinely performed at many trauma centers across the United States (12–19).
Cost-effectiveness analysis demonstrates that CT is cost effective in patients undergoing head imaging and who have a greater than 5% probability of cervical spine fracture (20). Factors such as severe closed head injury, a high-energy mechanism of injury, and neurologic deficits have been associated with an increased risk of cervical spine fracture; together, these risk factors form clinical prediction rules that can be used to identify patients who are at a sufficiently high risk of injury to justify CT screening (15,19).
Because of the biomechanic changes that occur in the cervical spine and the relatively low energy required for fracture in elderly patients, we hypothesized that clinical predictors of cervical spine fracture in the elderly may differ from predictors that have been established for the general adult population. Findings from several studies have demonstrated differences in the clinical presentation of elderly patients with cervical spine fractures (5–8,21–24). To our knowledge, no study has attempted to predict the probability of fracture by using readily available clinical and radiographic data. Thus, the purpose of our study was to determine the clinical predictors of cervical spine fracture in the elderly and to develop a clinical prediction rule to guide appropriate imaging in high-risk patients.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Study Design
We performed a retrospective case-control study of blunt trauma patients 65 years and older with cervical spine fractures (case patients) and randomly selected blunt trauma patients 65 years and older without cervical spine fracture (control subjects). We defined clinical and radiographic predictors of cervical spine fracture and determined the degree to which these predictors were associated with the presence of fracture. A clinical prediction rule was established to stratify patients into risk groups according to different fracture probabilities. The study design parallels that of a previous study of the clinical predictors of cervical spine fracture in all adults that encompassed the same institution during the years 1994–1995 (15).
Patient Selection
Case patients.—The inpatient trauma registry of our regional trauma center was queried for blunt (ie, nonpenetrating) trauma in patients 65 years and older with cervical spine fracture who presented between January 1, 1995, and December 31, 2002. The trauma registry is a computerized database of demographics, mechanisms, injury diagnoses, and other information for all trauma patients who were either admitted to or died in the emergency department of our urban level 1 referral trauma center (25). Cervical fractures were defined as fractures involving any aspect of the C1 through C7 vertebrae. Confirmation of fracture was based on the review of medical records and radiologic reports by a single investigator (L.D.B.). Patients who had penetrating trauma and those patients who died prior to cervical spine imaging were excluded. Patients who were transferred from other institutions were also excluded to avoid selection bias on the basis of referral patterns.
Control subjects.—Control subjects were randomly selected from a database of all patients who were evaluated in the emergency department between January 1, 1995, and December 31, 2002, regardless of whether the patients were admitted or discharged. Selection criteria included an age of 65 years and older, a blunt trauma mechanism of injury, and the absence of cervical spine fracture. Absence of cervical fracture was based on trauma registry coding; these findings were further confirmed through review of emergency department records and imaging reports by an investigator (L.D.B.). Patients who were transferred from outside medical centers or who died prior to imaging were excluded.
Data Collection
Background data were obtained from an emergency department database. A set of clinical and radiologic factors that were potentially associated with cervical spine fractures were identified by using literature review and clinical experience. Only information that would have been available to the treating emergency department physicians was included. These potential clinical predictors included aspects of patient history, physical examination, preliminary imaging, and laboratory results (obtained prior to definitive cervical spine imaging). A standard data form was completed for each patient by a single unblinded investigator (L.D.B.). Double review of the first 25 of 210 charts (12%) was performed approximately 1 month after the initial review to calculate intraobserver agreement while minimizing memory bias. Head CT results (including intracranial hemorrhage, skull fracture, and facial fracture) were included because our emergency department prioritizes head CT ahead of cervical spine evaluation, and the results of CT are often available to clinicians before definitive cervical spine imaging is performed. Also, previous analysis has shown that severe head injury is highly predictive of cervical spine fracture in the general population (15).
Statistical Analysis and Development of Clinical Prediction Rule
The data were analyzed in multiple stages in consensus by two of the authors (L.D.B., C.C.B.). Intraobserver agreement on data collection was calculated by using the 
statistic.
In the first stage of analysis, each potential clinical predictor was evaluated with simple logistical regression, and an odds ratio was generated (Stata 6.0; College Station, Tex). The odds ratio is the ratio of the odds of fracture in patients with a given clinical predictor to the odds of fracture in patients without the predictor. To account for multiple comparators, Bonferroni-corrected P values of .002 were used to evaluate for statistical significance (26).
Second, each variable was examined after accounting for the effects of the other predictors in a forward stepwise multivariable logistic regression model. In this model, independent predictors were combined into composite predictors that were based on previous research and clinical similarity. Intracranial hemorrhage, skull fracture, and unconsciousness at examination (including all subjects with an endotracheal tube) were combined into "severe head injury"; facial fracture and transient loss of consciousness were combined into "moderate head injury"; superficial facial trauma was considered "mild head injury"; high-speed motor vehicle crash (defined as an accident occurring at a speed of 
30 miles per hour), a fall from 10 ft (3 m) or greater, a car striking a pedestrian, and an airplane accident were combined into "high-energy mechanism;" low-speed motor vehicle crash (<30 miles per hour), a fall from less than 10 ft (3 m), and a skiing accident were considered "moderate-energy mechanisms"; and a fall from a standing or sitting position was considered a "low-energy mechanism."
The overall prevalence of cervical spine fracture in our elderly population was calculated from the number of eligible patients with fracture and the overall number of eligible patients in the study time frame. The probability of cervical spine fracture for patients with each of the various composite predictors was determined from the odds ratios and from the likelihood ratio form of Bayes theorem.
Age and sex were evaluated both as independent predictors and as "effect modifiers" because the effects of increasing age or sex may influence the predictive value of other variables. For example, a fall from 3 ft (1 m) may not be predictive of cervical spine fracture in a 20-year-old adult, but the same mechanism may predict fracture in an 85-year-old woman because of the modifying effects of age and sex.
The final clinical prediction rule was developed with recursive partitioning, which is a tree-based method of analysis that divides the sample into risk groups according to the diminishing probability of fracture. Various models were constructed and evaluated by their receiver operating characteristic (ROC) curves and clinical relevance. The area under the ROC curve was calculated by using the trapezoid method. Bootstrap validation was performed on the final model by using the Stata bootstrap module to correct for data overfit. Bootstrap validation is a method of determining the validity of a model by retesting the model on a large number of new samples. One thousand new samples are derived by sampling with replacement from the original study sample. The area under the ROC curve was determined for each resample. The bootstrap replications were used to obtain the bootstrapped confidence intervals (CIs), and the area under the ROC curve was adjusted for overfitting.
Because some authors recommend cervical spine CT only in patients who are to undergo head CT, we performed a subanalysis of the prediction rule on those patients who underwent head CT. The prevalence of cervical spine fracture was calculated separately for elderly patients who underwent head CT and was based on the number of cervical spine fractures in eligible elderly patients who underwent head CT versus the total number of eligible elderly patients who underwent head CT. The total number of eligible patients who underwent head CT was estimated from the number of patients in the control sample who underwent head CT.
Cervical spine fracture probabilities for subjects who underwent head CT and the various predictors for CT were estimated from the odds ratios by using the likelihood ratio form of Bayes theorem. The decision to perform head CT was not evaluated in the multivariate logistic regression model or incorporated into the multivariable prediction rule. The reason for this is that there are no specific criteria for performing head CT in our emergency department. Rather, head CT identifies a complex group of patients with multiple combinations of risk factors, including high-energy trauma, endotracheal intubation, loss of consciousness, confusion, superficial head injuries, and focal neurologic findings. We could not include head CT in the multivariate models without confounding the effects of the other predictors.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Medical records were obtained for 103 of the 104 case patients. One patient’s records were unavailable. Of 210 randomly selected control subjects, 103 were excluded because they were transferred from other institutions, died prior to imaging, had penetrating trauma or burns, or had unavailable medical records. A total of 107 control subjects were included in the study. Demographics and descriptive statistics of the study population are shown in Table 1. Intraobserver agreement for each potential predictor ranged from a of 0.77 to 1.0, which is excellent (27).
|
Simple Predictors
The results of simple logistic regression analysis are presented in Table 2, including odds ratios, 95% CIs, and P values. Significant predictors of cervical spine fracture included focal neurologic deficit, indicators of severe or moderate head injury (ie, unconscious at examination, endotracheal intubation, intracranial hemorrhage, facial and skull fractures, transient loss of consciousness, and Glasgow Coma Scale score less than 15), and high-speed motor vehicle crash. The use of seatbelts was associated with fewer cervical spine fractures but was not statistically significant because the numbers were small. Falls from a standing or sitting position were associated with decreased fracture risk compared with other mechanisms of injury. Age, sex, intoxication, confusion, and fractures not involving the face or skull were not significant independent predictors of fracture. Age and sex were not predictive of fracture when evaluated as effect modifiers. Performing head CT, regardless of CT results, was predictive of cervical spine fracture.
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
The current study developed from a previous clinical prediction rule that was derived and validated on a general adult trauma population at our institution (15,19). The previous study contained patients ranging in age from 18 to 88 years, with a mean age of 43 years. On the basis of clinical factors, adult patients were stratified into risk groups, with fracture risks ranging from 0.04% to 19.7%.
As with the general population, the three groups of elderly patients at highest risk for fracture included patients with neurologic deficit (24.2%), those with severe head injury (7.9%), and those with a high-energy mechanism of injury (3.4%). Low-energy trauma (ie, a fall from standing or sitting) in the absence of a focal neurologic deficit or severe head injury was associated with the lowest risk of fracture: 0.4%. When compared with the previously reported clinical prediction rule for all adult patients, fracture probabilities were higher for the elderly in every risk group.
For those patients in the highest risk groups, the increase was minimal, but elderly patients with moderate- or low-energy trauma had a substantially higher fracture risk than previously reported for all adult patients. While patients with low- and moderate-energy trauma have relatively low fracture probabilities overall, the elderly are still more likely to sustain a fracture from a low- or moderate-energy injury than are other adult patients. When the prediction rule is applied to patients who underwent head CT, fracture risks increased slightly. Elderly patients with blunt trauma injuries who did not require head CT had a low risk for fracture (0.2%).
Prediction of cervical spine fracture can be challenging, and the area under the ROC curve for the prediction rule in the elderly is somewhat less than was previously reported for all adults (0.82 vs 0.88). This may be explained by the importance of low-energy mechanism, including a fall from standing or sitting. In the elderly control population, a fall from standing or sitting caused 49% of injuries. However, in the general adult trauma population, low-energy falls caused only 21% of injuries (15). Therefore, despite the relative low probability of cervical spine fracture from a low-energy fall, a substantial number of fractures are the result of low-energy falls because such fractures are common mechanisms of injury among the elderly. In addition, fractures resulting from low-energy falls commonly had few associated injuries. Eleven patients (11%) in our study fractured their cervical spine in falls from standing or sitting. Seven of these 11 patients (64%) had no neurologic deficits, no head injuries, no loss of consciousness, and no associated injuries. These patients were not identified by using any of the evaluated clinical or radiographic criteria and, therefore, will be missed by the clinical prediction rule. These patients are one reason that cervical spine fractures are more difficult to predict in the elderly, who may also be the patients at highest risk for missed fractures at imaging.
The National Emergency X-Radiography Utilization Study (NEXUS) and the Canadian C-Spine prediction rules are two prospective cohort studies that identify patients at extremely low risk for fracture (28–30). Both studies enrolled large numbers of patients—34 000 and 9 000, respectively—and both studies developed sets of clinical criteria with 99.6%–100% sensitivity for fracture. These studies concluded that imaging does not need to be performed when patients meet all low-risk criteria.
Ngo et al (31) performed a subgroup analysis of patients 80 years and older from the NEXUS data. He reported a 4.7% prevalence of cervical fractures in the "very elderly," nearly double the 2.4% prevalence of injuries in the population as a whole. The estimate by Ngo et al of fracture prevalence in the elderly is higher than what we report. However, this finding is not surprising because we considered all elderly patients with blunt trauma injuries as our denominator, while Ngo et al considered only patients who underwent cervical imaging. The sensitivity and negative predictive value of the NEXUS prediction rule were 100% among the very elderly, indicating that, despite the higher frequency of injury, the NEXUS rule remains effective in selecting elderly patients who do not require imaging. While the Canadian C-Spine prediction rule, the NEXUS rule, and the Ngo et al subgroup analysis accurately identify low-risk patients, these studies do not address the stratification of high-risk patients.
Previous cost-effectiveness analysis demonstrated that CT screening of the cervical spine was cost effective in patients undergoing head CT who had an approximately 5% or greater risk of fracture (17). In the previous analysis, cervical spine CT screening was determined to be cost effective in all adult patients undergoing head CT with focal neurologic deficits, severe head trauma, and high-energy mechanisms of injury. We identified no additional risk groups in the elderly population with a sufficiently high risk of fracture to merit CT screening. It should be noted that the cost-effectiveness analysis that was described earlier was derived from a model that used the general population. Therefore, this model may or may not hold true in the elderly. Patients older than 65 years have a shorter life span than do the general population, and cost per quality-adjusted life-year would increase; however, sensitivity and specificity of radiographic screening would likely decreases in the elderly, thereby counterbalancing the effects of a shorter life expectancy. Further cost-effectiveness analysis would be required to better assess who should undergo primary imaging with CT in the elderly population.
Because our study was retrospective, we were limited to information contained in the chart, laboratory data, and radiographic reports. Only clinical factors that were consistently reported could be evaluated. Prospective clinical indicators of fracture, such as midline cervical spine tenderness (which was used in the NEXUS study), could not be assessed because these indicators were not reliably recorded. If we are to predict fractures in minimally injured patients, findings such as midline cervical tenderness may play an important role.
One potential source of bias in our study is the presence of a clinical decision rule already in place in our emergency department. Our blunt trauma clinical decision rule mandates that all patients who are intubated, unconscious, have high-energy trauma, or have severe head injuries and who will be undergoing head CT should also undergo CT screening of the cervical spine instead of radiography. This protocol increases the number of subtle fractures found in these patients; such fractures may not have been visible on radiographs and are of uncertain clinical importance. This protocol may therefore lead to an overestimation of the association between the above factors, head CT, and cervical spine fractures. While we cannot accurately assess the number of injuries that were missed when radiography was used to triage patients, the numbers are likely to be small. If the fractures are radiographically occult, patients with considerable or persistent neck pain are often further evaluated with magnetic resonance imaging or CT, particularly if the patient has osteopenia or degenerative disk disease. If a fracture was diagnosed after the initial evaluation, the patient would be added to the trauma registry at the time of diagnosis and identified in our search strategy. In addition, because the criteria for performing head CT were not formalized, the association between cervical fracture and head CT may vary between medical centers.
At our medical center, we admit a large number of high-energy trauma patients, which raises the question of whether these results can be generalized to centers that treat primarily patients with low-energy trauma. Despite the fact that falls were the dominant mechanism of injury in our total study population, motor vehicle crashes caused the highest number of fractures. However, because our case patients and control subjects were selected from the same population of elderly blunt trauma patients, the relative predictive values should remain the same, even in medical centers that admit mostly low-energy trauma patients. The ideal method for confirming the accuracy and generalizability of the prediction rule would be to prospectively validate it at another center.
In conclusion, a clinical prediction rule for cervical spine fracture in blunt trauma patients 65 years and older was developed that stratifies patients into subgroups with fracture risks ranging from 0.4% to 24.2%. This rule was essentially identical to a prediction rule already developed and verified on the general adult population (15).
Cervical spine fractures are more difficult to predict in the elderly than in other adults because fractures caused by low-energy trauma occur more frequently and without the associated clinical factors that are predictive of injury. A separate evaluation focusing on low-energy mechanisms of injury would be required to identify independent predictors in this subgroup.
|
FOOTNOTES |
|---|
Authors stated no financial relationship to disclose.
Author contributions: Guarantor of integrity of entire study, L.D.B., C.C.B.; study concepts, all authors; study design, C.C.B., L.D.B.; literature research, L.D.B., C.C.B.; clinical studies, L.D.B., C.C.B.; data acquisition, L.D.B.; data analysis/interpretation, L.D.B., C.C.B.; statistical analysis, C.C.B., L.D.B.; manuscript preparation and definition of intellectual content, L.D.B., C.C.B.; manuscript editing, revision/review, and final version approval, all authors.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  R. Dammers, H. W C Bijvoet, M. J Driesse, and C. C J Avezaat Occurrence of malignant vertebral fractures in an emergency room setting Emerg. Med. J., October 1, 2007; 24(10): 707 - 709. [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]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
).