At present, the optimal protocol for the evaluation of cervical spine injury, particularly in obtunded blunt trauma patients, remains controversial.1–7 In an alert, cooperative patient with a normal neurologic examination, negative computed tomography (CT) imaging is accepted as sufficient to certify that the cervical spine is free of significant injury.1–3 When a patient has impaired mental status, concomitant closed head trauma or other distracting injuries, there is persistent concern that CT evaluation may miss clinically significant injuries.1–3,7 Although it is well recognized that magnetic resonance imaging (MRI) is more sensitive than CT in detecting cervical injuries, particularly those involving the ligaments and soft-tissues, the incorporation of such imaging modalities as a standard part of an evaluation protocol adds expense as well as logistical challenges.1–3,7
Recent studies have reported that the addition of MRI in the evaluation of patients with suspected cervical spine trauma identifies injuries missed on CT in 6% to 46% of cases.7–14 However, several authors contend that the findings missed on CT and then identified on MRI are not clinically significant.5,8,10–13 At the same time, other research7,9 maintains that stand-alone CT is insufficient to certify that the cervical spine is free of clinically significant injury. Nearly all of these works suffer from limited sample size, retrospective design, and the potential for substantial confounding to result from selection as well as indication bias.3,15
Given ethical concerns, a randomized controlled trial to address the ideal imaging protocol for the surveillance of cervical injury following trauma is not possible.3,15 In this context, we sought to utilize propensity-based matching techniques to effectively adjust for selection and indication bias within our cohort. Propensity-score matching is a well described statistical technique that has gained broad acceptance in the literature in recent years and is thought to allow researchers to simulate the environment of a randomized controlled trial.16,17 Based on prior work,14 we hypothesized that cervical spine evaluation using CT-MRI would have a higher likelihood of detecting clinically significant injuries as compared to reliance on CT imaging alone.
MATERIALS AND METHODS
Patient data for use in this study was derived from the Partners Health System Research Patient Data Registry (RPDR). Begun in 1991, the methodology behind this initiative has been extensively described in other works.18–20 In brief, the RPDR is a registry that gathers clinical data, patient demographics, radiology results, and operative reports on all patients treated at Brigham and Women's Hospital (BWH) and Massachusetts General Hospital (MGH), two academic level I trauma centers located in Boston. The RPDR has been successfully utilized in the past to study patients treated for a variety of orthopaedic conditions, including spinal metastases,20 cervical spine fractures,18 and ankylosing spondylitis.19
Inclusion criteria for this study consisted of adult patients receiving CT alone, or CT-MRI, for the primary evaluation of cervical spine injury at BWH or MGH following trauma between 2007 and 2014. Patients with initial evaluations performed at outside centers that were then transferred for care, a prior history of spine surgery or spinal metastases, penetrating trauma, and those lacking complete radiologist reports, emergency room evaluation and/or surgical reports were excluded from further consideration. Individuals without a clear history of trauma, but who had cervical spine imaging ordered for other reasons, were similarly excluded.
Individuals meeting inclusion criteria had their medical records reviewed by five authors (A.J.S., D.G.T., H.V.L., D.A.L., P.R.) and demographic, as well as clinical characteristics were recorded. These included age at time of presentation, biologic sex, race, body mass index, medical comorbidities according to modified Charlson scale,21 Injury Severity Score (ISS), Glasgow Coma Scale (GCS) at presentation, whether CT alone or CT-MRI was used in cervical evaluation, radiology results regarding injury identification and ultimate management, whether operative or nonoperative. Patients in this analysis had CT imaging performed using 128-slice scanner (Somatom Definition AS+; Siemens, Forchheim, Germany; using tube voltage of 120 kVp and automated tube current modulation [CareDose4D] with reference mAs of 180 and 2 mm slice thickness) and MRI performed with 1.5 Tesla scanners (Signa Excite GE Healthcare, Waukesha, WI) with axial and sagittal sequences. Imaging results were recorded directly from the radiologist reports and injury characteristics were abstracted from clinical notes and operative reports. Individuals were stratified into two cohorts based on whether CT alone was used in the evaluation of the cervical spine or CT-MRI was employed. Race was classified as white or non-white (African American, Hispanic, Asian, Native American, Mixed Race, and others). The number of Charlson comorbidities was categorized as zero, one, two, or three or more. Identification of a cervical spine injury (e.g., fracture, dislocation, traumatic disc herniation, ligamentous disruption) on MRI, with associated change in clinical management or surgical intervention was considered the primary outcome variable. Change in management was used as a proxy for identifying clinically significant injuries and was defined, a priori, as those radiographic findings on MRI that resulted in a decision for surgery in patients previously deemed nonoperative based on CT results, or a decision to continue cervical orthoses in individuals where such braces would otherwise have been discontinued.3,15
A logistic regression was performed to identify clinical factors significantly associated with the decision to obtain MRI in addition to a CT. Model calibration to the source data was assessed using the Hosmer-Lemeshow test.22 A propensity score for CT-MRI was then developed using clinical factors that demonstrated P < 0.20 in the logistic regression analysis.16,17,22 The propensity score was used to assemble a matched cohort of patients evaluated using CT alone and CT-MRI with one-to-one matching and no replacement. Cohort imbalance was assessed using the L1 statistic.23
Baseline differences in the propensity-matched cohorts were assessed using chi-square testing for categorical variables and t tests or Wilcoxon rank-sum tests for parametric and nonparametric continuous variables, respectively. The odds of detecting a cervical spine injury were calculated using bivariate logistic regression. In a subsequent test, adjustment was performed for the detection of cervical spine injury on CT imaging within the CT-MRI cohort. The number needed to treat (NNT) was calculated using the absolute risk reduction in detection of clinically significant injury between the CT alone and CT-MRI cohorts. Statistical significance was maintained for those variables with odds ratios (OR) and 95% confidence intervals (CIs) exclusive of 1.0 and P values <0.05. A subset analysis limited consideration to patients with GCS ≤ 9. This study received Investigational Review Board approval prior to commencement.
In total, 12,691 patients were screened for inclusion in this analysis. Ultimately, 8753 were deemed eligible with 8060 having been evaluated using CT alone and 693 receiving CT-MRI (Figure 1). There were numerous significant differences between these initial groups including number of medical comorbidities, ISS, and GCS (Table 1). In the logistic regression analysis used to identify factors associated with the propensity to receive CT-MRI (Table 2), male sex, the number of comorbidities, ISS, and GCS all met criteria for inclusion in the propensity score. There was no evidence that the final model was poorly calibrated to the data (P = 0.84).
Following propensity score matching, two balanced cohorts of 668 patients evaluated using CT alone or CT-MRI were derived. The average age of patients in the CT alone and CT-MRI groups was 52.6 (SD 22.7) and 54.8 (SD 21.7), respectively. Sixty percent of both cohorts were male with white race approximating 75% in both groups. No statistically significant differences (Table 3) in demographic composition, medical comorbidities, ISS, or GCS were detected between the matched cohorts. The L1 statistic was 0.43, indicative of cohort balance.
Among individuals evaluated using CT alone, a cervical spine injury was detected in 36 patients (5%), with one requiring surgery (0.2%). In the CT-MRI group, a total of 248 patients (37%) had an injury identified, with 51 (8%) treated surgically (Figure 2). Among CT-MRI patients, 195 (29%) had a cervical injury previously recognized on CT, whereas 53 (8% of the entire cohort) had an injury identified on MRI that was not detected on CT. The majority of these were ligamentous injuries with two patients found to have fractures and one an occipital-cervical dissociation. Among these 53 patients, the findings on MRI resulted in a change in management in 47 cases, with only five (1%) requiring surgery. In the entire cohort, only one case of neurologic deterioration was encountered and this was attributed to a misplaced lateral mass screw following surgery.
Overall, the odds of identifying a cervical spine injury were significantly higher in the CT-MRI group (OR 10.4; 95% CI 7.2–15.0; P < 0.001) and remained so even after adjusting for injury recognition on prior CT (OR 2.6; 95% CI 1.7–4.0; P < 0.001). When consideration was limited to patients with GCS ≤ 9, the odds of identifying a cervical spine injury remained higher in the CT-MRI group (OR 4.4; 95% CI 1.96–10.0; P < 0.001). In this cohort, however, once recognition of prior injury on CT was taken into account, there was not a higher rate of injury detection after CT-MRI (OR 1.1; 95% CI 0.38–3.03; P = 0.90). The NNT to identify an injury necessitating a change in management was 50 and the NNT to identify an injury requiring surgery was 167.
The safe and effective evaluation of the cervical spine in a patient who has sustained blunt trauma necessitates balancing the risk of missing an occult soft tissue injury against relying only on the findings of screening CT imaging, especially when the patient is obtunded or otherwise unexaminable.1–3,7,9,14,15 Although many authors readily admit that MRI has superior sensitivity for detecting injuries to the cervical spine, they maintain that these findings are most commonly clinically insignificant in most cases and rarely alter management, thus unnecessarily adding logistical challenges and costs to the care of trauma patients.3,15 Although some studies report that in up to 40% of cases, the addition of MRI to a standard CT evaluation protocol will result in a change in management,6,7,9,11,14 there are also numerous efforts that demonstrate CT alone is sufficient to certify the cervical spine as free of clinically relevant injury.4,5,8,10–13 Many of these works suffer from a retrospective design, however, and lack the ability to effectively control for selection and indication bias in the decision to add MRI to a stand-alone CT protocol.3,15
To address this issue using a more judicious approach, we employed propensity score matching16,17 to assemble two cohorts where demographic, injury, and medical characteristics were approximated between the two groups and clinical equipoise could be assumed in terms of decision making regarding adding MRI to a stand-alone CT evaluation. The use of this technique has been maintained to allow researchers to simulate the environment of a randomized controlled trial in situations where only retrospective clinical data are available.16,17 Viewed in this context, our study could be considered akin to a randomized trial where 668 patients were each assigned to a CT alone protocol or one that utilized a combination of CT and MRI.
Our study is advantaged over prior research4–14 on this topic in that it utilized a large number of patients derived from two academic level I trauma centers and relied on statistical methodology that could adjust for confounding by selection as well as indication. It is encouraging that the rate of missed injury identified by MRI (8%) in our study, as well as the rate of surgical intervention in such settings (1%) approximates that reported in previous published research.3,6,7,11,14,15 We believe that these facts speak to the external validity of our results. Our study demonstrated that the addition of MRI to a stand-alone CT protocol for the evaluation of cervical spine injury significantly increased the likelihood of overall injury detection (OR 10.4; 95% CI 7.2–15.0; P < 0.001) and even following adjustment for injury recognition on prior CT evaluation (OR 2.6; 95% CI 1.7–4.0; P < 0.001). The rate of clinically significant injuries (i.e., those that necessitated a change in management) identified on MRI was quite low (47/668). Another way to interpret this data is through an NNT calculation, where it appears that 50 MRI studies would have to be ordered to avoid a missed injury that would necessitate a change in management. Similarly, the number of MRIs necessary to avoid a missed injury warranting surgical evaluation would be 167. Perhaps even more importantly, there were no instances of neurologic deterioration among those individuals where an injury was not detected on initial CT. These findings were robust and persisted in subset analysis that limited consideration to those with GCS ≤ 9. Among those patients with GCS ≤ 9, after taking into account injuries already identified on CT imaging, MRI did not increase the likelihood of detecting an injury over CT-alone. This indicates that stand-alone CT imaging is likely sufficient to detect most clinically significant injuries and nearly all those that might result in spinal instability and neurologic compromise. Although MRI certainly can detect subtle injuries not apparent on screening CT, these are rare enough that the incorporation of MRI into the standard protocol for the evaluation of cervical injury in blunt trauma patients cannot be substantiated.
We acknowledge that there are limitations associated with this study. Foremost, this remains a retrospective study subject to many of the drawbacks associated with this type of research, including errors in coding within the RPDR. There was no standard protocol regarding when MRI should be added to CT in the evaluation of patients with suspected cervical trauma at our centers, which could be confounded by unmeasured variables or physician intuition that cannot be ascertained in a registry study. To adjust for this to the fullest extent possible, we employed propensity-matched techniques to account for differences in selection as well as indication for MRI between the cohort evaluated using CT alone and those receiving CT-MRI. Nonetheless, there is the potential for residual confounding that cannot be meaningfully quantified. We are solely limited to consider patients who presented to BWH and MGH in the time-period under study. Although the clinical and demographic characteristics, including age, are comparable to most other studies in the literature,2,4–14 the results may not be generalizable to populations whose sociodemographic composition, injury characteristics and medical comorbidities lie outside the range of the individuals evaluated in this work. In light of the average age of patients in this study, the results may not be applicable to patients of advanced age or those with severe cervical spondylosis. Lastly, given that all patients were derived from two trauma centers in the same city who share a parent healthcare corporation and academic institution, as well as residencies and fellowship programs, there may be limited heterogeneity in terms of clinical practice and surgical decision making. This could directly impact our determinations regarding clinically significant injuries, as well as those necessitating surgery. If the decision making of surgeons, radiologists, and physicians at our centers are different from those elsewhere in the country, this could impact the translational capacity of our results. As far as we know, however, there is no empirical evidence of such regional variation in radiologic interpretation or surgical practice.
We believe that the results of this research are best applied to blunt trauma patients who have negative CT imaging of the cervical spine and no other clinical findings (e.g., abnormal initial neurological examination, progressive deterioration in function, history of malignancy, or ankylosing spondylitis) that might otherwise increase the likelihood of neurologic injury. We seek to emphasize that this work is the first we are aware of to apply propensity-matched techniques to evaluate the utility of adding MRI to stand-alone CT in the detection of cervical spine injury. In the propensity-matched cohort, the addition of MRI to a CT alone protocol identified missed injuries at a rate of 8%. Only a minority of these, however, were serious enough to warrant surgery. There were no instances of neurologic deterioration among those individuals where injury was not detected on initial CT imaging. Although MRI can certainly prove valuable in the detection of occult injuries following cervical spine trauma, decisions regarding its use are likely best made on a case-specific basis. In light of our study results, we maintain that the standard addition of MRI to CT imaging in the evaluation of all patients with suspected cervical spine trauma cannot be supported.
- This work is the first we are aware of to apply propensity-matched techniques to evaluate the utility of adding MRI to stand-alone CT in the detection of cervical spine injury.
- In the propensity-matched cohort, the addition of MRI to a CT alone protocol identified missed injuries at a rate of 8%.
- In light of our study results, we maintain that the standard addition of MRI to CT imaging in the evaluation of all patients with suspected cervical spine trauma cannot be supported.
The authors would like to acknowledge Wei Jiang, MS, and Ritam Chowdhury MD, PhD, of the Center for Surgery and Public Health at Brigham and Women's Hospital for their guidance and feedback on the approach to propensity-matched scoring. The authors also wish to thank Michael Van Hal, MD, of Beth Israel Deaconess Medical Center for his assistance in abstracting patient records. None of these individuals were compensated for their efforts.
1. Anderson PA, Gugala Z, Lindsey RW, et al. Clearing the cervical spine in the blunt trauma patient. J Am Acad Orthop Surg
2. Simon JB, Schoenfeld AJ, Katz J, et al. Are “normal” multidetector computed tomographic scans sufficient to allow collar removal in the trauma patient? J Trauma
3. Schoenfeld AJ, Harris MB, Davis M. Clinical uncertainty at the intersection of advancing technology, evidence-based medicine and healthcare policy. JAMA Surg
4. Chew BG, Swartz C, Quigley MR, et al. Cervical spine clearance in the traumatically injured patient: is multidetector CT scanning sufficient alone? Clinical article. J Neurosurg Spine
5. Como JJ, Leukhardt WH, Anderson JS, et al. Computed tomography alone may clear the cervical spine in obtunded blunt trauma patients: a prospective evaluation of a revised protocol. J Trauma
2011; 70:345–349. discussion 349–51.
6. Fisher BM, Cowles S, Matulich JR, et al. Is magnetic resonance imaging
in addition to a computed tomographic scan necessary to identify clinically significant cervical spine injuries in obtunded blunt trauma patients? Am J Surg
7. Kaiser ML, Whealon MD, Barrios C, et al. The current role of magnetic resonance imaging
for diagnosing cervical spine injury in blunt trauma patients with negative computed tomography scan. Am Surg
8. Patel M, Humble S, Cullinane D, et al. Cervical spine collar clearance in the obtunded adult blunt trauma patient: a systematic review and practice management guideline from the Eastern Association for the Surgery of Trauma. J Trauma Acute Care Surg
9. Pourtaheri S, Emami A, Sinha K, et al. The role of magnetic resonance imaging
in acute cervical spine fractures. Spine J
10. Raza M, Elkhodair S, Zaheer A, et al. Safe cervical spine clearance in adult obtunded blunt trauma patients on the basis of a normal multidetector CT scan—a meta-analysis and cohort study. Injury
11. Resnick S, Inaba K, Karamanos E, et al. Clinical relevance of magnetic resonance imaging
in cervical spine clearance: a prospective study. JAMA Surg
12. Satahoo S, Davis J, Garcia G, et al. Sticking our neck out: is magnetic resonance imaging
needed to clear an obtunded patient's cervical spine? J Surg Res
13. Soult MC, Weireter LJ, Britt RC, et al. MRI as an adjunct to cervical spine clearance: a utility analysis. Am Surg
14. Schoenfeld AJ, Harris MB, McGuire KJ, et al. Computed tomography alone versus computed tomography and magnetic resonance imaging
in the identification of occult injuries to the cervical spine: a meta-analysis. J Trauma
15. Schoenfeld AJ, Carragee EJ. Clearing the cervical spine in Plato's cave. Spine J
16. Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res
17. Haukoos JS, Lewis RJ. The propensity score. JAMA
18. Harris MB, Reichmann W, Bono CM, et al. Mortality in elderly patients after cervical spine fractures. J Bone Joint Surg Am
19. Schoenfeld AJ, Harris MB, McGuire KJ, et al. Mortality in elderly patients with hyperostotic disease of the cervical spine following fracture: An age- and sex-matched study. Spine J
20. Janssen SJ, van der Heijden AS, van Dijke M, et al. 2015 Marshall Urist Young Investigator Award: prognostication in patients with long bone metastases: does a boosting algorithm improve survival estimates? Clin Orthop Relat Res
21. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol
22. Long JS, Freese J. Regression Models for Categorical Dependent Variables Using STATA. 2nd ed.College Station, TX: STATA Press; 2006.
23. Blackwell M, Iacus S, King G, et al. Cem: Coarsened exact matching in STATA. The STATA Journal