MRI for Detecting Root Avulsions in Traumatic Adult Brachial Plexus Injuries: A Systematic Review and Meta-Analysis of Diagnostic Accuracy

T brachial plexus injuries affect 1% of patients involved in major trauma (1). Brachial plexus injuries may cause permanent disability (2–6), pain (6–8), psychologic morbidity (9,10), and reduced quality of life (2,4,11). Early reconstructive nerve surgery is associated with better functional recovery in the upper limb (2,3,11,12), which improves quality of life (4). Nerve reconstruction for patients with root avulsion(s) is a clinical priority for several reasons. After root avulsion, the cell bodies of motor neurons in the spinal cord recede (4,5,12,13), so the limb never reanimates spontaneously. Because reimplantation of avulsed roots yields no meaningful recovery (14–16), nerve transfers are performed, which significantly improves function (11,17–20). Nerve transfers are relatively minor and cost-effective procedures (21,22) with low morbidity (20,23–25). Furthermore, 95% of patients with traumatic brachial plexus injuries have neuropathic pain (6), and the evolving evidence suggests that early reconstructive nerve surgery reduces cortical reorganization and thus neuropathic or phantom limb pain (26–28). Therefore, correctly identifying patients with root avulsions is of paramount importance. MRI is the best noninvasive test for brachial plexus injuries (29). MRI is superior to nerve and muscle electrophysiology studies (30), US (31–34), and intraoperative somatosensory-evoked potentials (35). Currently, MRI MRI for Detecting Root Avulsions in Traumatic Adult Brachial Plexus Injuries: A Systematic Review and Meta-Analysis of Diagnostic Accuracy

radiology.rsna.org n Radiology: Volume 293: Number 1-October 2019 is unable to help reliably differentiate root avulsion (and other nerve injuries proximal to the intervertebral exit foramen) from injuries outside the foramen. The delay from injury to surgical reconstruction is the leading cause of poor functional recovery (36)(37)(38). To identify patients with root avulsion who need urgent reconstructive nerve surgery, most surgeons use preoperative MRI, although the accuracy of this test remains a topic of debate. Therefore, the prevailing clinical practice involves either protracted observation (for months) or surgical exploration of the brachial plexus.
Numerous studies have examined aspects of MRI for traumatic brachial plexus injuries, and this review aims to evaluate its diagnostic accuracy for the identification of root avulsion(s) in adults with traumatic brachial plexus injuries.

Materials and Methods
This systematic review and meta-analysis was performed in accordance with our protocol (39) and was written in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses-Diagnostic Test Accuracy, or PRISMA-DTA, Statement (40).

Participants and Studies
We included studies of adults with symptomatic brachial plexus injuries sustained from nonpenetrating trauma that reported the findings of preoperative MRI and surgical exploration of the roots of the brachial plexus. We excluded case reports and studies concerning bilateral injuries (41,42).

Target Condition
The primary target condition was avulsion of the root(s) of the brachial plexus. The secondary target condition was a pseudomeningocele, which is purported to be a surrogate marker of root avulsion. Our review was concerned with the ability of MRI to help distinguish between normal roots (no avulsions) and any frequency of root avulsion.

Index Test
The role of MRI is to depict root avulsion (Fig E1 [online]). This examination is typically performed within weeks of the injury and before surgery. The interpretation of MRI for root avulsion is binary, with an implicit threshold. Similarly, images are examined for the presence of a pseudomeningocele (sometimes erroneously termed meningocele), which is also a binary outcome with an implicit threshold. The target conditions can be observed at any spinal level, from C4 to T2, depending on the pattern of plexus fixation. Several systematic differences were expected, including the following: MRI unit brand and model, field strength, pulse sequences, software for postprocessing, and display hardware and viewing environment; these differences were investigated if possible.

Prior Tests
Patients typically undergo extensive clinical assessment and imaging. Some patients might undergo electrodiagnostic studies (nerve conduction and electromyography). All examination and medical test findings would typically be made available to the radiologist interpreting the MRI study.

Reference Standard
The reference standard was surgical exploration of the supraclavicular brachial plexus. This may be supplemented by hemilaminectomy, sensory evoked potentials, and bipolar motor nerve stimulation, and we consider these as part of the reference standard of "surgical exploration." Search Strategy MEDLINE and Embase were searched from inception to August 20, 2018, with no restrictions (Appendix E1 [online]). Citations were independently deduplicated by two authors (R.G.W. and J.C.R.W., both academic plastic and reconstructive surgery registrars with 8 and 6 years of experience, respectively) and were independently screened.

Study Selection
Two authors (R.G.W. and J.C.R.W.) independently screened all citations. The full texts of potentially eligible articles were obtained and were independently assessed by the same two authors (39). Disagreements were resolved by discussion. Eligible articles were imported to Review Manager, version 5 (RevMan; the Nordic Cochrane Centre, the Cochrane Collaboration, Denmark) and were categorized as included or excluded.

Data Extraction
Two authors (R.G.W. and J.C.R.W.) independently extracted data concerning patient demographics (age and sex); mechanism of injury; time frames between injury, imaging, and surgery; imaging protocols; surgical steps; and test statistics to construct 2 3 2 tables of the number of true-positive findings, false-positive findings, false-negative findings, and truenegative findings. The authors of two studies (43,44) provided missing data.
Abbreviations CI = confidence interval, IQR = interquartile range, QUADAS = Quality Assessment of Diagnostic Accuracy Studies, SROC = summary receiver operating characteristic Summary This systematic review and diagnostic test accuracy meta-analysis shows that the overall accuracy of MRI for traumatic brachial plexus root avulsion is modest.

Key Results
n A systematic review and diagnostic test accuracy meta-analysis showed that 72% of patients with traumatic brachial plexus injuries had at least one root avulsion (interquartile range: 53%-86%).
n On average, MRI had a mean sensitivity of 93% and specificity of 72% for root avulsion, with surgical exploration as the reference standard.
n On the basis of these limited data, MRI has modest accuracy for diagnosing root avulsions following traumatic brachial plexus injury, and early surgical exploration should remain as the method of diagnosis.

Risk of Bias and Applicability Concerns
The results of the risk of bias and applicability assessment are summarized in Figure 2. Six studies (43,60,61,(63)(64)(65) were at unclear risk of selection bias because the sampling strategy was unclear or because patients were inappropriately excluded. The study by Hayashi et al (62) was at high risk of selection bias because of the retrospective exclusion of (an unknown number of) patients whose MRI images were suboptimal. The study by Yang et al (67) was also at high risk of selection bias because they

Methodologic Quality Assessment
The risk of bias and the applicability of included studies was assessed by two authors (R.G.W. and J.C.R.W.) using a tailored version of the Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 (Appendix E2 [online]) (39,45), and there were no disagreements.

Statistical Analysis
We used RevMan to generate forest plots and summary receiver operating characteristic (SROC) plots. Estimates of sensitivity and specificity were provided on forest plots. In Stata, version 15 (Stata, College Station, Tex), a bivariate meta-analytical model was used to obtain summary sensitivities and specificities, with the nerve as the unit of analysis (46)(47)(48). A meta-analysis of patient-level data was not performed because of sparse and heterogeneous data. To determine the effect of a covariate on the sensitivity and specificity of MRI, we planned a bivariate meta-regression, but this was not possible because of limited data. We did not assess publication bias because the determinants are not well understood for diagnostic accuracy reviews (48), and the Deeks test has low power in the presence of substantial heterogeneity (49). The significance level was set at 5%. Analyses were performed by R.G.W. and Y.T. (a biostatistician with 12 years of experience).
The unit of analysis and therefore the prevalence of root avulsion varied between studies (Table 1). Five studies  unclear applicability concerns regarding the reference standard because surgeons had knowledge of the MRI findings. Figure 3 shows the forest plots of the sensitivity and specificity of MRI for root avulsion, according to the unit of analysis. There was no apparent influence of field strength on diagnostic accuracy when the patient was the unit of analysis (Fig 3). When the nerve was the unit of analysis, 3.0-T MRI systems appeared to improve the sensitivity and reduce the specificity (Fig 3). Of note, all studies of MRI performed at 3.0 T reported the diagnostic accuracy with the nerve as the unit of analysis, while studies at lower field strengths (0.5 and 1.5 T) used patients.

Synthesis of Results
With the nerve as the unit of analysis, the mean sensitivity of MRI for root avulsion was 93% (95% confidence interval [CI]: 77%, 98%), with a mean specificity of 72% (95% CI: 42%, 90%), as shown in Figure 4. The accuracy of MRI with the patient as the unit of analysis is summarized in a forest plot (Fig 3) and an SROC plot (Fig E2 [online]) and varied considerably; a meta-analysis was not possible owing to sparse and heterogenous data. The clinical consequences of using MRI to diagnose root avulsion are summarized in Table 2 using the median prevalence. There were insufficient data to allow us to estimate the effect of a covariate (eg, field strength) on the accuracy of MRI.
Forest plots of the sensitivity and specificity of pseudomeningoceles as a surrogate marker of root avulsion are shown in Figure 5. No meta-analysis was performed owing to the substantial heterogeneity and sparsity of data.

Discussion
MRI for traumatic brachial plexus root avulsion has a mean sensitivity of 93% (95% confidence interval [CI]: 77%, 98%) and a mean specificity of 72% (95% CI: 42%, 90%). This means that MRI fails to depict seven in 100 avulsed nerves and incorrectly classifies 28 in 100 nerves as avulsed when they are in continuity. Our data suggest that pseudomeningoceles are an unreliable marker of root avulsion. Moreover, the performance of MRI and the clinical consequences vary depending on the prevalence (Table 2).
MRI appears to have modest diagnostic accuracy for root avulsions in adults with traumatic brachial plexus injuries, which is unlikely to be acceptable to patients or surgeons given that a false-negative finding may cause irreversible morbidity. Moreover, it is probable that the diagnostic accuracy of MRI is worse than suggested because there were several methodologic concerns (68). Eight (73%) of 11 studies were at risk for selection bias (60)(61)(62)(63)(64)(65)67), meaning that the samples may not be representative. For example, one study excluded patients whose images were suboptimal (62), which would upwardly bias the accuracy (69). Furthermore, there were issues in the reporting of MRI in all studies that hamper critical appraisal and translation to practice. We decided to ascribe an unclear risk of bias when surgeons knew the MRI findings because the effect of reviewer bias is inconsistent (69). Additionally, the lack of detail regarding pulse sequences limits both the reliability and external validity.
Overall, the quality of the included studies in our review was inappropriately excluded individuals with concomitant injuries. Eight studies (43,44,59,60,62,63,65,67) were at risk for bias concerning the MRI examination; four studies (43,62,63,65) at high risk because no definition or threshold was described, and four studies (44,59,60,67) were at unclear risk because potential features of root avulsion were described but no criteria were provided. All 11 studies were at unclear risk of bias regarding the reference standard because the surgeons performing the exploration had knowledge of the MRI findings; additionally, partial verification bias was present in five studies (61)(62)(63)(64)67). There was an unclear risk of bias in the flow and timing of four studies (43,61,62,64) owing to a lack of information about the time between injury and MRI or surgery.
Five articles (43,60,61,64,67) were of unclear concern regarding the applicability of patient selection because of nonconsecutive recruitment. There were applicability concerns for MRI in nine studies (43,44,59,60,(62)(63)(64)(65)67); four studies (43,62,63,65) were of high concern because no MRI criteria or thresholds were described, and five studies (44,59,60,64,67) were of unclear concern because multiple radiologists reported the images (59)(60)(61)(62)(63)(64)67), the methods were not described (43,44,65), the description of the MRI protocol was different from the parameters shown in figures (60), or the pulse sequences were not adequately described (43,44,59,63-65,67). All 11 studies had and therefore the accuracy of MRI (71). Patients with a brachial plexus injury are typically victims of polytrauma who undergo several tests that may identify a problem obscuring or superseding the brachial plexus injury. Patients with polytrauma are typically treated by clinicians who do not specialize in nerve injury, so the intrinsic threshold for identifying the condition and referring the patient might be lower. Moreover, most surgeons do not perform hemilaminectomy when exploring the supraclavicular brachial plexus, so the status of the root proximal to the exit foramen is unknown, which might lead to further underestimation of the prevalence. Overall, several factors suggest that the true prevalence of root avulsion could be higher than observed, which may have important clinical ramifications.
Currently there is no evidence nor consensus to suggest whether a more sensitive or specific test is preferable. A more sensitive test would enable clinicians to rule out root avulsions (potentially avoiding unnecessary exploratory surgery), although the fiscal and functional costs of false-negative findings are substantial-notably, lifelong disability (2-6), pain (6-8), mental illness (9,10), and impaired quality of life (2,4,11). A more specific test might rule in root avulsion, enabling clinicians to avoid exploration and proceed directly to nerve transfer while eliminating prolonged surveillance (36)(37)(38). Furthermore, avoidance of unnecessary surgery would reduce morbidity for patients and costs for health services. We accept that a proportion of patients who are not found to have root avulsion may have other nerve injuries that warrant reconstruction; however, the diagnosis of root avulsion is of paramount importance. Therefore, improving the specificity of MRI for root avulsion appears to be most desirable.
In upper extremity nerve surgery, there is no consensus on the unit of analysis (patient, limb, or nerve). This is a common issue in surgical research but is profoundly important in upper extremity research (72,73). In this situation, if the patient is poor because 71% of all QUADAS domains were assessed as high or unclear risk, which is 15% worse than diagnostic test accuracy reviews in other fields (70).
Our review suggests that root avulsions are found in approximately three in four patients, but given the variability of the estimates and quality of the original data, this is unlikely to be reliable. There are several clinical and artifactual factors that might reduce the apparent prevalence of root avulsions  131 no meta-analysis was performed. Their review had no protocol and was not reported in accordance with recommended guidelines. Furthermore, CT myelography carries a demonstrable risk of cancer induction. Therefore, we disagree with the authors' findings and suggest that their conclusion is too optimistic (76) and that the hazards of CT myelography outweigh the potential benefits as compared with MRI.
Our review had limitations. While 20 potentially relevant articles were identified, data extraction was impossible for nine articles (50)(51)(52)(53)(54)(55)(56)(57)(58). Retrospective studies of medical tests tend to report inflated estimates of diagnostic accuracy (68); our review included four retrospective studies (59,60,66,67) and seven articles that incompletely described the time frames between injury, imaging, and surgery. Similarly, studies recruiting nonconsecutive (nonrandom) samples are prone to selection the unit of analysis, then MRI would classify patients as either negative (avulsion free) or positive (having between one and five avulsions), which fails to discriminate between distinct patterns of roots injury and wrongly apportions one to five avulsions the same result, introducing composite bias (74). Conversely, if the root is the unit of analysis, then composite bias is avoided, and five clinically meaningful and distinct assessments are provided. Therefore, we advocate using the nerve as the unit of analysis.
There is one systematic review related to this topic that compares the performance of MRI to CT myelography for the diagnosis of root avulsions (75). The authors concluded that MRI is "an excellent test for assessing traumatic brachial plexus injuries"; however, this statement is not substantiated because all three studies were judged to be at risk for methodologic bias and  bias and inflated estimates of accuracy (68); our review included four such articles (43,(61)(62)(63). We expected the accuracy of MRI to improve over time, in line with technologic and methodologic advances, although this is not appreciable (Fig 3). This discord is likely multifactorial, and certainly highquality prospective research is needed.
In conclusion, on the basis of limited and heterogeneous data, MRI offers modest diagnostic accuracy for traumatic brachial plexus root avulsion(s), and early surgical exploration should remain as the preferred method of diagnosis.