Meta-Analysis: Test Performance of Ultrasonography for Giant-Cell Arteritis

Giant-cell arteritis is a common inflammatory vasculopathy of large and medium-size arteries, especially those branching from the proximal aorta (1). It almost exclusively affects individuals older than 50 years of age. Disease susceptibility has been associated with European descent, and two thirds of those affected are women (1, 2). Prompt diagnosis and treatment are important to prevent serious vascular complications, particularly visual loss (3). Given the paucity of disease-specific manifestations and laboratory markers (2), temporal artery biopsy is the diagnostic gold standard (1, 4). Temporal artery biopsy is a generally safe procedure with a complication rate of approximately 0.5% (5). Serious possible complications include facial nerve damage, infection, skin necrosis, and stroke due to an interruption of collateral circulation (5-10). More important, giant-cell arteritis is characterized by segmental inflammation (11), and biopsy results may be negative in 9% to 44% of patients with a clinical diagnosis of the disease (4, 12). The American College of Rheumatology (ACR) has proposed classification criteria for giant-cell arteritis that are based on history, physical examination, and laboratory and biopsy findings (13). However, these criteria are mostly research tools and also have limitations (14). Ultrasonography has been proposed as a useful noninvasive diagnostic test for giant-cell arteritis. Early studies using a continuous-wave Doppler ultrasound flow detector focused on stenoses and occlusions (15-19), but diagnostic performance varied (15-22). Color duplex ultrasonography has better resolution for visualizing vascular structures; it combines real-time imaging with spectral Doppler flow-velocity determination and thus permits assessment of both vessel anatomy and luminal status. Several studies showed that a dark halo around the temporal artery lumen (a sign of arterial wall edema) was specific for giant-cell arteritis (23-30). Other studies, however, failed to reproduce these results (31-33). Many studies unavoidably had limited sample sizes, and their estimates had considerable statistical uncertainty. In addition, it is not known whether aspects of study design might affect estimated test performance. We performed a comprehensive meta-analysis of available evidence on the diagnostic performance of ultrasonography for giant-cell arteritis. Methods Data Sources and Study Identification We searched the MEDLINE, EMBASE, and Cochrane databases for articles published in any language through April 2004 that examined ultrasonography for diagnosis of giant-cell arteritis. The search strategy was based on combinations of index terms: giant cell arteritis, temporal arteritis, cranial arteritis, granulomatous arteritis, or Horton's disease and ultrasonography, ultrasound, duplex, Doppler, accuracy, false-negative, or specificity (34). We screened reference lists of retrieved articles and review articles and asked experts about additional studies. We contacted investigators of original studies to obtain clarifications and missing data. Study Eligibility We included studies that evaluated temporal artery ultrasonography for diagnosis of giant-cell arteritis, enrolled at least 5 patients, and used temporal artery biopsy or the ACR criteria (13) as the reference standard. The ACR requires 3 of the following 5 criteria to be met for the classification of giant-cell arteritis: age at least 50 years at disease onset, new onset of localized headache, temporal artery tenderness or decreased pulse, erythrocyte sedimentation rate of at least 50 mm/h, and an artery biopsy specimen showing vasculitis characterized by a predominance of mononuclear cells or granulomatous inflammation (13). Articles were eligible regardless of the ultrasound method used. Studies examining multiple cranial arteries were considered only if they provided separate data for the temporal artery. Meeting abstracts were not included because the results may not have been final and may not have been subjected to full peer review. Duplicate data were counted only once. Study Quality To assess methodologic quality, we adapted previous criteria for other diagnostic tests (35-37). The revised criteria included 13 items covering 6 dimensions of study quality: technical quality of ultrasonography, technical quality of biopsy, application of the reference test or tests, blinding, description of the study sample, and cohort assembly (Appendix Table 1). Criteria for technical quality of ultrasonography were adapted from a recent consensus statement on carotid ultrasonography (38). The technical quality of temporal artery biopsy is also important to ensure an accurate diagnosis of giant-cell arteritis (4, 11, 39). Two investigators independently graded each study's methodologic quality, and coefficients were calculated to express interrater agreement on each quality element based on the initial appraisal of each reviewer (40). Discrepancies were resolved by discussion. Data Extraction We abstracted data using a standard form that included demographic characteristics of participants, setting, inclusion and exclusion criteria, total number of participants, number of participants evaluated with the index test (ultrasonography) and reference standard (ACR criteria, biopsy, or both), time interval between the 2 procedures and order of examination, and number of patients with clinical subtypes of the disease (cranial arteritis, systemic inflammatory syndrome with arteritis, large-vessel arteritis or aortitis, and polymyalgia rheumatica) (1). We also recorded the ultrasound method, the ultrasonographic criteria used to define vessel abnormalities, the number of ultrasonographers who performed the examination, and the number of pathologists who interpreted the biopsy results. Finally, we recorded whether treatment had been started before diagnostic tests were performed. Two investigators independently abstracted the 2 2 tables needed to calculate sensitivity and specificity. Disagreements were resolved by discussion with a third investigator. Statistical Analysis We synthesized data on the sensitivity and specificity of a hypoechoic halo around the lumen of the temporal arteries or their rami, arterial stenosis or occlusion, and any of these types of vessel abnormality. We separately estimated test performance from studies that used biopsy as the reference standard and from studies applying ACR criteria. We used 2 meta-analytic methods to assess the overall diagnostic performance of each marker of vessel abnormality: 1) weighted independent estimation of sensitivity and specificity values and 2) summary receiver-operating characteristic (ROC) curve analysis. Both fixed-effects (MantelHaenszel) and random-effects (DerSimonianLaird) models (41, 42) were used to estimate weighted sensitivity and specificity across studies. Fixed-effects models weigh each study by the inverse of its variance, while random-effects models also incorporate between-study variation. Random-effects models tend to provide wider CIs and are preferable in the presence of between-study heterogeneity. We report random-effects estimates unless stated otherwise. Heterogeneity was assessed with the Fisher exact test, given the small number of patients in several studies. Because weighted sensitivity and specificity are interdependent, independent calculation may sometimes underestimate both variables. The summary ROC curve analysis is more appropriate because it accounts for this mutual dependence (43, 44). It uses a regression method to fit a curve describing the tradeoff between sensitivity and specificity across studies with different characteristics and thresholds. The regression is D= + S, where D is the difference of the logits of the true-positive rate (sensitivity) and false-positive rate (1 specificity) and S is the sum of these logits. Both weighted and unweighted summary ROC curves (43, 45) were estimated. We used the MannWhitney U test to examine the relationship between sample size and satisfaction of individual criteria for methodologic quality. Post-test probability of giant-cell arteritis was estimated by: Prevalence indicates the probability of the disease in the study sample. Post-test probability after positive and negative results on ultrasonography was calculated for different values of pretest prevalence. Positive and negative likelihood ratios were derived from the meta-analysis on the basis of weighted sensitivity and specificity. We examined the effect of year of publication, language, number of participants (studies that evaluated 30 patients with both ultrasonography and the reference standard [36] vs. those that evaluated <30 patients), and individual aspects of study quality on the diagnostic performance of ultrasonographic findings. Subgroups were compared with general variance methods estimating the standardized difference in effect sizes. Analyses were conducted by using SPSS, version 11.0 (SPSS, Inc., Chicago, Illinois); Meta-Test, version 0.6 (Lau J, Boston, Massachusetts); and Meta-Analyst, version 0.991 (Lau J). Data Synthesis Search Results Of 137 identified reports, 65 were potentially relevant to the meta-analysis (Appendix Figure). Of these, we excluded 31 because they presented duplicate or overlapping data (n= 12); involved fewer than 5 patients (n= 11); or were editorials, comments, or letters without original data (n= 8). Five additional studies were excluded because they described technical aspects of ultrasonography or were anatomic investigations. We also excluded 4 articles that examined ultrasonography in vascular territories other than temporal arteries and 2 studies that did not correlate imaging findings with clinical or histologic data (Appendix). Study Description Twenty-three studies met the inclusion criteria (15-22, 24-33, 46-50). Six articles were published in German (15, 19, 26-28, 48), 2 in French (16, 17), 1 in Italian (18), and 1 in