Diagnostic Tests for Renal Artery Stenosis in Patients Suspected of Having Renovascular Hypertension

In a general hypertensive population, the prevalence of renovascular disease varies between 1% and 5% (1, 2). In patients who exhibit specific clinical clues that indicate renovascular hypertension, the probability of having the disease increases to 20% to 40% (3). Renovascular hypertension is currently treated with drugs; surgery; or less invasive techniques, such as percutanous transluminal renal angioplasty with or without stent placement. A recent study found no significant difference between treatment with percutanous transluminal renal angioplasty and antihypertensive drug therapy. However, in that study, 44% of patients who received drug therapy still received percutanous transluminal renal angioplasty (4). Because a large proportion of patients are still preferably treated with angioplasty, a valid diagnostic test is needed in the presence of one or more clinical clues (5). Intra-arterial x-ray angiography is considered the gold standard; however, it is invasive and carries small risks for serious complications, such as arterial dissection or adverse contrast reactions. During the past few decades, many researchers have reported results of other, less invasive diagnostic tests. It is, however, difficult to compare the tests' diagnostic performance. Reviews have reported broad ranges of sensitivity and specificity for a certain test that overlap the ranges reported for other tests. Furthermore, studies differ in case-mix, specific test characteristics, and cutoff points for a positive test result, all of which may affect estimates of sensitivity and specificity. The objective of our meta-analysis was to compare five noninvasive or minimally invasive modalities currently used in clinical practice to detect renal artery stenosis. The tests were computed tomography angiography (CTA), magnetic resonance angiography (MRA), ultrasonography, captopril renal scintigraphy, and the captopril test. Intravenous renal angiography and conventional renography were not considered because earlier studies showed that their accuracy was limited (6, 7). Methods Data Sources We searched MEDLINE using the following terms: magnetic resonance imaging and angiography and (renal artery or hypertension, renovascular); tomography, x-ray computed and angiography and (renal artery or hypertension, renovascular); ultrasonography and angiography and (renal artery or hypertension, renovascular); and captopril and angiography and (renal artery or hypertension, renovascular). No beginning date limit was used. The search was updated until 1 August 2000. Only English-, German-, or French-language studies were considered because the investigators were familiar with these languages. The bibliographies of selected articles were screened for potentially suitable references, which were then retrieved. We also searched the EMBASE and Cochrane databases using the same search strategy. Study Selection Studies were included in the meta-analysis if they met the following criteria: 1) intra-arterial x-ray angiography was used as the gold standard; 2) clinical suspicion of renovascular hypertension was the reason for referral; 3) the criteria and cutoff values for a positive result for each test [for example, hemodynamically significant stenosis] were explicitly defined; and 4) the absolute numbers of true-positive, false-negative, false-positive, and true-negative results were available or could be derived from the data presented. Two investigators screened the full text of potentially relevant articles using the inclusion criteria. In all cases of disagreement, the investigators reached consensus. Studies that included patients who had received renal transplants were excluded. Another reason for exclusion was the possibility of verification bias, which appears when patients with a positive result on the index test (that is, the test under investigation) are referred to the gold standard test more often than patients with a negative result on the index test. The presence of verification bias was assumed if only some of the patients who had the index test were referred for intra-arterial x-ray angiography. If an author or a research group published more than one study about the same diagnostic test, the most recent study was included. All articles by the same author or research group were included for analysis only when it was obvious that a different patient sample had been used. In cases where an author or research group discussed more than one diagnostic technique within one journal article (for example, MRA vs. ultrasonography, or two-dimensional time-of-flight MRA vs. three-dimensional time-of-flight MRA), each modality was considered separately. Data Extraction We used a standard form to extract relevant data from the included articles. Several studies reported more than one pair of sensitivity and specificity estimates because they used several approaches to analyze data on accuracy. These approaches included inclusion compared with exclusion of the accessory renal arteries in the analysis; estimation of sensitivity and specificity using renal arteries, kidneys, or patients as the unit of analysis; exclusion compared with inclusion of missing observations (such as technical failures, poor examination quality, or arteries not identified at the index test); and presentation of sensitivity and specificity estimates based on evaluation by more than one reviewer. We preferred sensitivity and specificity estimates based on inclusion of accessory renal arteries, on patients as unit of analysis, or on inclusion of missing observations in the analysis. If a study presented results based on more than one reviewer, the accuracy data representing the highest estimates were extracted. For each ultrasonography study, we noted whether measurements were performed at the level of the intrarenal arteries or at the level of the extrarenal arteries. For captopril renal scintigraphy studies, we noted whether the analysis was based on changes between baseline (precaptopril) and postcaptopril measurements or only on postcaptopril measurements. When captopril renal scintigraphy studies presented estimates based on evaluation of postcaptopril renography and captopril-induced changes, the highest estimates were extracted. Data Synthesis We based our meta-analysis method on the construction of summary receiver-operating characteristic (ROC) curves. This method accounts for the mutual dependence between sensitivity and specificity; evaluates the extent to which the variation in sensitivity and specificity can be explained by variation in positivity thresholds among studies; makes it possible to assess the effects of other between-study differences, such as differences in test design and analysis characteristics, on the estimates of diagnostic accuracy; and is very useful in comparing different diagnostic tests (8, 9). Areas under the summary ROC curves were used as a measure of the diagnostic performance of the tests. We calculated a diagnostic odds ratio on the basis of the sensitivity and specificity reported in each study and then derived areas under the ROC curves for each study. We used linear regression analyses to compare the diagnostic tests. In these analyses, indicator variables represented the tests and the presence or absence of specific study characteristics. The coefficients resulting from the model were translated into areas under the summary ROC curves, differences between areas under the summary ROC curves, and 95% CIs of these differences. Details of the statistical analysis are discussed in the Appendix. All statistical calculations were performed by using SPSS (SPSS, Inc., Chicago, Illinois). Results Included Studies We identified 343, 306, 314, 172, and 172 studies for CTA, MRA, ultrasonography, captopril renal scintigraphy, and the captopril test, respectively. We retrieved the full text of 22, 39, 58, 25, and 13 studies. Five of 22 studies involving CTA (10-14), 16 of 39 studies involving MRA (15-30), 24 of 58 studies involving ultrasonography (7, 10, 18, 29, 31-50), 14 of 25 studies involving captopril renal scintigraphy (7, 16, 35, 37, 51-60), and 4 of 13 studies involving the captopril test (59, 61-63) met the inclusion criteria. Subsequent searches on the EMBASE and Cochrane databases did not reveal any additional references. A list of all studies that we excluded after reading the full text, as well as the reasons for exclusion, is available from the authors or at www.unimaas.nl/~radiolog/radish.html/. Ten journal articles studied more than one diagnostic modality (7, 10, 15, 16, 18, 21, 29, 35, 37, 59). In the MRA articles, a distinction was made between techniques. Studies evaluating gadolinium-enhanced MRA (n = 6) were combined and compared with studies evaluating nongadolinium-enhanced MRA, representing the two-dimensional and three-dimensional time-of-flight and the phase-contrast techniques (n = 12). In the ultrasonography studies, a distinction was made between studies performed at the level of the extrarenal arteries (n = 13) and studies performed at the level of the intrarenal arteries (n = 7); 4 studies performed measurements at both levels. Data Extraction and Synthesis We extracted the evaluated technique, the year of publication, test characteristics, patient characteristics, design characteristics, and results from the 55 included articles (evaluating 65 modalities); this information is partially summarized in the Table. The summary ROC curves for the compared diagnostic modalities are shown in Figure 1. The summary ROC curve representing the CTA studies and the summary ROC curve representing the gadolinium-enhanced MRA studies pass through the upper left quadrant of the ROC area, which indicates that high sensitivity could be reached at low false-positive rates. Figure 1 also shows that the captopril test has the lowest diagnostic performance. The areas under the ROC curves for the diagnostic modalities were 0.99 for CTA, 0.99 f