| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Editorial |
1 From the Department of Radiology, Dartmouth-Hitchcock Medical Center, 1 Medical Center Dr, Lebanon, NH 03756; and the Center for the Evaluative Clinical Sciences, Department of Community and Family Medicine, Dartmouth Medical School, Hanover, NH. Received August 1, 2000; accepted August 10. Address correspondence to the author (e-mail: William.Black@Hitchcock.org).
Index terms: Aneurysm, intracranial, 17.73 • Computed tomography (CT), angiography, 17.12116 • Digital subtraction angiography, comparative studies, 17.12483 • Editorials • Magnetic resonance (MR), vascular studies, 17.12142 • Ultrasound (US), Doppler studies, 17.12983, 17.12984
Several guidelines for the systematic review of diagnostic accuracy have been developed (1). The diagnostic examinations, clinical context, and reference standard(s) used should be clearly defined. The literature retrieval procedure and inclusion and exclusion criteria should be clearly described. Data should be extracted by two or more readers. Diagnostic accuracy should be estimated by using a method that accounts for the interdependence of sensitivity and specificity, such as the summary receiver operating characteristic curve (2). Finally, the effects of variation in study design and patient and examination characteristics on estimates of diagnostic accuracy should be assessed.
In this issue of Radiology, White and colleagues (3) systematically review the accuracy of noninvasive imaging examinations in the detection of intracranial aneurysms. Their review includes 38 studies involving the use of computed tomographic (CT) angiography, magnetic resonance (MR) angiography, and transcranial Doppler ultrasonography (US) in 1,765 patients. In all the studies, intraarterial digital subtraction angiography was used as the reference standard. Two of the authors independently extracted data from the articles, and the overall accuracy of CT angiography and MR angiography were expressed as summary receiver operating characteristic curves.
In addition, the authors assessed the effects of several variables that can affect the estimation of accuracy. These variables included the date of publication (before and after 1995), sample size (less than 50 subjects and greater than or equal to 50 subjects), aneurysm location (anterior and posterior circulation), and aneurysm size (less than or equal to 3 mm, greater than 3 mm, greater than 3 mm but less than 10 mm, and greater than 10 mm). Because the authors closely followed the recommended guidelines for systematic reviews (1), the reader can place a high level of confidence in their main finding that CT angiography and MR angiography have comparable accuracy in the detection of intracranial aneurysms.
The authors also found that sensitivity was much greater for the detection of large aneurysms (
10 mm) than for the detection of small aneurysms (
3 mm). For large aneurysms, the sensitivity of CT angiography and MR angiography were 100% and 99%, respectively; for small aneurysms, the sensitivity of CT angiography and MR angiography were only 61% and 38%, respectively. Although it may seem obvious that sensitivity will increase with the size of the aneurysm, the anatomic extent of disease is often neglected in reports of diagnostic accuracy (4,5), as well as in reports of disease incidence and prevalence. In fact, in the article in which the highest prevalence of intracranial aneurysms among asymptomatic adult patients with polycystic kidney disease (41%) was reported, there was no indication of aneurysm size (6). Such neglect to report the anatomic extent of disease has profound implications to our understanding of diseases and testing (5,7) that are not so obvious as the fact that large intracranial aneurysms are easier to detect than small intracranial aneurysms.
When comparisons of accuracy are not controlled for anatomic extent, real differences in accuracy may be obscured or spurious differences may appear. Although this problem can be avoided by using the same subjects—and a uniform reference standard—for all the examinations being compared, there are ethical, financial, and logistic limitations to this approach. In addition, one should be able to compare the examinations evaluated in studies across location and time. For these reasons, the accuracy of imaging examinations should always be stratified according to the anatomic extent of disease, ideally by using standard dimensional units such as centimeters (5,7).
The reason that stratification by anatomic extent is so important is illustrated in two reports on the accuracy of imaging examinations for detecting liver metastases. In 1982, sulphur colloid liver scintigraphy was reported to have a sensitivity and specificity of 96% and 98%, respectively (8). Nine years later, CT was reported to have a sensitivity and specificity of only 72% and 95%, respectively (9). The apparently lower sensitivity and specificity values of CT, however, can be easily explained by the fact that the surgeons in the latter study examined the liver more closely, with the aid of intraoperative US, and found many more small metastases than did the surgeons in the former study. In fact, the modal size of liver metastases in the 1991 study was between 1 and 2 cm, whereas the modal size in the 1982 study was probably around 4 cm (although this was not actually reported).
Consideration of the anatomic extent of disease is also highly relevant to the interpretation of examination results. According to the Bayes theorem, the posttest probability of disease is a function of the pretest probability and the accuracy of the examination. As already pointed out, sensitivity generally increases with the anatomic extent of disease. In addition, specificity also tends to increase, because large false-positive abnormalities are generally less common than small false-positive abnormalities. The relationship between pretest probability and anatomic extent is more complex. In symptomatic individuals, the pretest probability is usually high and the disease that is present tends to be severe. In the systematic review performed by White et al (3), which was based mainly on data in symptomatic patients, the prevalence of intracranial aneurysms was about 80%, and about 80% of the intracranial aneurysms were larger than 3 mm. In asymptomatic individuals, however, the pretest probability of disease usually is low and the disease that is present tends to be mild. In a study of adult patients with polycystic kidney disease who were being screened by means of MR angiography (10), the prevalence of intracranial aneurysms was only 11.7%, and less than one-third of the intracranial aneurysms were larger than 3 mm.
These relationships between pretest probability, accuracy, and anatomic extent can lead to an apparent violation of the Bayes theorem, which states that the posttest probability of disease increases with the pretest probability (11). For example, with a negative CT angiogram, the posttest probability of a small aneurysm in an asymptomatic individual with a pretest probability of 10% would be about 5%, on the basis of a sensitivity and specificity of 61% and 86%, respectively, for small intracranial aneurysms (3) and the dichotomous form of the Bayes theorem. However, the posttest probability of a large aneurysm in a symptomatic patient with a pretest probability of 80% would be 0% on the basis of a sensitivity of 100% for large intracranial aneurysms (3). Thus, if small and large intracranial aneurysms are not distinguished, then it would seem that the posttest probability of an intracranial aneurysm will be lower for those patients with a higher pretest probability. (A more precise quantification of probabilities can be achieved by stratifying the pretest probability and accuracy according to anatomic extent [5,12] and applying the generalized form of the Bayes theorem rather than the dichotomous form.)
Finally, consideration of anatomic extent is highly relevant to decisions about future testing strategies, especially screening. As pointed out by White et al (3), screening populations have a lower pretest probability of disease than do symptomatic populations, and those screening-eligible individuals who have disease tend to have milder forms of it. Consequently, the positive predictive value will usually be much lower in the screening setting. In addition, it must be remembered that early diagnosis is a "double-edged sword" (13) and should not be assumed to be beneficial with any disease, including intracranial aneurysm (14). Ideally, the effectiveness of a screening intervention should be established in a randomized clinical trial before the screening is incorporated into standard medical practice (15). However, decision modeling can also play a role in designing a randomized clinical trial and extrapolating the results of a randomized clinical trial to other populations or screening methods (16). Modeling the natural history of disease and the effects of screening requires data on disease prevalence and test accuracy, both of which need to be stratified according to anatomic extent of disease (14,17).
In summary, White et al (3) provide a sound systematic review on the accuracy of noninvasive imaging examinations in the detection of intracranial aneurysms, which reveals that CT angiography and MR angiography have comparable accuracy and that the sensitivities of these modalities are much higher for the detection of large rather than small intracranial aneurysms. Their review serves also as a reminder that the accuracy of imaging examinations should always be stratified according to the anatomic extent of disease and clearly reported in this manner.
FOOTNOTES
See also the article by White et al (pp 361–370 ) in this issue.
REFERENCES
Related Article
This article has been cited by other articles:
|
|
 |
|
  P MacEneaney Re: Communication of doubt and uncertainty in radiological reports Br. J. Radiol., December 1, 2002; 75(900): 1005 - 1005. [Full Text] [PDF]
|
|
|
|
 |
|
 T. The Evidence-Based Radiology Working Group Evidence-based Radiology: A New Approach to the Practice of Radiology Radiology, September 1, 2001; 220(3): 566 - 575. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. M. White, J. M. Wardlaw, E. Teasdale, S. Sloss, J. Cannon, and V. Easton Power Transcranial Doppler Ultrasound in the Detection of Intracranial Aneurysms Stroke, June 1, 2001; 32(6): 1291 - 1297. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
