Evidence in Diagnostic Imaging: Going Beyond Accuracy

Abstract

Because of the increasing concern about its excessive and inappropriate use, guidelines have been developed for the appropriate use of diagnostic imaging [1-3]. Guidelines for diagnostic imaging are also incorporated into more comprehensive clinical practice guidelines [4,5]. Guidelines should be evidence based [6], and in assessing the evidence, accuracy is usually the primary criterion used to evaluate the role of diagnostic imaging [7,8]. Unfortunately, the quality of the evidence is usually given a lower grade [4,5], and the methodology for assessing the evidence in imaging guidelines has been criticized [9]. In fact, the type of evidence needed for imaging guidelines needs to be reassessed. Accuracy is obviously a very important attribute of effective diagnostic imaging. A study designed to test the accuracy of diagnostic imaging must be methodologically rigorous to avoid bias, and the methodology must be clearly described in any report [10]. The ost difficult issue in the design of study of the accuracy of a diagnosic imaging procedure is the selecion of the reference or “gold” stanard against which the accuracy of he diagnostic test is measured. The est reference standard is usually athology. In the ideal world, all he patients in a study should be ssessed by the same reference stanard. However, in practice, this is irtually never possible. In assessing he accuracy of diagnostic imaging, atients in a study who are thought o have the condition being assessed ay undergo biopsy or some other ype of surgical exploration, and the ccuracy of the test for those patients can be assessed against the reference standard of pathology. However, patients who do not have any evidence of the disease on their imaging studies cannot ethically undergo biopsy and are therefore assessed using a different reference standard, usually follow-up. This results in what is called verification bias, and verification bias can result in overestimation of both sensitivity and specificity [10]. If 2 diagnostic imaging examinations are being compared, and the same reference standards are being used for both, verification bias may be less of a problem. There are also statistical methods available that can help correct for verification bias [11]. However, if the true accuracy of a diagnostic imaging study is being investigated, verification bias is a significant and usually unavoidable problem. Defining a reference standard for assessing the accuracy of imaging in the diagnosis of conditions that are not usually or cannot be confirmed pathologically becomes even more difficult. For instance, radiography has been used to diagnose fractures for more than a century [12]. We do know certain situations in which the diagnostic accuracy of an x-ray is limited, for instance, in the diagnosis of fractures of the femoral neck in the elderly [13], but there have been no large-scale studies to determine the overall accuracy of x-rays in the diagnosis of fractures. There are other conditions in which the establishment of a reference standard is extremely difficult or controversial. A very common indication for an x-ray of the chest is the diagnosis of pneumonia. However, we do not know the ac-