The assessment of diagnostic tests

Abstract

To determine whether improvements have occurred since a survey of the 1982 literature assessing diagnostic tests, the authors evaluated all English-language articles that assessed clinical diagnostic tests in abridged Index Medicus journals in 1985, and that had the terms sensitivity and specificity in the title, abstract, or key words. The 89 articles were assessed against seven methodologic criteria, including use of a well-defined “gold standard,” clearly defined test interpretation, blinding, clear data presentation, correct use of sensitivity and specificity, calculation of predictive values, and consideration of prevalence. In comparisons of 1985 vs. 1982 articles, there were significant improvements in five of the seven criteria. For example, the proportion of articles using a well-defined “gold standard” rose from 68% to 88%. Overall, the frequency of papers demonstrating five or more of the seven criteria increased from 26% to 47%. However, predictive values were discussed in only 54% of the articles without, necessarily, consideration of the influence of prevalence as well. This study raises the concern that while the concepts of sensitivity and specificity are now accepted, predictive values remain less well understood. Although there has been an improvement in the assessment of diagnostic tests in published research, attention to accepted methodologic standards is still needed on the part of researchers, reviewers, and editors.