Peculiarities of diagnostic test accuracy studies

Abstract

The performance of a diagnostic test is always appraised in comparison to one or several other competing tests, which may be other index or reference tests. The continuum in results of such tests is typically simplified using one or more threshold, eventually leading to a contingency table (e.g., a 2×2 table). Such format is the formal basis for the computation of several dimensions of comparative diagnostic accuracy, from sensitivity, to specificity, predictive values, likelihood ratios, and area under the curve of receiver operating curve, among the others. This is at odds with incidence studies, prognostic research, and controlled trials, which instead lead to more common measures of effect such as odds ratios, relative risks, and hazard ratios, among the others. Sources of bias and threats to the internal and external validity of diagnostic test accuracy studies also differ substantially from methodological issues pertinent to incidence studies, prognostic research, and controlled trials. The key design features, the complex interplay between index and reference test, and the theoretical distance between test results and clinical outcomes characterize the key peculiarities of diagnostic test accuracy studies. The present chapter provides an overview of design and analytical aspects of such studies, including a sample dataset and computing code for detailed and comprehensive analyses. Thorough knowledge and mastery of such peculiarities is crucial before facing a plethora of similar studies on diagnostic test accuracy when envisioning a systematic review and meta-analysis of the scholarly literature.