How to Analyze the Diagnostic Performance of a New Test? Explained with Illustrations

Diagnostic tests are pivotal in modern medicine due to their applications in statistical decision-making regarding confirming or ruling out the presence of a disease in patients. In this regard, sensitivity and specificity are two most important and widely utilized components that measure the inherent validity of a diagnostic test for dichotomous outcomes against a gold standard test. Other diagnostic indices like positive predictive value, negative predictive value, positive likelihood ratio, negative likelihood ratio, accuracy of a diagnostic test, and the effect of prevalence on various diagnostic indices have also been discussed. We have tried to present the performance of a classification model at all classification thresholds by reviewing the receiver operating characteristic (ROC) curve and the depiction of the tradeoff between sensitivity and (1-specificity) across a series of cutoff points when the diagnostic test is on a continuous scale. The area under the ROC (AUROC) and comparison of AUROCs of different tests have also been discussed. Reliability of a test is defined in terms of the repeatability of the test such that the test gives consistent results when repeated more than once on the same individual or material, under the same conditions. In this article, we have presented the calculation of kappa coefficient, which is the simplest way of finding the agreement between two observers by calculating the overall percentage of agreement. When the prevalence of disease in the population is low, prospective study becomes increasingly difficult to handle through the conventional design. Hence, we chose to describe three more designs along with the conventional one and presented the sensitivity and specificity calculations for those designs. We tried to offer some guidance in choosing the best possible design among these four designs, depending on a number of factors. The ultimate aim of this article is to provide the basic conceptual framework and interpretation of various diagnostic test indices, ROC analysis, comparison of diagnostic accuracy of different tests, and the reliability of a test so that the clinicians can use it effectively. Several R packages, as mentioned in this article, can prove handy during quantitative synthesis of clinical data related to diagnostic tests.