Statistical Evaluation of Diagnostic Tests: A Graphical Approach

Authors: Ergun KARAAĞAOĞLU

Abstract: Although sensitivity, specificity, false positive and false negative rates are very valuable measures for showing the performance of diagnostic tests, their usefulness is restricted by the fact that these test operating characteristics are subject to change as the cut off point changes. Therefore methods, which enable us to evaluate the overal performence of diagnostic tests are required. Two graphical methods; plotting posterior probabilities versus prior probabilities and ROC curves are introduced. These two methods not only help us to evaluate the overall performance of diagnostic tests, but also help us to compare many diagnostic tests at some defined interval of possible test results. ROC curves have some additional advantages, such as showing the positive likelihood ratio at each possible test result or cut off point. By means of this ratio one can evaluate the cost of falsely labeling a subjects as diseased and the benefit or correctly diagnsosing a subject as diseased.

Keywords: Sensitivity, specificity, ROC curves, positive likelihood ratio