Evaluating diagnostic tests

Anaesthesiologists are increasingly more involved in perioperative patient care wherein interpretation of special investigations is crucial to making therapeutic and prognostic decisions. Furthermore, anaesthetic journal publications increasingly rely on diagnostic tests, without paying sufficient attention to the methodology for evaluation of the predictive ability of these tests, particularly when conducting research. The purpose of this article is to provide an introduction to the principles underlying the objective appraisal of diagnostic tests and to provide basic tools for that purpose. Basic concepts about probability and conditional probability are introduced. An explanation is given how Bayes’ theorem can be used to apply new information to improve uncertainty about a diagnosis. Various indices of diagnostic accuracy are clarified using simple probabilistic notation applied to a 2x2 table from which they can be calculated. Two clinical examples are employed to illustrate these concepts. The calculations are made easier using a spreadsheet that may be downloaded from the Internet and by using Fagan’s nomogram. The use of receiver operating characteristic (ROC) curves for choosing a suitable cut-off point between positive and negative tests is explained.