Biostatistics in clinical decision making for cardiothoracic radiologists

Abstract

Cardiothoracic radiologists are intuitively aware of sensitivity and specificity as they pertain to diagnostic tests involving clinical information. However, many cardiothoracic radiologists are unfamiliar with odds ratios, likelihood ratios, predictive values, and receiver operating characteristic (ROC) curves, which provide more information about the performance of a test. Our article will first review the fundamental concepts of sensitivity, specificity, predictive values, and likelihood ratios. The ROC curve methodology will be covered with an emphasis on creating a look-up table, a straightforward table that communicates important information to the clinician to aid in diagnosis. The article reviews sensitivity and specificity, as well as predictive values, logistic regression, and ROC curves, using conceptual principles without unnecessary mathematical rigor. We will apply principles of sensitivity and specificity to continuous measurements by constructing ROC curves in order to tie together key ideas in diagnostic decision making. Three clinical examples are presented to illustrate these fundamental statistical concepts: predictors of pulmonary embolism in children, use of dobutamine-cardiac magnetic resonance imaging to identify impaired ventricular function in patients who have suffered a myocardial infarction, and diagnostic accuracy of 64-multidetector row computed tomography to identify occluded vessels in adult patients with suspected coronary artery disease. In addition, a glossary is provided at the end of the article with key terms important in diagnostic imaging. An understanding of the concepts presented will assist cardiothoracic radiologists in critically discerning the usefulness of diagnostic tests and how these statistics can be applied to make judgments and decisions that are essential to clinical practice.