The concepts sensitivity and specificity are critically evaluated in the light of case studies drawn from electrocardiographic literature. These terms are often misused, and the meaning of these elementary statistical concepts is often misunderstood in studies on ECG, VCG and PCG criteria. Specificity figures reported in literature commonly refer to the fraction of true negatives with a negative test in normals. Limiting the test to two-group analysis and eliminating other disease categories tend to give overoptimistic values for specificity and diagnostic accuracy in general. It is pointed out that various performance indices for diagnostic accuracy depend heavily on the composition of the test groups and the fraction of test cases in each group. Sensitivity and specificity appear inadequate performance indices for evaluation of event detection schemes such as classification of ectopic beats. Alternative performance indices are considered, including the error ratio, association index, accuracy of positive test, accuracy of negative test, and overall diagnostic accuracy. Increased utilization of simple statistical tests for significance estimation in ECG criteria evaluation is suggested. The development and application of better diagnostic performance evaluation schemes based on concepts of cost of misclassification, entropy and information is encouraged.
Bibliographical noteFunding Information:
From the Biophysics and Bioengineering Research Laboratory, Faculty of Medicine, Dalhousie University, Halifax, Nova Scotia, and the Laboratory of Physiological Hygiene, School of Public Health, University of Minnesota, Minneapolis, Minnesota. Supported in part by research grants from the Medical Research Council of Canada (MT-2828) and the Nova Scotia Heart Foundation. Reprint requests to: P. M Rautaharju, M.D., Ph.D., Sir Charles Tupper Medical Building, Dalhousie University, Halifax, Nova Scotia B3H 4H7, Canada.