Matchmaking and McNemar in the comparison of diagnostic modalities

Abstract

Comparative studies of radiologic techniques commonly yield matched data due to the ease and desirability of performing all of the techniques on each of the patients. The two-by-two matched data array and the McNemar analysis provide a succinct format for the presentation and proper analysis of matched comparisons of binary (positive and negative) test results. When comparing tests, it is essential not to rely on just the statistical significance of the differences in sensitivities (or specificities); the magnitude of the differences must also be assessed. Confidence intervals provide a useful form of estimation by providing a range of plausible values and an indication of the precision of the estimate. The matched data array also indicates the complementarity of the diagnostic tests being compared.