Empirical likelihood confidence regions for the evaluation of continuous‐scale diagnostic tests in the presence of verification bias

Abstract

AbstractIn a continuous‐scale diagnostic test, when a cut‐off level is given, the performance of the test in distinguishing diseased subjects from non‐diseased subjects can be evaluated by its sensitivity and specificity. Joint inferences for sensitivity and specificity as well as cut‐off level play an important role in the assessment of the diagnostic accuracy of the test. Most current studies on this topic focus on complete data cases. However, in some studies, only a portion of subjects given their screening test results ultimately have their true disease status verified. In addition, the verification may depend on the test result and the subject's observed characteristics. Directly applying full data methods to verified subjects results in biased estimates, known as verification bias. In this paper, based on a general framework that combines empirical likelihood and general estimation equations with nuisance parameters, we propose various bias‐corrected joint empirical likelihood confidence regions for sensitivity and specificity with verification‐biased data. Thorough simulation studies are conducted to compare the finite sample performance of the proposed confidence regions in terms of coverage probabilities, and some suggestions are provided accordingly. Finally, an example is provided to illustrate the proposed methods. The Canadian Journal of Statistics 41: 398–420; 2013 © 2013 Statistical Society of Canada