EM algorithm for comparing two binary diagnostic tests when not all the patients are verified

• Published in: Journal of statistical computation and simulation 2008-01, Vol.78 (1), p.19-35
• Main Authors: Roldan Nofuentes, J. A., Luna Del Castillo, J. D.
• Format: Article
• Language: English
• Subjects: EM and SEM algorithms; Partial verification; Predictive values; Sensitivity; Specificity; Verification bias
• ISSN: 0094-9655; 1563-5163
• DOI: 10.1080/10629360600938102

The comparison of the efficiency of two binary diagnostic tests is one of the most interesting problems in the study of the methods of diagnosis of disease. In the presence of partial disease verification, in the case of some patients, the true disease status is not known and, therefore, comparison of the accuracy of the two binary diagnostic tests cannot be carried out through the traditional method. In this study, an EM algorithm is proposed to compute the values of the maximum likelihood estimators of sensitivity, specificity and the predictive values of the diagnostic tests. Hypothesis tests have been deduced to compare these measurements in two diagnostic tests, estimating the variances-covariances by applying the SEM algorithm and the delta method. Simulation experiments have been carried out to study the asymptotic behaviour of the hypothesis tests obtained. The results have been applied to the diagnosis of coronary stenosis.