Evaluating Diagnostics in the Absence of a Gold Standard

In 2004, Drs. Dodd and Albert published a paper on potential problems from estimating the diagnostic error of binary tests without a gold standard using latent class modeling. They showed that these approaches are sensitive to the dependence structure between tests, yet it is generally nearly impossible to distinguish between competing models. In a follow-up paper, they examine the robustness of the estimation procedures when, in a fraction of cases, we observe the gold standard test. They propose semi-latent modeling approaches for this problem and show that, even with a small percentage of gold standard information, estimates of diagnostic error are insensitive to the assumed dependence structure between tests.

Albert PS, Dodd LA. Cautionary note on the robustness of latent class models for estimating diagnostic error without a gold standard. Biometrics 2004:60;427–35.

Albert PS, Dodd L. On estimating diagnostic accuracy from studies with multiple raters and partial gold standard evaluation. In revision at J Am Stat Assoc.

Albert PS. An imputation approach for estimating diagnostic accuracy from partially verified designs. Submitted to Biometrics.

Albert PS. Misclassification models. In: Encyclopedia of Biostatistics. 2nd ed. Armitage P, Colton T, eds. New York: John Wiley & Sons; 2005.