Diagnostic Accuracy of a Binary Test in the Presence of Two Types of Missing Values

Date of Completion

January 2011


Statistics|Health Sciences, Epidemiology




This thesis compares several methods for estimating the sensitivity and specificity of a binary diagnostic test in the presence of missing data. Naïve estimation, Begg-Greenes, Logit Begg-Greenes, expectation maximization, and markov chain monte carlo estimation are compared to multiple imputation under five different nonignorable imputation models. This work is motivated by an Alzheimer Disease example. We estimate test accuracy for the examples, and produce a simulation study to determine which (if any) method is better. We also consider test accuracy when data are missing for two reasons, and when there is an additional observed covariate. The Multiple imputation methods whose imputation model matched the simulation model performed pretty well, but did not do well for misspecified models. The methods gave very different estimates. ^