| In this dissertation; we focus on the statistical analysis of diagnostic accuracy with correlated data. In Chapter 1, we review the literature on the related area. In Chapter 2, we propose a nonparametric approach for comparing diagnostic accuracies in multi-rater Receiver Operating Characteristic (ROC) studies. The approach constructs a test statistic froze each rater by extending the conventional nonparametric method and then combines all the individual test statistics to draw an overall conclusion on the relative accuracies of different diagnostic tests. The method can handle both continuous and ordinal data. Furthermore, the choice of the optimal weights in combining the individual test statistics is discussed. Compared to the existing nonparametric methods, the proposed method is robust and effectively deals with the possible heterogeneity among raters. The method is applied to a real example and its finite sample performance is examined through simulation studies.; In Chapter 3, a nonparametric approach for evaluating the diagnostic accuracy of a monitoring test has been proposed. In a monitoring program, patients are examined periodically with monitoring tests until the onset of the designated condition or the termination of the program. Therefore, different patients may contribute different numbers of measurements of a monitoring marker. The proposed method provides optimal estimates of diagnostic accuracy index in the monitoring test setting. The asymptotic properties of the estimates are discussed. Simulation studies are conducted to examine the finite sample performance of the proposed estimates. Future research opportunities are discussed in Chapter 4. Examples include the thoughts of improving ROC regression models. |
