Semiparametric and nonparametric modeling of diagnostic data
Year: 2012
Type: dissertation
Abstract: Diagnostic tests play an important role in health care and the statistical evaluation of their accuracy is imperative before they are used in practice. The receiver operating characteristic (ROC) curve is the most widely used measure to evaluate the discriminatory performance of a continuous diagnostic test. In some diagnostic situations, covariate information that affects the test's performance is also available and this additional information should be taken into account when evaluating the accuracy of the test. When there are more than two possible disease status, ROC curves give rise to ROC surfaces. The statistical analysis of diagnostic data has traditionally used parametric methods. Nonparametric and semiparametric approaches are advantegeous because they provide exible and robust inferences. In this thesis we developed nonparametric and semiparametric estimators as well as new methodologies for the evaluation of continuous diagnostic tests. In the first part of this thesis, we developed a exible and robust Bayesian nonparametric approach based on mixtures of finite Polya trees priors to estimate the ROC surface. We thus relaxed the strong distributional assumptions of the existing approaches. In the second part, we proposed a Bayesian nonparametric ROC regression estimator based on dependent Dirichlet processes, which allows for modeling directly the entire conditional distribution in the healthy and diseased populations. This model also easily accomodates multiple predictors, either categorical or continuous. In the third part of the thesis, we have developed ROC regression methodology for the case where the covariate is functional, rather than univariate or multivariate. To this end, semiparametric and nonparametric ROC regression estimators were proposed. A large number of simulations and example analysis illustrate the performance of the proposed estimators.
Author: Vanda Calhau Fernandes Inácio
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2229218927