Analysis of Data with Complex Misclassification in Response or Predictor Variables by Incorporating Validation Subsampling

The problems of misclassification are common in epidemiological and clinical research. Misclassification may be present in either an exposure or outcome variable, or both. It is well known that the validity of analytic results (e.g., estimates of odds ratios of interest) might be questionable when no correction effort is made. Therefore, valid and accessible methods with which to deal with these issues are still in high demand.;In this dissertation, we first consider the situation when correlated binary response variables are subject to misclassification. Building upon prior work that extended McNemar's test to correct paired-data odds ratio estimation, we propose a nonlinear mixed model-based approach to adjust for potentially complex differential misclassification in correlated binary responses via internal validation sampling.;In the second topic, we shift gears toward predictor misclassification, for which we develop likelihood-based approaches based on generalized linear and generalized linear mixed models that can efficiently incorporate internal validation data in univariate and multivariate settings, respectively. We discuss the use of the approach both in the case when a baseline predictor is misclassified and when a time-dependent predictor is misclassified.;In the final topic, we elucidate extensions of well-studied methods in order to facilitate misclassification adjustment when a binary outcome and binary exposure variable are both subject to complex differential misclassification in the 2-by-2 table scenario. We develop maximum likelihood approaches to accommodate a broad range of complexity in the joint misclassification process while incorporating various types of internal validation observations. We then generalize the method to a more standard binary regression setting, allowing the incorporation of covariates both in the main health effects model of interest and in misclassification models for both the binary outcome and exposure variable. Throughout, illustrative examples are presented via detailed analyses of bacterial vaginosis and trichomoniasis data from the HIV Research Epidemiology Study (HERS).;Key Words: Differential; Misclassification; Internal Validation; Likelihood.