THESIS DEFENSE ABSTRACT
Modeling Composite Outcome and Jointly Modeling Its Components
Xianbin
Li, PhD Candidate, Johns Hopkins Department of Biostatistics
Composite outcomes defined by logical (Boolean) operations on
mixed original outcomes arise often in biomedical research. For example,
hypertension is often defined as a systolic blood pressure greater than or equal
to 140 mmHg, a diastolic blood pressure greater than or equal to 90 mmHg, or the
use of antihypertensive medication. When there are no missing values in the
original outcomes, the estimation of the proportion of successes from a
composite outcome is straightforward; however, when there are missing values in
the original outcomes, the estimation is less clear and common estimators can be
biased, even if the missingness is completely at random. Motivated from the
study of hypertension, we propose estimators of prevalence, methods of joint
regression modeling of continuous and binary outcomes, and conduct a fully
Bayesian longitudinal analyses of these outcomes. This dissertation comprises
three distinct papers. In the first paper, the logically defined outcome
(composite outcome) is defined and four estimators of the prevalence are
proposed and compared. The maximum likelihood estimator, using all available
data, is shown to be consistent and efficient while the naive estimator is
arbitrarily biased. In the second paper, we jointly model two continuous
outcomes and one binary outcome using shared random effects (intercept) models
with a probit model for the binary outcome and propose the use of the propensity
score as a way to balance confounding variables in order to obtain the
proportion of successes of the composite outcome associated with covariates. In
the third paper, Bayesian joint modeling of longitudinal continuous and binary
outcomes is proposed to analyze a novel hypertension data set and Markov chain
Monte Carlo algorithms are used to sample from the posterior distributions of
parameters. The proposed statistical methods for composite outcome and its
components in this dissertation perform reasonably well and can be used in many
epidemiologic studies and clinical trials with continuous and binary outcomes.
Return to
Biostatistics Calendar
|
Return
to Home Page
|
