Department of Biostatistics

People

Prospective Students

Academics

Research

News & Events

Calendar

Consulting

Employment Opportunities

Resource Quicklinks

Computing Environment

Contact

THESIS DEFENSE ABSTRACT

Analysis of Differentiated Risk Factor Associations with Multiple Outcomes

 Hongfei Guo PhD Candidate, Johns Hopkins Department of Biostatistics

Multiple outcomes are commonly used to characterize the effect of a treatment or exposure in medical research, for instance, multiple symptoms to characterize potential remission of a psychiatric disorder. Often a global, i.e., symptom-invariant, treatment effect is evaluated. Such a treatment effect may overgeneralize the effect across the outcomes. On the other hand individual treatment effects, varying across all outcomes, are complicated to interpret, and their estimation may lose precision relative to a global summary. An intermediate and potentially more effective way to summarize the treatment effect may be through patterns of treatment effects, i.e., "differentiated effects." In this dissertation, we propose a multi-category effects model to differentiate treatment effects into a few groups for continuously scaled multiple outcomes. In our approach, outcome-specific treatment effects are assumed to partition according to follow mutually independent multinomial distributions, so that the outcomes are differentiated into a few groups by their strengths of association with the treatment or exposure. We develop a feasible MCEM fitting algorithm to obtain maximum likelihood estimation of the parameters.

First, we introduce a two-category effects model to differentiate treatment effects into more, or less, effectively treated groups. The model specification, parameter interpretations, model fitting algorithm and inference are described in detail. The two-category analysis is evaluated through a simulation study and illustrated by an analysis of schizophrenia symptom data. Second, we extend our model to allow more than two categories by generalizing the assumption of random effects. We also evaluate the performance of information criteria for determining the number of categories. We apply the multi-category effects model to an analysis of the association between physiological dysregulation, characterized by multiple biomediators, and frailty in older adults. We aim to identify a few groups of biomediators, each of whose means differ similarly across frail and non-frail populations, and estimate the magnitude of mean difference per biomediator group. Using data from the Women's Health and Aging Study (WHAS), we contrast our methodology with two alternative methods: summarizing the biomediators into a few scores and then applying multivariate regression, and analyzing the biomediators using multivariate regression and then summarizing the coefficients. Finally, we evaluate procedures to describe and test for heterogeneity in treatment effects on multiple outcomes. Graphical and quantitative methods are adopted for distinguishing homogeneity of effects from heterogeneity, and item-wise normally distributed treatment effects from non-normal heterogeneity. We report a Monte Carlo simulation to compare the size and power of different testing procedures. Recommendations for appropriate test procedures for multi-category treatment effects on multiple outcomes are given.