Title

A comparison of propensity score and pattern-mixture models for non-random attrition with CD4 cell count decline in HIV infection

Date of Award

2000

Availability

Article

Degree Name

Doctor of Philosophy (Ph.D.)

Department

Epidemiology and Public Health

First Committee Member

Maria M. Llabre, Committee Chair

Abstract

The use of a "growth" curve framework to model a chronic disease process is complicated when the probability of continuing to observe that process is a function of the outcome being modeled. A propensity score model for non-ignorable missing data in a growth curve framework was developed. This model and a pattern-mixture model were evaluated using Monte Carlo simulation and then utilized to examine the relationship of two psychological traits, multivariate locus of control and optimism, to the decline in CD4 cell count over 3½ years of follow-up in HIV+ gay men. The simulations showed that even small amounts of non-random truncation seriously biased both simple estimates of the intercept and slope (Level 1) as well as estimates of the relationship of traits to the intercept and slope (Level 2) from the simple mixed model. Neither the propensity score nor the pattern-mixture estimators showed consistent improvement of Level 1 estimates. At Level 2, the propensity score model showed substantial improvement over the simple mixed model and the pattern-mixture model in the amount of bias related to non-random missing data in the predictors of the slope in the growth curve. In the data from gay men (from the period 1986--1991), the propensity score model showed that men who believed that powerful others controlled their lives had a slower decline in CD4 cell count and men who believed that their lives were controlled by chance had a much quicker decline in CD4 cell count. These findings may be related to the men with chance attributions being less likely to seek and adhere to medical regimen, whereas believing life is controlled by powerful others might be associated with more integration with the health care system.

Keywords

Biology, Biostatistics; Health Sciences, Public Health

Link to Full Text

http://access.library.miami.edu/login?url=http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqdiss&rft_dat=xri:pqdiss:3001166