Addressing Selection Bias in Observational Event History Data, with Application to HIV Data from Western Kenya

Mwangi, Ann W.

Full Metadata

Overview

Year:: 2011
Contributor:: Mwangi, Ann W (creator); Hogan, Joseph (Director); Liu, Tao (Reader); Linkletter, Crystal (Reader); Kantor, Rami (Reader); Carter, E (Reader); Brown University. BIOMED: Biostatistics (sponsor)
Genre:: theses
Subject:: HIV/TB; Inverse Probability weighting; Informative censoring; Cox proportional hazards model; Recurrent Events; Double sampling; Proportional hazards models
Extent:: xiv, 83 p.
DOI: https://doi.org/10.7301/Z0HT2MKT

Files

Description

Abstract:: Globally over 34 million people with HIV, with 67% in sub-Saharan Africa alone (UNAIDS) 2009. Data from randomized trials to inform on clinical decisions scarce. Several sources of observational data are being compiled on HIV treatment, outcomes, and modes of care, such as Academic Model Providing Access to Health care (AMPATH) Medical Records System (AMRS). Data from these observational studies, might play a major role for some clinical decisions and monitoring and evaluation of programs. However, appropriate methods are required in the use of these datasets. This project addresses two major complications in the use of large observational studies both for treatment decision and M&E: Non-random allocation to treatment and exposure and Informative loss to follow up (LTFU)In the first chapter we apply methods for marginal structural models to assess the causal effect of combination antiretroviral treatment (cART)timing on mortality among adult HIV patients coinfected with tuberculosis (TB): HIV treatment is complicated by TB due to high pill burden and drug to drug interaction etc. The decision on when to initiate cART in this population is crucial for survival and for slowing AIDS progression. We make use of observational data adjusting for non-random allocation to treatment and LTFU using marginal structural models applying the inverse probability weighting (IPW) technique. In the dataset some patients were LTFU or died before initiating treatment leading to censoring of the exposure. To address this we use an imputation approach.The second part we consider models for drawing causal inference when dealing with recurrentevents in the presence of death using observational data with high informative drop out. We make use of partly conditional model where our target for inference is the population that survive to a specific time point of interest.Lastly we consider models for estimating survival distribution when event times are informatively censored. In the presence of double sampled data we show how to use the double sampled data to estimate a model for the informative censoring distribution, and in turn how to use the censoring model to construct an IPW estimator of the survival distribution.
Notes:: Thesis (Ph.D. -- Brown University (2011)

Content

Access Conditions

Rights: In Copyright
Restrictions on Use: Collection is open for research.

Citation

Mwangi, Ann W., "Addressing Selection Bias in Observational Event History Data, with Application to HIV Data from Western Kenya" (2011). Biostatistics Theses and Dissertations. Brown Digital Repository. Brown University Library. https://doi.org/10.7301/Z0HT2MKT

Relations

Collection:

Biostatistics Theses and Dissertations

Theses and Dissertations for the Biostatistics department.
...