ANALYTIC METHODS FOR HIV TREATMENT AND COFACTOR EFFECTS

点击翻译按钮获取中文摘要

My principal aim is further development of new methods for analyzing observational data bases and randomized trials of HIV-infected persons. The proposed approaches are based either on (i) the estimation of a new class of causal models, the structural nested models using a new class of estimators, the G-estimators, or (ii) new methods for analyzing semi- or non-parametric models in the presence of both informative and non- informative missing data. Many of the new methods are fundamentally "epidemiologic" in that they require data on time-dependent confounding factors, that is, risk factors for outcomes that also predict subsequent treatment with the drug or co-factor under study. The proposed methods of analysis will improve upon previous methods in the following ways. First, the new methods are the best methods available to estimate the effect of a treatment (e.g., AZT) or a co-factor (e.g., marijuana) on an outcome of interest (e.g., time to AIDS or CD4-count level) from observational data available, when symptoms of HIV disease (e.g., thrush, fever) are simultaneously confounders and intermediate variables. We shall use the new methods to estimate treatment and co-factor effects on the evolution of CD4-counts and on time to progression of HIV-disease among subjects in the San Francisco Men's Health Study (SFMHS), and the Multicenter AIDS Cohort Study (MACS). Results will be compared with results obtained using standard methods. Second, the new methods are the best methods available to adjust for dependent censoring, non-random non-compliance, treatment cross-over or termination, and the concurrent effect of additional non-randomized treatments in randomized clinical trials. For example, in ACTG trial 002 of the effect of high-dose versus low-dose AZT on the survival of AIDS patients, patients in the low-dose arm took more aerosolized pentamidine (a non-randomized treatment). The new methods are the best methods available to efficiently incorporate information on surrogate markers (e.g., CD4-count) in order to stop, at the earliest possible moment, randomized trials of the effect of a treatment on a survival time outcome (e.g., time to AIDS). Specifically, we to construct a valid alpha-level test of the null hypothesis of no effect of treatment on survival that incorporates data on the evolution of multivariate surrogate markers, e.g., (CD4-count and HIV antigen level), with power exceeding that of the log rank test. We shall use our new methods to further analyze ACTG trial 0002 as well as ACTG trials 021 and 081. These latter two trials compare the effects of various chemotherapeutic agents on the opportunistic infection and survival experience of HIV-infected subjects.