Department of Biostatistics
HIV Working Group
2013 - 2014
ABSTRACT: Achieving universal antiretroviral treatment (ART) coverage for HIV-infected ART-eligible patients remains a central goal for global HIV efforts.
In many developing countries, a main barrier to achieving this goal is the lack of human resources to treat HIV/AIDS (HRHA), rather than drugs, equipment or facilities. Any strategy for achieving universal ART coverage must therefore invest heavily in increasing HRHA numbers. However, previous studies found that such strategies can fall victim to their own success because of the significant life-prolonging effects of ART – i.e. increasing the number of people receiving ART leads to an increase in the number of people needing ART in the future. The sobering conclusion was that in many developing countries, universal ART coverage cannot be achieved under realistic increases in the numbers of HRHA, without reducing HIV incidence or changing the organization of ART delivery, e.g. task-shifting from physicians to other health workers.
The conclusions from previous studies were based on evidence about the mortality reduction effects of ART, but at that time we had little evidence of the impact of ART on HIV transmission. Recent results from the HPTN 052 study have provided strong evidence of the preventive effects of ART (96% reduction in new infections among HIV sero-discordant couples). Even if the population level preventive effects of ART are lower, it is plausible that the HIV incidence reduction due to ART may outweigh the mortality reduction – i.e. while people on ART live longer, substantially fewer new HIV infections occur – so the challenge of achieving universal ART coverage may be surmountable.
We investigate this question. We incorporate the HIV transmission-reducing effects of ART into a mathematical model and find that substantially fewer HRHA are needed for achieving universal ART coverage than had been previously estimated. However, currently planned transitions to local management of global ART programs may disrupt HRHA recruitment and delay achieving universal coverage. Our results suggest that a "surge capacity" approach that aggressively scales up HRHA to reach universal coverage quickly will be better for population health outcomes. As the HRHA needs decline post-universal coverage, this surge capacity can be reassigned to other health needs. Further, our results show that without significant additional HRHA, implementing treatment-as-prevention (TasP) may "crowd out" sicker patients and result in adverse health outcomes. It is thus advisable to plan for a HRHA "surge capacity" specifically for delivering TasP, over and above the capacity planned for ART, which can again be reassigned as the burden of the disease dissipates.
ABSTRACT: It sometimes happens in epidemiologic practice, that the outcome in view is not observed for a subset of the sample. The outcome is then said to be missing not at random (MNAR) when conditional on fully observed data, the missingness process and the outcome remain dependent. In such settings, a regression analysis for the outcome cannot be identified without an additional assumption. Identification is sometimes possible, if for all subjects, an exogeneous instrumental variable (IV) is observed, known to satisfy an exclusion restriction, that variation in the IV induces variation in the missingness process, without directly influencing the outcome. This is a core assumption, which together with a parametric specification of the effects of covariates and Gaussian latent variables, essentially produces Heckman's classical specification of a selection model for a regression with outcome MNAR. Heckman's selection model however, is known to be sensitive to the distributional assumption, and can sometimes perform poorly when the assumptions is not met exactly. In this talk, I will present an identification strategy recently proposed as an alternative to Heckman's approach. Maximum likelihood estimation and novel semiparametric methods are described, and the methods are illustrated with an application to estimation of HIV prevalence using household survey data subject to substantial testing refusal rates in Zambia.
ABSTRACT: The treatment of missing values in epidemiological studies typically relies on the assumption of missing at random, which is often unrealistic. For example, in data from household surveys and Demographic Surveillance Sites, rates of refusal to test for HIV are often as high as 50%. Longitudinal data supports the view that respondents are systematically selecting out of testing on the basis of knowledge of HIV status. Then conventional estimates of HIV prevalence, including those from imputation based approaches, will be biased.
I provide an overview of an on-going project to account for missing values in HIV research using Heckman-type selection models, which under certain assumptions, can provide consistent estimates of the parameter of interest, even in the presence of systematic non-response due to selection on unobservables. I review the evidence on the possible extent of selection bias in this context, and discuss extension of the Heckman model to account for finite sample bias, identification of exclusion restrictions in the presence of collinearity, and relaxation of the parametric assumption of joint normality. Finally, I discuss research in progress which aims to validate the implementation of this approach in practice.
ABSTRACT: None Given
ABSTRACT: Little is known on the incidence of immune reconstitution inflammatory syndrome (IRIS) among HIV-positive individuals initiating combined antiretroviral therapy (cART). We explored the changes in incidence of AIDS-defining events previously associated with IRIS—tuberculosis, mycobacterium avium complex (MAC), cytomegalovirus retinitis (CMVR), progressive multifocal leukoencephalopathy (PML), herpes simplex virus (HSV), Kaposi Sarcoma (KS), Non-Hodgkin Lymphoma (NHL), cryptococcosis and candidiasis—after cART initiation. We identified 96,562 individuals in the HIV-CAUSAL Collaboration, which includes data from 6 European countries and the US, who were HIV-positive between 1996-2013, ART naïve, aged ≥18 years, had CD4 count and HIV-RNA measurements, and had been AIDS-free for at least 1 month between those measurements and the start of follow-up. For each AIDS-defining event, we estimated the hazard ratio (HR) for no cART versus <3 and ≥3 months since cART initiation, adjusting for time-varying CD4 count and HIV-RNA via inverse probability weighting. We found that that, with the potential exception of some mycobacterial infections, unmasking IRIS does not appear to be a common complication of cART initiation in high-income countries.
ABSTRACT: Characterizing the genetic determinants of complex traits continues to represent a significant challenge as a large portion of their heritability remains unexplained. In turn, there is increasing interest in developing novel analytic and computational approaches that allow investigators to draw strength from the vast base of existing knowledge regarding underlying genetic structure and biological mechanisms in characterizing genetic association. During this seminar, I will present a gene-level testing framework for genome-wide association studies (GWAS), termed Mixed Modeling of Meta-Analysis P-values (MixMAP). The underlying premise of this approach, similar to many clustered data methods, is that single nucleotide polymorphism (SNP)-level effects are influenced by latent locus or gene level variables and thus all SNP-level p-values within a gene can inform the significance of that gene. This approach is intended to complement post-hoc characterization of gene-level association based on whether the minimum SNP level p-value within the gene or locus is SNP-level significant (the ``min-P" approach) through incorporating knowledge about how SNPs fall within gene regions. Applications to the Global Lipids Gene Consortium (GLGC) and the Meta-Analysis of Glucose and Insulin-related Traits Consortium (MAGIC) publicly-available GWAS metadata are presented for illustration. Additionally, results from ACTG 5202, as well as an alternative quantile-regression approach that considers fully the raw p-value distribution, will be presented. All statistical analysis is performed using R version 2.15.2 and the open-source, publicly-available MixMAP package.
ABSTRACT: We present a detailed description of an individual-based mathematical model of HIV transmission, called the CEPAC Dynamic Model (CDM), calibrated to sexual behavior, HIV prevalence, and HIV incidence in South Africa. We use the CDM to better understand the impact of combination HIV prevention interventions on prevalence, incidence, survival, and cost. We also explain the discrepancy between the HIV prevalence trajectories found in large scale observational cohorts in South Africa and previous mathematical modeling analyses.
ABSTRACT: None Given
ABSTRACT: None Given
|Back to HSPH Biostatistics||
Maintained by the
Last Update: March 3, 2014