|
Department of Biostatistics HIV Working Group 2012 - 2013 |
ABSTRACT: Having an accurate estimate of HIV incidence is important for understanding transmission dynamics, evaluating the efficacy of prevention programs, and identifying high-risk groups for targeted interventions. HIV incidence is very difficult to measure because individuals can be infected for a long time and not know when the infection event occurred. Longitudinal studies tracking cohorts of seronegative individuals are prohibitively expensive and subject to bias. An alternative to such studies involves cross-sectional assays that distinguish between recent and long-term infections on the basis of a host or viral marker. We describe a novel marker based on the increase in viral diversity that occurs over time within individuals infected with HIV.
ABSTRACT: With the increased availability of highly active antiretroviral therapy (HAART) in many regions, the prognosis for HIV patients has improved substantially and it is increasingly apparent that regimens must be both effective and tolerable for patients for successful long-term treatment. To this end, we utilize the data from three recent AIDS Clinical Trials Group (ACTG) studies to develop and validate a risk prediction model for regimen failure among HIV patients beginning treatment. In addition to investigating factors associated with regimen failure, we investigate whether the predicted risk of regimen failure can be utilized in the context of personalized treatment decisions by applying the proposed model to the patients in another recent ACTG study and assessing the treatment effects among patients with differing levels of predicted risk.
ABSTRACT: We present a framework to construct sexual contact networks using sampled data collected from ego-centric surveys. The constructed networks not only target mean estimates of local network properties (e.g. density, degree distributions, and mixing frequencies) but also incorporate the uncertainty due to sampling. We will utilize this framework in two applications. The first is designing a large cluster-randomized trials (CRTs) to evaluate the effectiveness of HIV prevention strategies in reducing HIV incidence in Botswana. The second is combining epidemic data and sexual histories to understand transmission networks.
ABSTRACT: Currently under development are three cluster-randomized trials (CRTs) sponsored by CDC, NIH and USAID to evaluate the effectiveness of HIV prevention strategies in reducing HIV incidence in different settings in Africa. We discuss design and analytical challenges for one of these studies—the Botswana Combination Prevention Project (BCPP). In CRTs, a major driver of power is the intraclass correlation; approaches for calculating power have been derived from models in which cluster-level effects are assumed to be independent as are individual outcomes within a cluster. In HIV prevention studies, correlation structure for HIV infection endpoint is driven at least in part by the sexual network; correlation between partners would be expected to be higher than that among individuals who are far apart in the sexual network. Because sexual network is unobserved, it is not generally possible to obtain samples within each cluster in which outcomes are independent. We investigate analytically and through simulations how different sampling strategies and mechanisms for generating correlation would affect study power. For the BCPP, we use simulation studies to predict effect size and investigate power under different study designs and conditions regarding individual behaviors and community characteristics. We develop a new method to generate a robust collection of sexual networks utilizing both the estimated degree mixing matrix and its sampling variability. In order to model realistic community level correlation structure, we first generate a collection of sexual networks using data from a BHP pilot study in Mochudi, Botswana and a network study in Likoma island, Malawi, and then propagate disease on these networks. Simulations allow us to take into consideration some sexual network characteristics, such as mixing within and between communities, as well as coverage levels for different prevention modalities in the combination prevention package under study. We also consider how to handle the issues of interval censoring of the outcome of interest (time of infection with HIV) and to adjust for covariate imbalance at baseline.
This is joint work with Rui Wang, Ravi Goyal, and Vladimir Novitsky.
ABSTRACT: The impact of in utero antiretroviral (ARV) exposure on fetal and postnatal growth among perinatally HIV-exposed uninfected (HEU) children is not well understood, and is of increasing importance as new ARV medications are prescribed to pregnant women. The US-based Surveillance Monitoring for ART Toxicities Study in HIV-uninfected Children Born to HIV-infected Women (SMARTT) protocol of the Pediatric HIV/AIDS Cohort Study (PHACS) is a prospective cohort study designed to estimate the incidence of conditions and diagnoses potentially related to in utero exposure to antiretroviral therapy and/or exposure in the first two months of life among children born of HIV-infected mothers. Thus, we evaluated growth trajectories in HEU children up to age 12 years in SMARTT compared to US norms. Age- and sex-adjusted z-scores for weight, length or height, and weight-for-length or body mass index were calculated using Centers for Disease Control and Prevention 2000 reference standards. For children born preterm (< 37 weeks), gestational age-adjusted z-scores were calculated at birth and at one year old. The proportion of children with low growth (< 3rd percentile) for each parameter was calculated at each age. The mean (95% confidence interval, 95% CI) z-score by age was estimated for each growth parameter using mixed models for repeated measures. Although HEU infants were small at birth, we found that they caught up with US norms by 2 years of age and exceeded norms through 12 years of age where the mean WT and BMI were almost 1 SD higher than normal. The higher than expected rates of small for gestational age and prematurity in HEU children may increase the risk of adverse outcomes later in life (obesity, diabetes, and/or neurodevelopmental problems). Because childhood obesity is an epidemic in the US, it will be important to determine how obesity rates in HEU children compare to contemporary cohorts of children. The next step is to evaluate the effect of individual ARVs on fetal growth and growth trajectories over childhood.
ABSTRACT: A single patient infected by HIV will carry a large population of related, diverse viral strains usually described as a quasi-species. This population diversity complicates both drug resistance profiling and the development of broadly-reactive HIV vaccines. Early detection of minority variants is critical for identifying novel mutations contributing to drug resistance. Next-generation sequencing has the potential to increase our ability to resolve genetic diversity by more deeply sampling HIV populations. But the inherent error rates of the sequencing platform establish a lower bound for detecting low-frequency variants.
We describe the analysis of an HIV quasi-species using Illumina sequencing technology at >50,000-fold coverage. Using a control population of five different HIV genomic sequences present in the sample at known concentrations, we have developed an automated computational pipeline that reliably separates sequencing errors from real variations. The workflow automates quality control, alignment of sequencing reads, re-alignment around insertions and deletion and classification of sequencing artifacts using a novel evaluation scheme that takes into account sequence and alignment quality and uniqueness of neighbouring positions. This enables us to reliably distinguish minority variants at a lower boundary of 0.2% clonal variation from common sequencing errors while minimizing false positive variant calls.
ABSTRACT: Our goal is to estimate the response to antiretroviral therapy (ART) among HIV-positive patients who start ART in sub-Saharan Africa. Due to different strains of the virus and different treatment practices, with treatment initiation at lower CD4 counts and limited access to ART beyond the program evaluated, the response to ART in sub-Saharan Africa may be different from the response to ART in the US. A useful marker of both disease progression and response to ART is the CD4 count. Previously, we estimated the trajectory of the median CD4 count after ART initiation in patients in the US. There, it seemed reasonable to assume that the data were Missing At Random (MAR): the probability of dropout at any point in time depends on past observed information only, and not further on the prognosis of the patients. We then applied Inverse Probability of Censoring Weighting (IPCW) to account for dropout that is MAR.
In sub-Saharan Africa the assumption of MAR is not reasonable. This has been observed on the basis of ``outreach data'': additional data on a (hopefully representative) subsample of patients lost to follow-up, who were outreached and successfully located after dropout. More of the patients in the outreach sample died shortly after dropout than expected on the basis of their initially observed covariates, and more patients were off treatment, in part because of limited access to ART outside the program evaluated. We will present an extension of Inverse Probability of Censoring Weighting (IPCW), which can be used to account for dropout that is Missing Not At Random (MNAR). We will illustrate this method by estimating the trajectory of the median CD4 count after ART initiation in western Kenya.
This is joint work by Judith Lok, Constantin Yiannoutsos, Agnes Kiragga, and Ronald Bosch.
ABSTRACT: We will briefly describe the structure of our individual-based mathematical model: the CEPAC Dynamic Model (CDM). We will discuss the methods, similar to Approximate Bayesian Computing, used to calibrate the CDM to the South African epidemic and the challenges we've faced in the process. We will also discuss the seeding of the CDM with initial HIV cases in South Africa in 1990 prior to surveillance data and discuss the potential to seed the model with "patient zero." Throughout the talk, we will give reference to future problems and questions we will be able to answer with the CDM model structure.
ABSTRACT: Perinatally HIV-infected (PHIV) children have historically shown deficient growth and pubertal delay. Current combination antiretroviral treatment (cART) regimens have been associated with improved growth, but have also been linked to certain metabolic abnormalities which could interfere with normal pubertal development. Thus, the effects of cART on pubertal onset and maturation are unknown, and no prior studies have addressed this issue. We assessed the timing of pubertal onset and sexual maturation in two large US longitudinal cohort studies conducted 2000-2012. We compared the age at pubertal onset and at sexual maturity based on Tanner stage criteria between PHIV and HEU youth. Because pubertal assessments were typically only conducted once each year, we used interval-censored models to conduct comparisons of PHIV with HEU and to evaluate associations with HIV disease severity and cART. Our primary models adjusted for race/ethnicity and birth cohort. Sensitivity analyses were also conducted to assess adjustment for body mass index (BMI) and height z-scores, although these growth measures may be on the causal pathway between cART and pubertal development. Evaluations for sexual maturity also included comparisons of self-reported age at menarche by HIV status using adjusted Cox regression models. In this large cohort, pubertal onset and sexual maturity (including menarche) occurred later in PHIV than in HEU youth, but the association with cART was modified by secular trends in pubertal development. Early initiation of cART in perinatally infected youth, as currently recommended, may result in more normal timing of pubertal maturation.
ABSTRACT: In this talk we consider the semi-competing risks problem in which scientific interest lies time to some non-terminal event but where observance of the event is prevented by the occurrence of death. In this setting, analyses proceed by either (i) treating death as a censoring mechanism or (ii) redefining the outcome to include death explicitly (e.g. disease-free survival). Crucial, however, is that these analyses strategies ignore the fact that death is a truncation mechanism, rather than a (dependent) censoring mechanism. In this work we propose a Bayesian semi-parametric regression model for semi-competing risks data. Specifically, an illness-death model is adopted to represent three transitions: (1) time to the non-terminal event, from the origin, (2) time to the terminal event, from the origin, and (3) time to the terminal event, from the time of the non-terminal event. Dependence between the two event times is induced via a subject-specific shared frailty. For each of three hazard functions, the log-baseline hazard function is modeled as a mixture of piecewise constant functions, defined on separate time partitions. Model parameters including covariate effects, the baseline transition hazards, frailty terms, and their variance are jointly formulated and estimated via a Metropolis-Hastings-Green algorithm. The proposed framework is applied to data from Medicare Part A on n=16,051 individuals diagnosed with pancreatic cancer between 2005-2008, although is likely to have application outside cancer and, in particular, in studies of HIV/AIDS.
| Back to HSPH Biostatistics |
Maintained by the
Biostatistics Webmaster
Last Update: May 14, 2013 |