|
Department of Biostatistics HIV Working Group 2011 - 2012 |
ABSTRACT: With the success of trials of individual HIV prevention methods, such as treatment for infected members of discordant partners, PrEP for uninfected at-risk individuals, microbicide for uninfected women, and circumcision for uninfected men, the potential for control of HIV has never been greater. The current challenge is to understand how best to implement these measures in a way that is cost-effective in different epidemic settings throughout the world. This goal can be aided by development of realistic agent-based models of the propagation of HIV and of the effects of combination intervention strategies; such models can be used to suggest the optimal design of studies and to inform their use in developing policies for settings outside those that have been studied. To do so requires in turn, understanding of the factors that most impact on cost-effectiveness. Important information will be developed from both observational and cluster randomized studies, both of which present challenges for design and analysis.
Specific Issues for discussion: What are the local features, including network characteristics, which are most relevant for the cost-effectiveness of a package of interventions? How can we estimate those features from different sources of available data? What is the role of spatial factors in mixing of individuals in networks? What are the optimal designs for studying combination prevention studies (randomized and observational)? How do we incorporate baseline and post-baseline covariates, including network features, in analyses of cluster randomized trials. How do we monitor randomized studies, and how can we make adjust for changes made to study designs from information that develops within and outsides studies.
Working with Chris Barr, Joe Blitzstein, Ravi Goyal, Miranda Lynch, Eric Tchetgen Tchetgen, Alisa Stephens, Rui Wang.
ABSTRACT: Recent methodological advances in covariate adjustment in randomized clinical trials have used semiparametric theory to improve the efficiency of inferences by incorporating baseline covariates; these methods have focused on independent outcomes. We modify one of these approaches, augmentation of standard estimators, for use within randomized trials with correlated outcomes, such as cluster randomized trials or longitudinal studies. Furthermore, we derive and implement semiparametric locally efficient estimators of marginal mean treatment effects for correlated continuous or binary data. Locally efficient estimators are compared to various existing estimators through simulation. To improve small-sample estimation in the case of cluster-randomized trials, we consider several variance adjustments. We demonstrate the potential for imbalance correction and efficiency improvement through application to the Young Citizens Study, a cluster randomized trial of a behavior intervention in Tanzania, and AIDS Clinical Trial Group Study #398, a longitudinal randomized trial comparing the effects of various protease inhibitors in HIV-positive subjects with antiretroviral therapy failure.
ABSTRACT: Method comparison studies are often conducted in medical research to determine whether a novel method of clinical measurement can replace or be used interchangeably with the reference standard method. An increasing number of these studies have recently appeared in the HIV literature, as novel technologies have been developed to measure a critical biomarker for HIV care, the CD4 count. These studies need to be rigorously designed and analyzed to ensure that patients receive optimal care. As such, we describe some fundamental issues for investigators undertaking these studies, including sample size considerations, sampling schemes, and methods of data analysis. We focus on the highly-cited Bland-Altman methods, including the rationale underlying their use and modifications required for different study designs. We illustrate the proposed methods using data from a cross-sectional method comparison of a novel CD4 enumeration technology and conclude with a discussion of our findings and some problems that need further work.
ABSTRACT: Competing risks occur in a time-to-event study in which a patient can experience one ofseveral types of events. Traditional methods for handling competing risks data presuppose one censoring process, which is assumed to be non-informative. In a controlled clinical trial, censoring can occur for several reasons: some non-informative, others informative. We propose an estimator of the cumulative incidence function in the presence of multiple types of censoring mechanisms. The relationship between each censoring process and the recorded prognostic variables can differ, especially if some censoring processes are informative (e.g. treatment non-adherence) and others not (administrative censoring). Consequently, we incorporate this information by modeling each censoring process separately. We rely on semi-parametric theory to derive an augmented inverse probability of censoring weighted (AIPCW) estimator and its standard errors. We demonstrate the efficiency gained when using the AIPCW estimator compared to a non-augmented estimator via simulations. We then apply our method to evaluate the safety and efficacy of three HAART regimens in a study conducted by the AIDS Clinical Trial Group Network, ACTG A5142.
ABSTRACT: The importance of the exchangeability assumption for valid causal inference using observational data is well-described in the literature. However, in practice, more than one version of this assumption may be considered by the analyst. Reasons include (1) lack of complete knowledge of the temporal order of treatment assignment with respect to updates in the measured covariates and (2) the possibility of estimators that are "simpler" from a computational and/or pedagogic perspective. Causal effect estimates reported in various published analyses of both the Nurses' Health Study and a combination of the Multicenter AIDS Cohort Study and the Women's Interagency HIV Study were constructed under particular versions of exchangeability when alternative versions were equally or more justified. In this talk, we will describe the potential impact of such choices on causal inference in these and similar cohorts.
ABSTRACT: As perinatally HIV-infected (PHIV+) youth enter adolescence/young adulthood, factors associated with initiation of sexual activity and the risk for sexual transmission of antiretroviral drug-resistant HIV remain poorly understood. In this talk we present results from an analysis of the US-based Pediatric HIV/AIDS Cohort Study (PHACS), a study that is following PHIV+ youth to examine the impact of HIV infection and antiretroviral treatment. The objectives of this analysis were to (1) estimate the prevalence and incidence of vaginal or anal intercourse, (2) identify factors associated with initiation of intercourse and with unprotected sex, and, (3) for sexually active youth, estimate the proportion of youth with drug resistance and the proportion who disclosed their HIV status to their sexual partners.
ABSTRACT: In recent years, an increasing number of predictive markers have been identified as useful for risk prediction. When interest lies in predicting long term survival, it has often been argued that intermediate event information may be very helpful in improving the prediction. Most existing methods for incorporating potentially censored intermediate event information in predicting long term survival focus on modeling the disease process and are derived under restrictive parametric models in a multi-state survival setting. When such model assumptions fail to hold, the resulting prediction of long term survival may be invalid or inaccurate. We propose flexible landmark prediction modeling frameworks to incorporate intermediate event information and demonstrate that such frameworks could be very useful in two different settings. In the first setting where the overall goal is to predict long term survival at a landmark point, t_0, we demonstrate that a more accurate prediction can be achieved by incorporating intermediate event information up to t_0, along with baseline covariates, using a flexible varying-coefficient model. In the second setting where the goal is efficient estimation of long term survival, we demonstrate that efficiency can be gained via a semi-nonparametric two-stage estimation procedure by incorporating intermediate event information up to t_0 . We further derive a more powerful testing procedure based on these estimates to test for a treatment difference in a randomized clinical trial setting. Simulation studies suggest that the proposed procedures perform well in finite samples and demonstrate substantial potential gains in prediction accuracy and estimation efficiency in these two settings. We illustrate our proposed procedures using an AIDS Clinical Trial Protocol 175 dataset.
ABSTRACT: Between 2004 and 2011, 104,295 patients were enrolled in HIV/AIDs Care and Treatment clinics in Dar es Salaam, Tanzania, with the support of the PEPFAR program, in partnership with the Harvard School of Public Health. Of these, 47,006 met the Tanzanian government's eligibility criteria for initiation of first line anti-retroviral therapy (ARV1) and had sufficient data to be included in analysis, and of those, 8406 met the eligibility criteria for switching to second line ARVs (ARV2). 924 were switched to ARV2s, of whom 52% met the formal eligibility criteria at the time they were switched. The goal of this analysis is to assess the effectiveness and cost-effectiveness of 2ARVs in this large- scale community-based setting, by comparing outcomes among those who were switched to those who were eligible for switching but were not switched. The Kaplan-Meier estimate of the 6-month mortality among switchers was 64% lower than eligible non-switchers, and the 1 year mortality was 46% lower. We will adjust these findings for confounding by indication, time-varying confounding, and immortal person- time bias, accounting for the lag between eligibility for switching and second line initiation (median 11 months). Results will be presented from several analytic options using standard and newer causal methods.
ABSTRACT: As perinatally HIV-infected (PHIV+) youth enter adolescence/young adulthood, factors associated with initiation of sexual activity and the risk for sexual transmission of antiretroviral drug-resistant HIV remain poorly understood. In this talk we present results from an analysis of the US-based Pediatric HIV/AIDS Cohort Study (PHACS), a study that is following PHIV+ youth to examine the impact of HIV infection and antiretroviral treatment. The objectives of this analysis were to (1) estimate the prevalence and incidence of vaginal or anal intercourse, (2) identify factors associated with initiation of intercourse and with unprotected sex, and, (3) for sexually active youth, estimate the proportion of youth with drug resistance and the proportion who disclosed their HIV status to their sexual partners.
ABSTRACT: An agent-based stochastic HIV transmission model, called the Cost Effectiveness in Preventing AIDS Complications (CEPAC) Dynamic Model (CDM) was created and is currently being calibrated to the South African HIV epidemic. This model examines the behavioral and biological factors that influence HIV transmission and dynamics on both the individual and population levels. The model also calculates the cost-effectiveness of HIV prevention programs when implemented individually and in combination. We will present a description of the model structure, initialization, and calibration as well as discuss the potential uses of this model for other epidemics and applications.
ABSTRACT: Review of interim clinical trial data has many potential advantages: an ethical attractiveness with potentially fewer patients exposed to possibly harmful or ineffective therapies; economic savings with smaller expected sample sizes and shorter trial durations saving money, time, and other resources; and public health advantages as answers are available more quickly to the medical community. Methodologies have been developed to control error rates associated with these interim analyses but many have limitations including having binding decision rules, being test-statistic driven without regard to the clinical relevance of the effect magnitude, and not providing an evaluation of all decision alternatives (e.g., stopping vs. continuing a trial). We present use of prediction and predicted interval plots (PIPs) for flexible quantitative monitoring of clinical trials [[1] Evans SR, Li L, Wei LJ, Drug Inf. J., (2007): 41:733-742; [2] Li L, Evans SR, Uno H, Wei LJ, Stat. Biopharm. Res., (2009):1:4:348-355]. We demonstrate the methods using real examples from the ACTG and NARC. We also show two developing extensions to Bayesian predicted intervals and predicted rings for trials with co-primary endpoints (e.g., benefit and risk). Software for implementing the methods is available: http://cran.r-project.org/web/packages/PIPS/index.html. The methods provide a valuable tool for DSMBs and other decision-makers when evaluating interim data.
ABSTRACT: Combination Anti-retroviral Treatment (ART) is currently the first line therapy for HIV-positive patients. It is very successful in reducing AIDS-defining events and death, so successful that most clinical trials no longer use events as their outcome of interest. However, AIDS-defining events still cause hospitalizations and deaths, also in patients who are virologically suppressed. In addition, HIV-positive patients may also experience serious morbidity and death due to non-AIDS-defining diseases, like liver, cardiovascular, renal diseases, and cancers; more than would be expected in the general population. The first target for therapies is to reach virologic suppression. Next, it would be of interest to know who is at the highest risk for developing AIDS-defining and serious non-AIDS-defining events while being virologically suppressed. This would help guide research targeted at therapies to prevent serious events in virologically suppressed patients.
I will present an analysis of clinical events in virologically suppressed patients on ART in four ACTG clinical trials. Many of these patients have long-term follow-up in the observational ALLRT study. The events we analyze are death, AIDS-defining events, and serious non-AIDS-defining events. Because of the many successful ART regimens available, both first-line and second-line therapies, we study predictors of clinical events had no-one experienced confirmed virologic failure or discontinued treatment for more than 21 days. In clinical practice, the decision process regarding additional therapies could start at around one year after treatment initiation. Therefore we start our analysis at year one of suppressive ART. For this analysis, we censor patients after they experienced virologic failure or went off treatment longer than 21 days. This leads to informative censoring, which we take into account using Inverse Probability of Censoring Weighting. Candidate predictors include the obvious candidate CD4+ T-cell count at year one, but also CD8 T-cell activation, which was measured from fresh blood samples after one year of suppressive ART in the 1036 patients analyzed. I will present the details and preliminary results of this analysis.
This is joint work with Ron Bosch, Peter Hunt, Steven Deeks, Ann Collier, and Connie Benson.
| Back to HSPH Biostatistics |
Maintained by the
Biostatistics Webmaster
Last Update: April 27, 2012 |