06/04/2008: Issues in Non-Inferiority Trials: Simulation Designs and a Proposed New Method

Katie Odem-Davis

Subjects are often randomized in clinical trials to one of two or more regimens.  In the simplest case, the new experimental treatment (EXP) is evaluated for superiority relative to an inert placebo (PLA).  However, when a drug has been established to be effective for treatment, use of a placebo for evaluating efficacy of the EXP may be deemed unethical and treatment is therefore compared to this “standard” treatment (STD) in an active-control trial.  Showing superiority of EXP relative to STD requires techniques similar to those for EXP relative to PLA.  However, when EXP has some desirable features, such as improved side-effect profile, increased ease of use, or lower cost, it may be satisfactory to show that EXP is simply not unacceptably worse than STD.  In this setting, non-inferiority (NI) trials may be employed.

Methods used in NI trials are different from those for superiority trials, and there are outstanding issues unique to this field.  In my talk, I will outline four main issues in NI trials, simulation designs for assessing conditions contributing to these issues, and a proposed new method to adjust for violation of a key assumption inadequately addressed by current methods used in this setting.

05/28/2008: Using Bernstein Polynomials to Model Misclassification in BI-RADS Breast Density Measurements

Charlotte Gard

 

Mammographic breast density (BD) measures the extent of radiodense fibroglandular tissue in a woman’s breast.  Numerous studies have shown an association between BD and breast cancer risk.  Recently developed breast cancer risk prediction models include BD measured categorically on the four-point American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) scale.  Because BI-RADS measures are subject to misclassification, it is thought that use of a continuous measure of BD such as percent density could improve the predictive accuracy of the risk models.  Unlike BI-RADS measures, however, continuous BD measures are not routinely collected in clinical practice.  It is, therefore, of interest to better understand the relationship between BI-RADS BD and continuous BD and how this relationship differs by radiologist who assigns the BI-RADS score. 

 

Using data from a subset of Group Health Cooperative enrollees for whom both categorical and continuous measures of BD are available, we will model the radiologist-specific relationship between BI-RADS BD and percent density.  As simple parametric forms may be inadequate to describe this relationship for a particular radiologist, we will explore use of Bayesian nonparametric models based on Bernstein polynomial priors. 

 

In this talk, I will describe BD, how it is ascertained, and how measurement of BD varies within and between raters.  I will briefly summarize epidemiologic literature regarding the association between BD and breast cancer risk and will present several breast cancer risk prediction models.  I will introduce Bernstein polynomials and the use of Bernstein polynomials for estimation of distribution and density functions.  I will demonstrate how BI-RADS BD relates to percent density for radiologists in the current study.  I will present plans for future research, to include the development of breast cancer risk models that incorporate predicted percent density for women for whom only BI-RADS BD and radiologist are known. 

05/21/2008: Conditional Estimation of ROC(t) After a Phase II Group Sequential Diagnostic Biomarker Study

Joe Koopmeiners

The development of diagnostic biomarkers for disease screening is a several phase process.  In phase II, performance of the biomarker is compared to established levels of performance to determine if further study is warranted in larger Phase III studies.  It is expected that most biomarkers identified in phase I will have inadequate performance.  This, combined with the limited availability of samples for rare diseases, make group sequential designs that allow for early termination due to futility an attractive option for Phase II biomarker studies.  An option for early termination will preserve specimens when a marker has inadequate performance but will lead to biased estimates at study completion for studies that do not terminate early.  Estimates from studies that do not terminate early will be used to determine if the marker moves on in development and to design future studies.  Therefore, we desire estimates that are unbiased conditional on not terminating early.  We present a general framework for conditional estimation after a two-stage group sequential trial that allows for early termination. Three bias-corrected conditional estimators are proposed along with conditional confidence intervals.  Estimators and confidence intervals are used to estimate ROC(t) after a two-stage group sequential trial that allows for early termination and simulation results are presented to evaluate their performance.  Finally, use of these conditional estimators is illustrated by estimating ROC(t) for DCP, a new biomarker for liver cancer, after a two-stage trial that allows for early termination.

05/14/2008: HIV-1 and HSV-2 Biological Support for the Epidemiologic Link

Giancarlo Sal y Rosas

Herpes Simplex Virus type 2 (HSV-2) is the main caused of Genital Herpes, although HSV-1 accounts for about half of new cases in developed country. HSV-2 infection is endemic and transmission most frequently occurs by contact with a person who is shedding virus at a peripheral area, mucosa surface, in genital or oral secretions. The biological cycle of HSV-2 infection covers initial infection, latency, reactivation and transmission.

HSV-2 primary infection or symptomatic reactivation induce two potential biological bridges for HIV acquisition: (a) a break in the genital mucosa will bring HIV free cells virus in direct contact with its susceptible targets such as activated CD4 T cells and macrophages that live in the sub epithelium. (b) The presence of HSV-2 virus induces a response of the immune system and therefore an increase in the number of HIV susceptible target cells.

Moreover, in the last few years, scientists have discovered that most of HSV-2 reactivations are short and asymptomatic. In a group of HSV-1 (n=18 ) and HSV-2 positive patients (n=25), the median number of reactivations was 1.5 per moth and 50% of anogenital last less than 12 hours and only 7% of those were symptomatic. We will discuss how the immune system reacts under asymptomatic reactivations and the hypothesis that even asymptomatic reactivations potentially increase HIV acquisition.

To understand the association between HIV-1 and HSV-2, we will discuss, in the first part of my presentation, a little bit of HIV virion, the mechanics of HIV acquisition and what the principal HIV target cells are. The second part of my biology project will discuss HSV-2 infection, in particular primary infection and how the immune system responds to the infection. We will also discuss latency and reactivation (frequency of reactivations) of the virus. Finally, we will discuss potential mechanics that could increase HIV susceptibility in HSV-2 positive patients. Moreover, we will discuss the results of a double blind, randomized; phase 3 clinical trial that tested whether using acyclovir to suppressed HSV-2 in HSV-2 positive patients can reduce HIV acquisition.

05/07/2008: Estimating causal treatment effect in randomized clinical trials with noncompliance and outcome nonresponse

Leslie Taylor

Randomized clinical trials are a powerful tool for investigating causal treatment effects, but in human trials there are oftentimes problems of noncompliance and missing data which standard analyses, such as intention-to-treat, either ignore or incorporate in such a way that the resulting estimand is no longer a causal effect. One alternative to these analyses utilizes the principal stratification framework (Frangakis and Rubin, 2002) where we focus on the complier average causal effect (CACE) which estimates the average causal treatment effect among a subpopulation that would comply under any treatment assigned. In the first part, we estimate the CACE for a binary outcome and focus on the development of a moment estimator that relaxes the assumption of latent ignorability and incorporates sensitivity parameters that represent the relationship between potential outcomes and associated potential response indicators. In the second part, we develop estimators for the CACE using multiple imputation methods, which have been successfully applied to a wide variety of missing data problems, but have not yet been applied to the potential outcomes setting of causal inference. In the third part, we extend the multiple imputation methods to a clustered encouragement design study exploring the role of computer-based care suggestions in managing patients with chronic heart failure.