02/06/08: Assessing the surrogate and predictive value of a marker measured after randomization

Julian Wolfson

The identification and evaluation of surrogate endpoints is a major goal of many clinical studies. Pinpointing a reliable surrogate endpoint, often a biomarker, can help predict individual outcomes at an early stage, potentially shortening study length and cost. It may also suggest biological targets for new treatments. Though the search for useful surrogates is relevant to many different research areas, in this talk I will focus on surrogate endpoint assessment in the HIV vaccine context. In this domain, the aim is to identify vaccine-induced immunological responses which can be used as surrogates for HIV infection. A key feature of these immune responses is that they are generally measured some time after vaccination; for example, the T-cell response to a vaccine may take several weeks to fully develop. Analyses which treat such post-randomization measurements as baseline covariates are based on net effects, a mixture of the causal effect of treatment on the outcome and any systematic differences between the groups sharing similar values of the biomarker.

 

In this talk, I will introduce and explain the concept of principal stratification and apply it to isolate causal effects and assess surrogate value in this context. As with most techniques based on counterfactuals, there is a substantial amount of missing data which must be imputed. I will describe some assumptions and a novel vaccine trial design (due to Follmann) which help to identify Surrogate Risk (SR), the estimand of interest. Even with these assumptions and additional data, SR is a function of several non-identifiable terms, so I propose a sensitivity analysis approach to estimation. I will conclude with a discussion of a closely related estimand, Predictive Risk (PR), which is completely identifiable from observed data.

01/30/08: Protein biomarker detection and cardiovascular disease

Sangsoon Woo

The field of “proteomics”, which can be defined as the study of the set
of all proteins present in a cell, is rapidly becoming a key component
of modern biology. Blood plasma exhibits an exceptional proteome in
many respects. It is the most complex human derived proteome,
containing other tissue proteomes as subsets. It is the most sampled
proteome for medical diagnosis. Proteins in plasma have been studied
since before we knew genes existed. Related to a disease, some
proteins are used as biomarkers. For example, many previous researchers
proved that measuring some proteins levels such as LDL, HDL, and CRP
provides information about relative risk of cardiovascular disease.
In this talk, I will address basic concept of biomarker and human
blood plasma and talk about how LDL and CRP are related to cardiovascular disease.

01/23/08: Causal Modeling in Quantitative Genomics

Lin Chen

In this talk, I will propose a causal framework that utilizes the naturally
randomized genotypes to rigorously infer causal regulatory relationships
among genes and traits. Based on experiments in which genotyping and
expression profiling are performed, we propose a non-parametric empirical
Bayesian method to estimate the posterior regulatory probability for any
directed pair of gene transcripts in the genome. The resulting posterior
probabilities can further be used to build transcriptional regulatory networks.
We demonstrate this method on an experiment in yeast, in which genes
known to be in the same processes and functions are recovered in the
resulting transcriptional regulatory network. Building on the causal
framework, I will present a second method to identify the genes whose
transcription levels are causal for a quantitative trait of interest, resulting in a
p-value calculation for each potential causal relationship. This method also
involves extending the theoretical framework to accommodate hidden
variables with large-scale effects. I will present applications of this method
to experiments in yeast and mouse.

Call for Speakers

Several dates remain unfilled for Winter and Spring Quarter. If you’re giving a talk at a conference this summer, or presenting a Biology/General Exam, come practice at Student Seminar first! Check the Upcoming Seminars page for a list of available dates (denoted as TBA) in Winter quarter; contact Julian (julianw@u) or Joe (jskoopm@u) if you’d like to speak during Spring Quarter.

01/16/08: Estimating the capacity for improvement in risk prediction with a marker

Jessie Gu

Consider a set of baseline predictors X to predict a binary outcome D and let Y be a novel marker or predictor. This paper is concerned with evaluating the performance of the augmented model P(D=1|Y,X) compared with the baseline model P(D=1|X). The diagnostic likelihood ratio, DLR_X(Y), quantifies the change in risk obtained with knowledge of Y=y for a subject with baseline risk factors X. The notion has been promoted as a useful way of capturing the increment in risk prediction due to Y. It is contrasted here with the notion of covariate adjusted effect of Y in the augmented risk model.

We propose methods for making inference about DLR_X(Y). We also demonstrate how the population performance of baseline and augmented risk models can be compared using baseline data from a cohort and marker data from a nested case-control study. The key step is to estimate DLR_X(Y) with the case-control data. Finally, we show how the methodology yields estimates of covariate specific predictiveness curves that can be used by an individual to decide if ascertainment of Y is likely to be informative or not for him. Policy makers could also use these curves to identify subpopulations for whom Y is, or is not, likely to be useful. We illustrate with data from a study of renal artery stenosis using the serum biomarker creatinine in addition to standard demographic and clinical factors.