Alt Text
Office of Neuroscience Research > Neuroscience Calendar > Division of Biostatistics Seminar: Peter Wang (WashU Biostatistics)

Division of Biostatistics Seminar: Peter Wang (WashU Biostatistics)

"Using the Whole Biomarker Profile to Predict the Rate of Change in a Primary Trial Outcome"

Add to your
When: Friday, November 10, 2017 at 12:30 PM to 1:30 PM
Where: Becker Library 502 (Medical Campus, 5th floor)


Predictive biomarkers can be used to identify a subgroup of patients who will benefit from or become resistant to a treatment agent and are frequently utilized in randomized clinical trials to guide treatment. To establish the predictability of the biomarkers on the therapeutic primary trial outcome, 3 types of analyses have been widely used: (i) correlation analysis and general linear mixed effects models are applied using only the baseline biomarker as one of covariates; (ii) bivariate linear mixed effects models are fit to evaluate the correlation between the rate of change of the biomarker and that of the primary trial outcome; (iii) the two-stage method which estimates the rate of change in the biomarker at the 1st stage and then use it as a covariate in the 2nd stage to predict the change in the primary outcome. However, these methods either do not use the whole biomarker profile or lead to biased conclusions or cannot estimate the rate of change in the primary outcome. 

Inspired by the well-established joint modeling concept from the joint longitudinal and survival data modeling, we propose a joint model for handling two longitudinal variables simultaneously to take advantage of the whole biomarker profile (baseline + longitudinal change) to evaluate the predictability of biomarkers. This joint model includes two sub general linear mixed effects (LME) models and is able to incorporate both fixed effects and random effects. The joint model can be readily estimated by the method of maximum likelihood and is very flexible for handling missing data. Through simulation, we demonstrate that our method yields unbiased estimator and keeps the Type I error well under-control. We demonstrate its application using the DIAN observational study. 


Coffee, water, and cookies will be provided

Visit the calendar for the full list of Division of Biostatistics seminars.

For inquiries contact Chengjie Xiong.   

Medical Campus Map  |  Danforth Campus Map