Quantitative Methods for Drug Discovery and Development
(19 Jun - 14 Jul 2017)

~ Abstracts ~


Medical Product Safety: Biological Models and Statistical Methods
Tze Leung Lai, Stanford University, USA

Over the past decade there has been a greatly increased focus on the safety evaluation of medical products. Safety data are routinely collected throughout preclinical in-vitro and in-vivo experiments, clinical development and post-approval studies and monitoring. While the majority of clinical studies are designed to investigate the hypothesized efficacy of a compound, safety outcomes are not generally defined a priori. This brings a number of challenges on how to best analyze the high-dimensional safety data, in order to detect safety signals earlier, and at the same time, reduce the rates of false signals and false non-signals. Depending on the questions of interest and the systems for collecting safety data, statistical methods applied to safety data analysis could differ dramatically.

This tutorial will introduce cutting-edge biological models for safety evaluation and prediction (e.g., in silico models, models for organ-specific toxicity) as well as the commonly used and innovative statistical methods that are tailored for specific objectives and data types for safety signal detection and benefit-risk assessment. Some frequently encountered issues and challenges in safety data analysis are discussed. Examples are given throughout the presentation.

« Back...


Regression tree methods for precision medicine
Wei-Yin Loh, University of Wisconsin-Madison, USA

A major difficulty in developing treatments for current diseases, such as cancer, is that treatments seldom benefit all patients equally; a treatment may be beneficial for some patients but it may be ineffective, besides causing unpleasant side effects, in others. A goal of precision medicine is to find the subgroups of patients who respond positively to a treatment. Examples of subgroup-specific treatments include Herceptin and Perjeta, which are effective treatments for breast cancers that test positive for the HER2 protein. These patients are less sensitive to hormone therapy, a common treatment for the disease.

The task of finding subgroups defined by biomarkers, including genetic variables, and other patient characteristics is made harder by the increasing amount of data that are made possible by technological advancement. As a result, research is turning towards the use of machine learning algorithms to help sort through the information. One class of such methods is regression trees, which are well suited for identification of patient subgroups with differential treatment effects. The tutorial will mainly focus on one regression tree algorithm called GUIDE. It will explain the statistical ideas behind GUIDE, illustrate them on real examples, compare GUIDE with other algorithms, and demonstrate the software. Published literature on GUIDE and its free software may be obtained from http://www.stat.wisc.edu/~loh/guide.html.

« Back...

Best viewed with IE 7 and above