Spring 2009 Meeting

Held on March 5, 2009 at the Crowne Plaza Hotel.

The Program consists of three presentations:

  1. Semi-parametric Joint Modeling of Longitudinal and Survival Data  Liang Li, Ph.D., Department of Quantitative Health Sciences, Cleveland Clinic
  2. Advances in Machine Learning: Tree-Based Algorithms: Alternatives to Standard Statistical Modeling CART, TreeNet/MART and Random Forests  Dan Steinberg, Ph.D., Salford Systems
  3. Crossover Versus Parallel Design: Dose-Escalation Design Considerations for First-In-Human (FIH) Studies Zhiwu Yan, Ph.D., Abbott Laboratories

________________________________________________________________________________________________________________________________

Semi-parametric Joint Modeling of Longitudinal and Survival Data by Liang Li, Ph.D., Department of Quantitative Health Sciences, Cleveland Clinic

Biographical Background Dr. Liang Li is currently an Assistant Staff (equivalent to Assistant Professor) of Biostatistics at the Department of Quantitative Health Sciences, Cleveland Clinic. He received his Ph.D. in Statistics from the University of Wisconsin- Madison in 2003. His research interests include measurement error models and analysis of longitudinal and survival data. He has over 20 publications on biostatistics methodology (published in journals including Biometrics, Statistics in Medicine, and Journal of Pharmaceutical Statistics), as well as on medical applications (published in NEJM, JAMA, Circulation, etc.).

Abstract In many longitudinal clinical studies, the level and progression rate of repeatedly measured biomarkers on each subject quantify the severity of the disease and that subject's susceptibility to progression of the disease. It is of scientific and clinical interest to relate such quantities to a later time-to-event clinical endpoint such as patient survival. This is usually done with a shared parameter model. In such models, the longitudinal biomarker data and the survival outcome of each subject are assumed to be conditionally independent given subject-level severity or susceptibility (also called frailty in statistical terms). I will review some of the latest developments for shared parameter models and present a new estimation method that can handle a general class of problems in which the conditional distribution of longitudinal data is modeled by a linear mixed-effect model, and the conditional distribution of the survival data is given by a Cox proportional hazard model. We allow unknown regression coefficients and timedependent covariates in both models. The proposed estimators are maximizers of an exact correction to the joint log-likelihood with the frailties eliminated as nuisance parameters, an idea that originated from correction of covariate measurement error in measurement error models. The corrected joint log likelihood is shown to be asymptotically concave and leads to consistent and asymptotically normal estimators. Unlike most published methods for joint modeling, the proposed estimation procedure does not rely on distributional assumptions of the frailties. The proposed method was studied in simulations and applied to a data set from the Hemodialysis (HEMO) Trial. This is joint work with Tom Greene and Bo Hu.

________________________________________________________________________________________________________________________________

Advances in Machine Learning: Tree-Based Algorithms: Alternatives to Standard Statistical Modeling CART, TreeNet/MART and Random Forests by Dan Steinberg, Ph.D., Salford Systems

Biographical Background Dr. Dan Steinberg, the President of Salford Systems, founded the company in 1982 just after receiving his Ph.D. in Economics at Harvard. He also served as a Member of Technical Staff at AT&T Bell Laboratories and Assistant Professor of Economics at the University of California, San Diego, and has participated in dozens of consulting projects for Fortune 100 clients. He has been honored by the SAS User's Group International (SUGI) and led the modeling teams that won the KDDCup 2000 and the 2002 Duke/Teradata Churn modeling competition. Dr. Steinberg has published articles in statistics, econometrics, computer science, and marketing journals, and has been a featured data mining issues speaker for the American Marketing Association, American Statistical Association, the Direct Marketing Association and the Casualty Actuarial Society.

Abstract Recent methodological developments using entirely new technology are showing enormous promise. Dr. Dan Steinberg, Salford Systems' CEO, will discuss the classic advanced data mining methodology developed by Stanford University Professor Jerome Friedman (Statistics, and Stanford Linear Accelerator Center) and former University of California Professor Emeritus, Leo Breiman. TreeNet/MART and Random Forests methodologies will be presented and the application and results of real-world analyses will be shown.

________________________________________________________________________________________________________________________________

Crossover Versus Parallel Design: Dose-Escalation Design Considerations for First-In-Human (FIH) Studies by Zhiwu Yan, Ph.D., Abbott Laboratories

Biographical Background Zhiwu Yan received his Ph.D. in mathematical statistics in 2006 from the University of Illinois at Chicago. He joined Abbott in September 2006 and is a Research Statistician II. His research interests include optimal design of experiments, specializing in crossover designs.

Abstract First-in-human studies present particular risks to human subjects. Consequently, such studies are typically small in sample size and are conducted in a timelagged, dose-escalation fashion. On account of the smallnumber of subjects, it is important to select efficient designs to increase the power of statistical analyses. Crossover designs are often recommended since they generally provide more precise inferences about the treatment effects as compared to parallel designs. On the other hand, various potential problems associated with crossover designs, such as carryover effects, complicated analyses, more dropouts and ethical issues are also reasons for which people prefer parallel designs. Due to the restrictions on the dosing orders in an FIH study, a crossover design we choose may lose considerable amount of power to an optimal crossover design and, in the worst case scenario (e.g., there are carryover effects), it could even be inferior to a parallel design. We will give discussions on this issue. Another important statistical issue that we are going to talk about is that: in a crossover study do we really gain power by inclusion of 3 the baseline (pre-dose) measurements as a covariate in statistical analyses?