Two special seminars

Tuesday, 6 February 2018
12noon-12.45pm & 4.00pm-4.45pm
University of Melbourne (venue details below) Parkville

PROFESSOR ANDREAS ZIEGLER studied Statistics and Mathematics culminating in a doctoral degree from the University of Dortmund, Germany, in 1994, followed by further studies in genetic epidemiology. Since 2001 he has been head and professor of the Institute of Medical Biometry and Statistics at the University of Luebeck. In 2009, he became director of the newly founded Centre for Clinical Trials at the University of Luebeck. Andreas is also CEO of StatSol, a consulting company which he founded in 2004. In 2014, he became honorary professor in the School of Mathematics, Statistics and Computer Science at the University of KwaZulu-Natal (South Africa). Professor Ziegler’s research interests include the planning and conduct of clinical trials with a focus on medical devices, biomarker studies, genetic epidemiology and machine learning. He has published more than 400 peer-reviewed journal articles and 8 books, including a textbook on genetic epidemiology and a monograph on generalized estimating equations. He has been President of the German Region of the International Biometric Society (DR-IBS) and the International Genetic Epidemiology Society (IGES). 

Seminar 1:
Biostatistical Contributions to the Planning of Clinical Trials: Examples from Three Studies
12.00noon – 12.45pm
Melbourne School of Population and Global Health
Room 515, Level 5, 207 Bouverie St, Carlton

Proper methodological planning and excellent logistics is the key to success or failure of clinical trials. They also have a major impact on total study costs. In this presentation, I will illustrate the impact of biostatistical planning on clinical trials using three randomized controlled trials as examples. The first example deals with precision medicine. Aim of the trial is to show a gradient in response by genetic status to treatment with coenzyme Q10 (CoQ10) in patients with Parkinson's disease. Patients are either randomized to high dose CoQ10 or placebo. The genetic status is assumed to have a gradient, and four different genetic constellations are considered. The challenge in this trial is the creation of a proper statistical hypothesis and analysis model. Using the ideas from time-dependent regression models, we propose a slope test for the analysis of the primary endpoint. In the second example, I will consider a randomized controlled trial for comparing insulin pumps with standard insulin injections in children and adults with type 1 diabetes. The challenge in this trial was that the required sample size could not be achieved within the planned and funded recruitment period. However, effect sizes were assumed to be heterogenous for different age strata. Since recruitment by strata differed from the originally intended composition of strata, we re-estimated the required sample sizes and showed that sufficient power could be obtained with an almost 25% lower sample size. This allowed to complete the trial. In the final example, we consider biodegradable metal-based screws which are modern alternatives to conventional metal implants (titanium, steel) or absorbable polymer-based implants. They are in use for various surgical procedures. In this example trial, biodegradable implants are to be compared with conventional metal implants. The first question is the choice of the primary endpoint. A natural choice would be complete bone healing (yes/no), and the aim of the trial could be to demonstrate non-inferiority of the novel biodegradable metal implant compared to the standard non degradable metal implant. However, the non-inferiority margin would be either relatively large or the sample size would be very high with this approach. I therefore suggest a different approach to formal testing of non-inferiority. In addition, I suggest to also demonstrate superiority. This endpoint makes use of specific properties of the novel material.

Seminar 2:
On Trees, Forests and Machines -- or: Do new Brooms Clean Better?
4.00pm – 4.45pm
University of Melbourne, Evan Williams Theatre G03, Peter Hall Building 160

Classical regression models, such as linear or logistic regression are the standard approach in biostatistics. In the past decade the statistical properties of several machine learning approaches, such as random forests or support vector machines have been better understood. For example, for random forests there are results available on consistency, convergence rates and asymptotic normality. However, machine learning approaches will only be used if the approaches are available in simple to use and fast implementations. In this presentation, I will focus on random forests as learning machine. In the part of the presentation, I will intuitively introduce classification trees and probability estimation trees. Trees will next be generalized to random forests. The statistical properties of random forests are sketched. A specific problem in machine learning is how probability estimates should be updated to make predictions for other centers or for different time points. In the second part of the presentation I will show that both a general approach by Elkan and a novel approach specifically developed for random forests can be used for calibrating probability estimates. The approach will be illustrated by use of data from the German Stroke Study Collaboration.