54. Biometrisches Kolloquium/25. ROeS SeminarStatistik in den Lebenswissenschaften: Perspektive und Herausforderung

Gemeinsame Tagung der Deutschen Region (IBS-DR), der Region Österreich/Schweiz (IBS-ROeS) und der polnischen Gruppe (IBS-Polen) der Internationalen Biometrischen Gesellschaft (IBS).

10.03.2008 - 13.03.2008 München

 

Download

Poster (248 kb)

Flyer (152 kb)

Call for Papers (121 kb)

 

Themen

Schwerpunktthemen

 

Adaptiv nahtlose Designs zur Kombination von klinischen Studien der Phase II/III

(Willi Maurer, Mike Branson, Basel)

Bayesianische Modelle in der Biostatistik
(Leonhard Held, Zürich)

Boosting für hochdimensionale biomedizinische Daten
(Peter Bühlmann, Zürich)

Multiple, sequentielle und flexible Methoden
(Werner Brannath, Wien)

Genomweite Assoziationsstudien
(Inke R. König, Lübeck)

Kosteneffiziente Designs
(Holger Dette, Bochum)

Nichtunterlegenheitsstudien
(Axel Munk, Göttingen)

 

Wissenschaftliche Themen

 

Biomarker und Surrogatendpunkte

Genomweite Assoziationsstudien

Hierarchische Modelle

Joint Modelling

Meta-Analyse und Meta-Regression

Genomanalyse

Methodische Entwicklungen in der statistischen Modellierung

Netzwerke und Graphen zur Strukturierung hochdimensionaler Daten

Nichparametrische Methoden

Additivity tests for the mixed model

Räumliche Statistik

Statistische Methoden in Systembiologie und Bioinformatik

Statistische Methoden im Umweltmonitoring

Statistische Methoden für klinische Studien

Biostatistikausbildung an Fachhochschulen

Freie Themen

 

Tutorien

 

Creating R Packages

Termin: 10.03.2008; 8.30 -11.45 Uhr
Organisator: Friedrich Leisch, München
Kosten: 50 € / 25 €
Raum: Hauptgebäude LMU 

Abstract:

One of the key factors to the success of R has been its package system, which allows the R community to develop state of the art statistical software in a decentralised way. R would not be what it is today without the numerous contributions to software repositories like CRAN or Bioconductor.
However, R packages are not only an easy way to share software with the rest of the world, they are also an ideal tool to transparently maintain private collections of functions for personal or workgroup usage, or to distribute data sets to students.

In this tutorial we show how to create an R package from a simple collection of R functions and/or data sets. This includes the structure of an R package, package meta data, name spaces, writing help files and package vignettes using Sweave, building packages for different computer platforms, automatic checking, and regression tests.

About the lecturer:

Friedrich Leisch is Professor of Computational Statistics at the University of Munich and member of the R Development Core Team. He is one of the main designers of the R packaging, documentation and quality control systems, administrator of the Comprehensive R Archive Network CRAN, and author of Sweave. He is main or co-author of CRAN packages like e1071, flexclust, flexmix, mlbench, modeltools, or strucchange.

 

 

Messfehler und Fehlklassifikation bei Regressionsmodellen: Grundlagen und neuere Entwicklungen

Termin: 09.03.2008; 14.00 -18.00 Uhr
Organisator: Thomas Augustin and Helmut Küchenhoff, Statistik LMU München
Kosten: 50 € / 25 €
Raum: Hauptgebäude LMU - A 021

Abstract:

In many practical situations the available data at hand do not exactly convey the information one is looking for: Frequently, the variables of material interest cannot be observed directly or measured correctly, and one has to be satisfied with so-called surrogates or proxies, i.e., with somehow related, but different variables. Typical examples reach from the error of technical measurement devices, of, e.g., radon exposure, to the handling of complex construct variables like dietary intake or quality of life. If one ignores the principle difference between the ideal variables and their observable counterparts and just plugs in the surrogates instead of the ideal variables (`naive estimation'), then all the estimators must be suspected to be severely biased, resulting in deceptive conclusions.

This problem of non-ascertainability of certain ideal variables is referred to as measurement error if the variables are continuous and misclassification if they are discrete variables. In the last years there has been a considerable progress how to adjust for measurement error and misclassification in statistical models. In the first part of the tutorial we give a state of the art overview, where the methods are illustrated by examples form epidemiologic studies. In the second part we focus in more detail on two recent developments:

1. Measurement error corrections for different types of survival models including the Cox model and parametric survival models
2. The simulation and extrapolation (SIMEX) approach as a general tool for handling measurement error and/or  misclassification

References:

Carroll, RJ, Ruppert, D, Stefanski, LA and Crainiceanu, CM (2006). Measurement Error in Nonlinear Models. A Modern Perspective. Chapman  & Hall,  Boca Raton. 2nd edition.

Gustafson, P. (2004). Measurement Error and Misclassification in Statistics and Epidemiology. Impacts and Bayesian Correction, CRC Press, Boca Raton.

Augustin, T. (2004). An exact corrected log-likelihood function for Cox's proportional hazards model under measurement error and some extensions. Scandinavian Journal of Statistics 31, 43-50.

Küchenhoff , H, Mwalili, SM and Lesaffre, E (2006). A general method for dealing with misclassification in regression: The misclassification SIMEX. Biometrics 62, 85-96.

 

Geoadditive Regression

Termin: 09.03.2008; 14.00 -18.00 Uhr
Organisator: Thomas Kneib, München
Kosten: 50 € / 25 €
Raum: Hauptgebäude LMU 

Abstract:

Due to the increasing availability of complex regression data including spatial or spatio-temporal information, extensions of classical generalised linear models are of interest in several fields of applications and in particular in the life sciences. Such extensions should allow to incorporate spatial and temporal correlation in combination with flexible modelling of covariate effects and interactions.
In this tutorial, we discuss geoadditive extensions of generalised linear models that comprise the following features:

- estimation of nonparametric effects of continuous covariates and time scales based on penalised splines,

- estimation of spatial effects based on Markov random fields and bivariate extensions of penalised splines,

- inclusion of cluster- or individual-specific random effects,

- modelling of interactions via varying coe±cients terms and interaction surfaces.

The tutorial will introduce these modelling alternatives in a unified framework to stress their similarities and to enable for unified inferential procedures. Case studies will be presented to discuss questions of practical interest as well as interpretation of estimation results. To facilitate the application of the modelling ideas presented, some comments on software will also be included.

About the lecturer:

Thomas Kneib is a Postdoc at the Department of Statistics at Ludwig-Maximilians-University Munich. During the summer term 2007 he has been Visiting Professor for Applied Statistics at the University of Ulm. His research focuses on semiparametric regression models for non-standard situations including geoadditive regression for categorical responses and survival time regression.

 

Organisation

Tagungsleitung

Dr. Hans Ulrich Burger (Basel)

Prof. Dr. Ludwig Hothorn (Hannover)

Prof. Dr. Stanislaw Mejza (Poznan)

 

Programmkomitee

Prof. Dr. Andrea Berghold (Graz)

Prof. Dr. Heike Bickeböller (Göttingen)

Prof. Dr. Sylvia Frühwirth-Schnatter (Linz)

Prof. Dr. Leonhard Held (Zürich)

Dr. Jürgen Kübler (Basel)

Prof. Dr. Martina Mittlböck (Wien)

Prof. Dr. Hans-Peter Piepho (Hohenheim)

Prof. Dr. Rainer Spang (Regensburg)

Prof. Dr. Gerhard Tutz (München)

Prof. Dr. Andreas Ziegler (Lübeck)

 

Lokale Organisation

Prof. Dr. Ulrich Mansmann

Lehrstuhl für Biometrie und Bioinformatik

IBE, LMU
Marchioninistrasse 15
81377 München

Tel.: +49 (0)89 7095 4491
Fax: +49 (0)89 7095 7491

E-mail: lifestat2008 [a] ibe.med.uni-muenchen.de

Prof. Dr. G. Tutz

Institut für Statistik
Seminar für Angewandte Stochastik

Ludwig-Maximilians-Universität München
Akademiestr. 1
80799 München

Tel.: +49 (0)89 2180 3044
Fax: +49 (0)89 2180 5308

E-mail : tutz [a] stat.uni-muenchen.de

 

Wissenschaftliches Programm

 

• Abstract Volume ( 1 MByte )
  If you are registered as conference participant, you will get the Abstract Volume as printed version at the registration.

• Program ( 1 MByte )
  If you are registered as conference participant, you will get the program as printed booklet at the registration.