Quantitative Assessment of Probabilistic Forecasts with Applications in Epidemiology
Freitag, 10.11.06, 11:15-12:15, Raum 404, Eckerstr. 1
In this talk I will describe methods for model choice and model criticism based on probabilistic forecasts of external data. Special emphasis will be given to multivariate predictions and predictions of count data. The methodology will be illustrated through a case study from chronic disease epidemiology.
Quantitative Assessment of Probabilistic Forecasts with Applications in Epidemiology
Freitag, 10.11.06, 11:15-12:15, Raum 404, Eckerstr. 1
In this talk I will describe methods for model choice and model criticism based on probabilistic forecasts of external data. Special emphasis will be given to multivariate predictions and predictions of count data. The methodology will be illustrated through a case study from chronic disease epidemiology.
Survival analysis in high dimensions
Freitag, 8.12.06, 11:15-12:15, Raum 404, Eckerstr. 1
One important topic of current research on prognostic factor studies is the development of methods that can can be employed to analyse high-dimensional data, where the number of explanatory variables is much larger than the number of observations. This is mainly driven by the requirements of biomedical applications such as DNA microarrays. The major problem of analyzing such data is the danger of overfitting. Methodological challenges arise in using large sets of covariates, e.g. patients gene expression profiles, to predict survival endpoints on account of the large number of variables and their complex interdependence.\n\nThe aim of this talk is to show how biostatistical procedures can be employed to analyse high-dimensional data. This include penalized Cox regression, but also boosting proportional hazards models and random survival forests.\n\nWe illustrate the different approaches using simulated data as well as real data from clinical microarray studies including gene expression and array CGH data. The results will be compared with respect to the prediction error and interpretability of the results. Comparisons of predictive accuracy are done using time-dependent ROC curves and related methods.
Semiparametric Mixed Models and Boosting
Freitag, 19.1.07, 11:15-12:15, Raum 404, Eckerstr. 1
Common approaches to the fitting of additive mixed models are based on the representation of additive models as mixed models. In the talk an alternative approach based on boosting techniques is presented. Boosting originates in the machine learning community where it has been developed as a technique to improve classification procedures by combining estimates with reweighted observations. In linear mixed models as well as in additive mixed models the advantage of the proposed componentwise boosting technique is that it is suitable for high dimensional settings with many potentially influential variables. The approach allows to fit additive models for many covariates with implicit selection of relevant variables and automatic selection of smoothing parameters. Moreover, boosting techniques may be used to incorporate the subject-specific variation of smooth influence functions by specifying "random slopes" on smooth effects. This results in flexible semiparametric mixed models which are appropriate in cases where a simple random intercept is unable to capture the variation of effects across subjects.
Model-driven characterization of gene-regulatory networks in T helper lymphocytes
Freitag, 26.1.07, 11:15-12:15, Raum 404, Eckerstr. 1
Cytokines regulate immune responses, cell growth, and cell differentiation. Cytokine communication is organized in networks that link multiple cell types releasing or taking up these messengers.\n\nTo rationalize how the readily diffusible, pleitropically acting cytokines can elicit specific and coordinated regulatory effects presents formidable problem. We developed a mathematical model of a prototypical small cytokine network: intercellular IL-2 signalling between regulatory T cells (Tresp) and conventional antigen-responsive T cells (Tresp). While Tresp activate immune responses, Treg inhibit them. Paradoxically, IL-2 produced by Tresp can act as a growth factor for both cell types. The model shows how a binary decision - activation of either Treg or Tresp - is achieved through feedback-enhanced competition for the growth factor. Positive feedback and spatial coupling emerge as critical processes from the analysis. We have verified part of these theoretical predictions experimentally.
Multivariate Frailty-Modelle - ein Überblick
Freitag, 9.2.07, 11:15-12:15, Raum 404, Eckerstr. 1
Die traditionellen Methoden der Ereigniszeit-Analyse gehen von zwei grundlegenden Annahmen aus: Erstens, die beobachteten Ereigniszeiten sind voneinander unabhängig, und zweitens, alle Individuen sind unter Risiko und erleiden letztendlich das interessierende Ereignis, wenn die Beobachtungsdauer lang genug ist.\n\nUm die erste Annahme zu umgehen, wurden zahlreiche Modelle in der multivariaten Lebensdaueranalyse entwickelt. Diese Modelle fallen in zwei große Klassen marginale und Frailty-Modelle. Der Vortrag handelt von Frailty-Modellen, die insbesondere dann angewendet werden, wenn die Assoziation zwischen den Lebensdauern selbst von Interesse ist und nicht als störender Parameter behandelt wird. Die Frailtyvariable agiert dabei multiplikativ auf der Hasardfunktion. Dazu wird auf das verbreitete Shared-Frailty-Modell und dessen Erweiterung, das Korrelierte-Frailty-Modell eingegangen.\n\nZur Abschwächung der zweiten obigen Annahme wurden in der Literatur so genannte cure-Modelle etabliert. In diesen Modellen wird davon ausgegangen, dass nicht alle Individuen der Studienpopulation gegenüber dem interessierenden Ereignis anfällig sind. Diese Modelle sind Spezialfälle der Frailty-Modelle und werden anhand von Datenbeispielen näher diskutiert.