|
Activity Number:
|
359
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #307778 |
|
Title:
|
Semiparametric Event History Models for Analyzing Human Sleep Data
|
|
Author(s):
|
Thomas Kneib*+ and Ludwig Fahrmeir and Alexander Yassouridis
|
|
Companies:
|
Ludwig-Maximilians-Universität München and University of Munich and Max-Planck-Institute for Psychiatry
|
|
Address:
|
Ludwigstrasse 33/II, Munich, International, 80539, Germany
|
|
Keywords:
|
Bayesian smoothing ; MCMC ; mixed models ; multi-state models ; penalised splines
|
|
Abstract:
|
Most sleep studies at the Max-Planck-Institute for Psychiatry in Munich focus on sleep structure and its relation to nocturnal hormone secretion or to psychiatric diseases like depression. Raw EEG signals are recorded at several sites during the night together with concentration of certain hormones in the blood measured repeatedly every 10, 20, or 30 minutes. The EEG signals are classified in several stages such as awake, rapid eye movement and states of non-rapid eye movement sleep. We show how flexible semiparametric event history models based on penalized splines can be applied to analyze categorized sleep data. Similar concepts can be used to analyze raw EEG data through penalized spline based signal regression. Both types of models will be formulated in a Bayesian framework with inference based on either penalized likelihood (posterior modes) or MCMC (posterior means).
|
