"Everything should be made as simple as possible, but not simpler." - Albert Einstein

Seminar Announcement

Statistical Methods for Modeling Time Series of Hormone Data

Nichole Carlson, Department of Biostatistics & Informatics, University of Colorado, Denver

Monday, April 25, 2011

4:00 p.m., room 223, Weber Bldg

ABSTRACT

Many hormones are intermittently secreted in boluses, called pulses, rather than continuously over time.  Pulsatile release often regulates the entire endocrine system.  Therefore, researchers are interested in understanding how pulsatile secretion, in particular the pattern of pulse locations, differs between diseased and healthy subjects.  To study pulsatile hormones, researchers collect frequent blood samples (e.g., every 10 minutes for 24 hours), giving a time series of hormone concentrations on each subject.  This talk starts with an overview of this biology and how this class of data is currently analyzed.  We then present a new approach to estimating the underlying pulse generating mechanism that dominates the time series of hormone data collected in these types of studies.  We develop a new model for pulse generation based on a Bayesian Cox cluster process.  We integrate this pulse generator model into an existing Bayesian deconvolution model for characterizing pulsatile hormone data, offering a fully integrated approach to characterizing pulsatile secretion.  The combined model includes a set of biologically relevant parameters that greatly expands the features of the pulse generator that can be quantified.  We use spatial birth-and-death Markov chain Monte Carlo to estimate the number and locations of the pulses along with the parameters defining the pulse generator model.  We further exhibit the strengths of this model on Cortisol data collected to study the HPA-axis in depressed and non-depressed women.