Nonlinear multivariate time series methods and a generalized estimating equation approach with application to sudden infant death syndrome (SIDS) and climate
by
Michelle Dalrymple
Department of Mathematics and Statistics; University of Canterbury
Coauthors: Dr Irene Hudson (Department of Mathematics and Statistics, University of Canterbury), Assoc Prof Rodney Ford (Community Paediatric Unit, Canterbury Health)

This study examines the incidence of SIDS in Canterbury over the years 1968 - 1999 in relation to climate. An intervention analysis, based on an overdispersed parametric survival model, found two significant points of change in the SIDS series, predominantly attributable to changes in infants' sleep position; showing a significant epoch factor that was modelled in methods 1 and 2 below.

Method 1 involves a three-step non-linear multivariate time series (NLMVTS) approach to produce a model relating SIDS to five climatic time series, that of humidity, pressure, temperature and deviance temperature and sinusoidal time functions. A profile of at risk SIDS days is produced.

Method 2 analyses SIDS data in a way that incorporates serial correlation found with the time ordered count data similar to the approach of Campbell (1994, J. R. Statist. Soc. A, 157, 191-208). It is derived from an assumption that correlation arises from an unobservable dynamic latent process, as in Zeger (1988, Biometrika, 75, 621-629). We test whether an iterative, weighted, and filtered, least squares algorithm, from a generalized estimating equation (GEE), approach is as optimal as the NLMVTS approach in method 1 for parameter estimation.

Date received: July 30, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cadt-12.