Poisson mixture methods and change point analyses to study the relationship between temporal profiles of sudden infant death syndrome and climate.
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
Sudden infant death syndrome (SIDS or cot death) is the leading cause of infant death in the developed world (Byard et al, 1996), and is unique in the medical lexicon in that it represents a pathology defined by categories of exclusion. Historically, Canterbury has had one of the highest rates of SIDS in New Zealand, and New Zealand has had one of the highest SIDS rates in the western world (Nelson, 1996). SIDS is one of the most catastrophic events that can occur within a family, and is particularly traumatic due to parents invariably placing blame upon themselves for their infant's death. Historically, statisticians and epidemiologists have played a major role in SIDS research. Their epidemiological approach has resulted in the identification of modifiable risk factors, including prone sleeping, smoking in an infant's environment, formula feeding, and bed sharing. Public health 'back-to-sleep' campaigns, have directly resulted in a sharp drop in SIDS incidence from the early 1990s. The aim of this thesis is to model and predict the incidence of SIDS in Canterbury, New Zealand (1968-1999), in terms of complex weather patterns, characterised by a diverse array of climatic variables. This is achieved by linking the temporal sequence of SIDS counts with a comprehensive climate profile. The association between climate and SIDS has a long history, with the first reference to a seasonal variation in SIDS (a peak in the incidence of SIDS in colder months) published nearly 150 years ago (Wakley, 1855). Many studies have related SIDS to various meteorological measures throughout the world, yet the only consistent relationship found is between SIDS and seasonality (for example Douglas et al (1998) or Mitchell et al (1999). This study is the first to systematically analyse a multiplicity of climate data at different temporal levels, with an accurate extensive time series of SIDS. Results from change point analyses showed that the Canterbury SIDS profile was constructed of three distinct temporal periods. Logistic regression on seasonality measures, confirmed that seasonality existed in the Canterbury profile. This annual variation in the incidence of SIDS was best measured by different variables for the different periods. Short term relationships between the incidence of SIDS and climate, over and above seasonality, were found for various climatic profiles including humidity, wind (speed, direction and velocity) and pressure. These relationships were identified using regression techniques based on the Poisson distribution, including Poisson regression, Poisson mixture models, Poisson regression with an autoregressive latent structure, and generalised additive models. Three separate aspects of this study have not previously been seen in the literature, and result from novel statistical applications of mixture and change point methods to the incidence of SIDS. Firstly, this study applies mixture methods to a temporal sequence of discrete SIDS counts. Secondly, the study identifies significant points of change in the chronological profile of SIDS counts, which correspond to structural shifts in the underlying distribution. Thirdly, this study methodologically analyses a vast array of climate data at different temporal levels, with an accurate extensive time series of SIDS.