Ms Caroline Samuels
Office: 721 |
Modelling survival of older Australians: the Australian Longitudinal Study of Ageing
The Australian Longitudinal Study of Ageing (ALSA) is an on-going study of older Ade- laide residents that began in 1992. This thesis uses proportional hazards models to model the probability of 18-year survival for the participants. A broad range of variables collected at study baseline were considered as potential predictors of survival in the analyses.
Previous similar research using the ALSA data has focused on fitting models using rela- tively few variables at a time. In contrast, our approach was to consider a larger number of variables in our models. We partitioned the variables into ÃÂÃÂ¢ÃÂÃÂÃÂÃÂdomainsÃÂÃÂ¢ÃÂÃÂÃÂÃÂ, broadly charac- terised by the contextual definitions of the variables. These domains are: demographics, health, mobility, social and activity, mental health, behaviour, and sleep.
For each domain, we calculated measures of association between the variables to quantify the extent of association between the variables within each domain. Since the variables within each domain often have strong associations, we first fit separate proportional haz- ards models to each domain. We then conducted backwards elimination to obtain a ÃÂÃÂ¢ÃÂÃÂÃÂÃÂminimal representationÃÂÃÂ¢ÃÂÃÂÃÂÃÂ for each domain.
Finally, the significant variables from each domain were combined to fit a multiple domain model to the data. Variables from five of the seven domains were significant in the multiple domain model. We discuss the goodness of fit of the multiple domain model, and explore how collinearity between domains affected which variables were retained in the final model.