The application of ordinal regression models in quality of life scales used in gerontology
Aim of the Thesis: The area of health-related quality of life has received increasing attention particularly in gerontology. As this area grows in importance, issues such as the design and analysis of instruments that measure this multi-dimensional outcome need to be addressed. Ordinal regression models are statistical methods that can be used to analyse ordered health-related quality of life measures. However, their use is limited in the literature. The aims of this thesis are (i) to compute all ordinal regression models and compare these models with other statistical methods (such as linear regression and binary logistic regression models) and (ii) assess the use of the stereotype ordinal regression model. Procedure: The data used to implement the regression models was from the Medical Research Council Cognitive and Function Ageing Study (MRC CFAS). In particular, two measures were chosen: the Townsend Disability Scale and the Health Status question. Results: Linear regression models were found to summarise the ordinal data inadequately given both ordinal measures. Binary logistic regression models were only adequate for analysing ordinal quality of life scales, if one could assume that the odds ratios were the same over all the binary groupings of the ordinal scale. However, one may still encounter other problems related to multiple testing or different effects in different models. Ordinal regression models provide a more sensitive and comprehensive analysis. These methods are easily adapted to different types of ordinal quality of life data. The 'best-fit' ordinal regression model for the health status ordinal categories was the partially constrained adjacent category model. The 'best-fit' model for the Townsend Disability Scale was the fully constrained continuation ratio model. Conclusions: This study has provided a method (based on first principles) of implementing all ordinal regression models. The comprehensive results from this thesis, suggest that ordinal regression models are indeed superior compared to other methods for analysing ordinal quality of life data. Evidence suggested that the stereotype model was of little use. Key words: Gerontology, health-related quality of life, ordinal regression models.