ONeill_2019_Profile.pdf (1.01 MB)
Profile development in population studies: clustering and latent class methodologies
thesisposted on 2022-08-17, 10:08 authored by Aoife O'Neill
Identifying subgroups within a population can provide important insights to decision makers. When concerned with health data, identifying profiles of individuals who share similar characteristics may inform targeted treatment or education. Categorical data are often prevalent in large population studies. Therefore statistical profiling methods capable of handling these data are required. In this dissertation, TwoStep cluster analysis (TSCA) and latent class analysis (LCA), were used to identify cross-sectional profiles, while latent transition analysis (LTA) was used to identify profiles longitudinally. Data from two Irish longitudinal studies; Growing Up in Ireland (GUI) and The Irish Longitudinal Study on Ageing (TILDA), and one European study of Rheumatology health professionals (HPs), were examined. TSCA was used to examine activity behaviours in 9-year-old Irish children from GUI. Cohesive activity profiles, associated with weight status, were found for boys but not for girls. LCA identified three classes of rheumatology HPs; ‘Traditional’, ‘Reluctant’ and ‘Early Adopters’, based on their methods of measuring physical activity in their patients. Next, a LCA model of risk behaviours in older adults, was developed from a theoretical model of biopsychosocial behaviours, using TILDA data. The analysis identified four classes; ‘Low Risk’, ‘Physical Health Risk’, ‘Mental Health Risk’, and ‘High Risk’, and examined associations with pain development. A traditional classify-analyse approach and a model-based distal outcome approach, were compared. The LTA methodologies were extended by developing a bootstrap approach to calculate 95% confidence intervals for the odds ratios generated by the LTA with covariates model, multiple random seeds were generated to identify the maximum likelihood, and sample weights were incorporated into the analysis. LTA examined changes in pain class over time, using in TILDA data, and investigated how transitions between classes where related to biopsychosocial variables and healthcare utilisation. These results add new findings to the literature and extend the latent class method ologies applied to population data. These findings have important implications for the identification, and potential moderation, of a range of risk factors, as well as potential to inform targeted treatment or education programmes.
Development of a structure identification methodology for nonlinear dynamic systems
National Research FoundationFind out more...