The aims of this proposal are the development, evaluation and application of methods for the statistical analysis of longitudinal data, with emphasis on observational follow-up studies in epidemiologic and other health related research. The specific focus is to realistically model various departures from the assumptions of commonly applied models, evaluate the effect, consider new estimation techniques that are robust to the possible misspecification and develop methods for detecting the misspecification. Misspecifications to be considered are omitted confounders, measurement error and informative missing data. Special attention will be paid to the relationships between these, and a unified framework will be introduced whenever possible. The proposed work includes: - Fitting models with latent random effects geared to capture possible misspecifications - Developing simple methods for fitting models for informative missing data - Adapting semiparametric random effects models for measurement error problems - Undertaking a comprehensive investigation of the effect of confounding in generalized linear models, and - Developing of a pseudo score test to detect misspecification with GEE The methods will be applied in the analysis of large amounts of longitudinal data on sleep disorders, diabetes, falls among the elderly, neonatal lung disease and eye disease in an aging population.
Showing the most recent 10 out of 14 publications
