A key ingredient of successful rigorous research is the employment of appropriate techniques for analyzing research data. The overall goal of this proposal is the development of new statistical methodology for analyzing data from longitudinal studies. Such studies are conducted in almost all areas of medicine. Our proposal will address three issues which have not been adequately addressed in the literature to date.
Aim 1 is concerned with methods for quantifying the potential for accumulated data to predict future observations. This is particularly important for early detection of high risk individuals.
Aim 2 is concerned with the errors in variables in longitudinal data analysis. This problem arises, for example, when observations are made at varying times for individuals, or when measurements fluctuate over time.
Aim 3 is concerned with standardizing measurements for patient characteristics which vary across patients, and/or time. Appropriate standardization reduces variability and enhances interpretability of study results. The methodology we propose for each of our aims, employs marginal regression models and recently described extensions of them called partly conditional regression models. Model parameters are allowed to vary with time in Aim l and with patient characteristics in Aim 3. The methods have several features which are attractive from a technical point of view. Inference is based on estimating equations. New methodology will be evaluated by simulation studies and by application to real data. Two longitudinal datasets will be available to this project. The first pertains to a cohort of over 1000 subjects enrolled at a HMO and followed for over 21 years from birth. Data consist of all recorded height and weight measurements for these individuals. The second is a clinical research database for cystic fibrosis patients cared for at a cystic fibrosis care facility in Seattle. Nutritional, pulmonary function and microbiological data items will be available.
