||5R01CA096885-03 Interpret this number
||University Of Pennsylvania
||Longitudinal Analysis for Diverse Populations
DESCRIPTION (provided by applicant): Community based interventions (e.g. to reduce obesity and increase physical activity) can play an important role in reducing the risk and overall mortality and morbidity of diseases such as coronary heart disease and cancer. They are especially important for African Americans, who are disproportionately at risk for a wide range of negative health conditions, including cancer of the breast, colon, esophagus, prostate, pancreas, ant stomach; mortality from cardiovascular disease; hypertension; and elevated serum cholesterol. This project will develop more efficient and cost-effective methods for analysis of longitudinal studies using quasi-least squares (QLS), with special emphasis on studies in diverse populations. Our aims are: (1) To develop more efficient and informative methods for analysis in longitudinal studies and community-based interventions, by applying QLS for a wide range of correlation models not currently implemented for generalized estimating equations (GEE) and constructing confidence intervals and tests of hypotheses for the parameters of the new structures, for data with one or more levels of within-cluster associations (e.g. both within families and within subjects over time). (2) To develop methods for planning more powerful studies and taking advantage of re-computing interim power, by (i) assessing loss in efficiency in estimation for different study designs and correlation models and (ii) providing explicit formulas for sample size and power calculations for several correlation structures, including the structures implemented in Aim 1. This aim will consider both the regression and the correlation parameters. (3) To apply our methods in analyses of several studies in female, pediatric, and African-American Populations at the University of Pennsylvania, to further refine and tailor their development to the characteristics of data for diverse populations and to answer new questions that our methods make possible. (4) To compare and contrast our approaches with alternative methods, including methods based on random effects models and recent extensions of GEE, via simulations to assess small sample efficiency and bias and data analyses to compare results of the different approaches. (5) To implement the methods for analysis (Aim 1) and planning (Aim 2) in Stata programs, for use by other statisticians. Further, to widely disseminate the programs, and their documentation, on a web site developed for this project.