STATISTICAL TECHNIQUES TO ADDRESS ERRORS IN MEASUREMENT OF PHYSICAL ACTIVITY IN A CASE-CONTROL STUDY by

Regular physical activity, fitness, and exercise are crucial for the health maintenance and well being of people of all ages. However, increased exercise or work intensity may lead to certain medical conditions. It is conjectured that increased activity intensity may be associated with pelvic floor disorders. Pelvic floor disorders including urinary leakage and pelvic organ prolapse (dropping of pelvic organs such as the uterus into or out of the vagina) are very prevalent. Limited data suggest that to date physical activity may be beneficial in terms of its general effects and may be harmful to the pelvic floor; however, previous studies examining this association are methodologically flawed. In two case-control studies (collectively termed the PhActs study, [Physical Activity Study]), we will examine the association between lifelong physical activity and different disorders of the pelvic floor. Primary analyses of the PhActs Study will focus on the analyses of strenuous physical activity as an exposure variable for the outcome of stress urinary incontinence (SUI case-control study) and pelvic organ prolapse (POP case-control study). 2 Secondary analyses will be performed using logistic regression incorporating physical activity measured in different ways, graphical time trends display, subcategories of physical activity and sensitivity analyses to examine and correct for ‗errors in measurement' where the data permit. Odds ratios will be adjusted for ‗errors in measurements', most likely using regression calibration methods. The main focus on this MStat project is the analyses which involve adjustment for ‗error in measurement'. In this study, existing data from a reproducibility sub-study of women enrolled in the case-control study will be analyzed by statistical techniques, Regression calibration and Simulation/extrapolation (SIMEX), allowing adjustments to be made to produce correct standard errors and test statistics. Regression calibration and simulation/extrapolation will be compared in this secondary study. SAS and STATA will be used in this study. Regression calibration is a statistical method for adjusting point and interval estimates of effect obtained from regression models commonly used in epidemiology for bias due to measurement error in assessing nutrients or other variables. Regression calibration is appropriate when a gold standard is available in a validation study and a linear measurement error model with constant variance applies or when replicate measurements are available in a reliability study and linear 3 random within-person error can be assumed. The PhActs study includes a sub-study involving replicate measurements of physical activity. The simulation extrapolation (SIMEX) is a method for addressing …