Longitudinal Analysis and Modeling Approach for Mammographic Percent Density and Dense Area

• Activity Number: 106
• Type: Contributed
• Date/Time: Monday, August 8, 2005 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Carol A. Janney*+ and Celine Vachon and James Cerhan and V. Shane Pankratz and Zach Fredericksen and Terry M. Therneau and Thomas A. Sellers
• Companies: Mayo Clinic and Mayo Clinic and Mayo Clinic and Mayo Clinic and Mayo Clinic and Mayo Clinic and H. Lee Moffitt Cancer Center
• Keywords: longitudinal ; mammography ; proc mixed

Abstract

In the Minnesota Breast Cancer Study, serial mammograms (average of three per woman [range of 1 to 10]) were collected on 1794 women (40 years or older). Longitudinal changes in both percent density and absolute dense area were examined. Spatial power was selected for the covariance structure since adjacent measurements should have a higher correlation than measurements further apart. We assumed a random intercept and slope. Our model building approach involved several stages. First, we examined the scatterplots of breast density or absolute dense area by age. Percent density and dense area declined until age 60, then plateaued. We chose a piecewise spline at age 60 to model the functional form of percent density and dense area. Next, we examined the association between covariates (BMI, hormone replacement therapy, age at mammogram, and menopausal status) known to be associated with percent density or dense area. Significant interactions between age and the known covariates were added to the model. Finally, we added other covariates to the model that might influence the outcome over time. Similar results were observed for dense area and percent density.