Estimating Relative Risk Parameters in Presence of Missing Covariates

• Activity Number: 55
• Type: Contributed
• Date/Time: Sunday, August 8, 2004 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Jinbo Chen*+ and Nilanjan Chatterjee and Mitchell H. Gail
• Companies: National Cancer Institute and National Cancer Institute and National Cancer Institute
• Keywords: missing covariate ; logistic regression ; pseudo-likelihood ; piecewise exponential distribution ; EM algorithm ; weighted average

Abstract

It was of interest to estimate the relative risks of breast cancer for various risk factors, including reproductive factors (R) and mammographic density (MD), using data from BCDDP study. BCDDP followed a cohort of 280,000 women for their breast cancer status for up to five years (screening phase, 1975-1980); then a subsample was selected for long term follow-up (follow-up phase, 1980-1995). For the screening phase, R were available for a nested case-control sample, a portion of which also had MD data. For the follow-up phase, all subjects had R, some of which were updated over time, but only a small subset had the MD data. The missingness of MD information imposed a challenge for our analysis, and it was desired to make efficient use of all available data. We assumed logistic regression model for the screening phase and adopted a pseudo-likelihood approach for estimation. A piecewise exponential model was assumed for the follow-up phase and the maximum likelihood estimates were obtained via the EM algorithm. The two sets of results were then combined by taking their weighted average.