Incorporating Evidence for Population Stratification Bias in Combined Analysis of Individual and Family Data

• Activity Number: 12
• Type: Topic Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lucia Mirea*+ and Lei Sun and James Edward Stafford and Claire Infante-Rivard and Shelley Brenda Bull
• Companies: Dalla Lana School of Public Health and Dalla Lana School of Public Health and Dalla Lana School of Public Health and McGill University and Samuel Lunenfeld Research Institute
• Keywords: population stratification bias ; candidate gene ; statistical accuracy and positive predictive value ; transmission disequilibrium ; case-control studies ; family-based association

Abstract

Analysis integrating case trios (CT) and independent case-control (CC) individuals can increase statistical power when the latter is free from population stratification bias (PSB). Established methods initially test for PSB by comparing results from CT and CC using an arbitrary testing level, typically 5%. Combined analysis are performed if no PSB is detected, otherwise analysis are restricted to CT. As a novel alternative, we use the PSB p-value directly in weights to combine CT and CC estimates, and standardize with a bootstrap variance estimate. In simulations across a range of PSB levels, the weighted approach has reduced type I error compared to testing PSB at the 5% level, increased (decreased) accuracy for larger (smaller) PSB, and overall increased positive predictive value. To illustrate, the weighted approach is applied to a candidate gene study of childhood leukemia.