Hui-Wen Lin1, Yi-Hau Chen. 1. Biostatistics Research and Consulting Center, Taipei Medical University, Taipei, Taiwan, ROC.
Abstract
OBJECTIVE: The association analysis based on a population-based case-control study is convenient and powerful, but may be biased under population stratification (PS), namely the study population consists of strata heterogeneous in disease rates and allele frequencies. On the other hand, a family-based (e.g. case-parents) study is robust against the PS bias, but may be less convenient to implement. We propose an association analysis that preserves the full robustness property of the family-based analysis while allowing for borrowing information from a population-based analysis. METHODS: A two-stage procedure is proposed. In the first stage, one selects a population-based case-control sample and performs a traditional case-control association analysis. In the second stage, one randomly selects a subset of the first-stage cases and recruits their family controls (e.g. parents), and performs a family-based association analysis. An overall two-stage analysis is then performed to utilize information from the two stages. RESULTS: The proposed two-stage analysis achieves higher power than the second-stage family-based analysis by utilizing information in the first-stage population study, while maintaining the full robustness of the family study and hence is still valid under PS. The proposal can also accommodate parental missingness when the case-parents study is used as the second-stage family study. CONCLUSION: The two-stage analysis facilitates efficient and robust association analysis under PS. Its computation- and cost-effectiveness render it very promising in genome-wide association studies. Copyright 2009 S. Karger AG, Basel.
OBJECTIVE: The association analysis based on a population-based case-control study is convenient and powerful, but may be biased under population stratification (PS), namely the study population consists of strata heterogeneous in disease rates and allele frequencies. On the other hand, a family-based (e.g. case-parents) study is robust against the PS bias, but may be less convenient to implement. We propose an association analysis that preserves the full robustness property of the family-based analysis while allowing for borrowing information from a population-based analysis. METHODS: A two-stage procedure is proposed. In the first stage, one selects a population-based case-control sample and performs a traditional case-control association analysis. In the second stage, one randomly selects a subset of the first-stage cases and recruits their family controls (e.g. parents), and performs a family-based association analysis. An overall two-stage analysis is then performed to utilize information from the two stages. RESULTS: The proposed two-stage analysis achieves higher power than the second-stage family-based analysis by utilizing information in the first-stage population study, while maintaining the full robustness of the family study and hence is still valid under PS. The proposal can also accommodate parental missingness when the case-parents study is used as the second-stage family study. CONCLUSION: The two-stage analysis facilitates efficient and robust association analysis under PS. Its computation- and cost-effectiveness render it very promising in genome-wide association studies. Copyright 2009 S. Karger AG, Basel.