Abstract
Identifying genetic variants associated with complex diseases is an important task in genetic research. Although association studies based on unrelated individuals (ie, case-control genome-wide association studies) have successfully identified common single-nucleotide polymorphisms for many complex diseases, these studies are not so likely to identify rare genetic variants. In contrast, family-based association studies are particularly useful for identifying rare-variant associations. Recently, there has been some interest in employing multilevel models in family-based genetic association studies. However, the performance of such models in these studies, especially for longitudinal family-based sequence data, has not been fully investigated. Therefore, in this study, we investigated the performance of the multilevel model in the family-based genetic association analysis and compared it with the conventional family-based association test, by examining the powers and type I error rates of the 2 approaches using 3 data sets from the Genetic Analysis Workshop 18 simulated data: genome-wide association single-nucleotide polymorphism data, sequence data, and rare-variants-only data. Compared with the univariate family-based association test, the multilevel model had slightly higher power to identify most of the causal genetic variants using the genome-wide association single-nucleotide polymorphism data and sequence data. However, both approaches had low power to identify most of the causal single-nucleotide polymorphisms, especially those among the relatively rare genetic variants. Therefore, we suggest a unified method that combines both approaches and incorporates collapsing strategy, which may be more powerful than either approach alone for studying genetic associations using family-based data.