To explore the degree to which stratification can cause spurious positive association results, we tested for association between BMD and 373 genetic markers using 381 white and 126 black females. The rate of positive results doubled as the proportion of stratification increased, showing the importance of controlling for stratification in association studies.
Introduction: Population-based association studies are commonly used to test the relationship between polymorphisms in a candidate gene and a disease or trait of interest. Although the collection of samples for this type of study design is relatively cost-effective, the statistical analysis may be susceptible to false positive results because of the effects of population stratification. Such results may occur when the underlying populations differ in both the polymorphism allele frequency and mean trait value.
Materials and methods: To explore the degree to which stratification can cause spurious positive association results, we analyzed femoral neck BMD data from an unrelated sample of 381 white and 126 black premenopausal females. As part of a previous genome screen, 373 microsatellite markers had been genotyped for each individual. For simplicity of interpretation, each multiallelic marker was reduced to a biallelic marker, with the most common allele as one allele and all other alleles combined as the second allele. As expected, the black women differed substantially for marker allele frequencies and had significantly higher mean femoral neck BMD than their white counterparts. Random subsets of the white and black samples were sampled, with increasing proportions of stratification (0%, 1%, 2%, 5%, 10%, 15%, and 20% black subjects) in the total analyzed sample. ANOVA was used to test for association between the recoded marker and femoral neck BMD.
Results and conclusions: The rate of positive results for the association test were observed to double as the proportion of stratification increased, with substantial increases in the frequency of false positives even for stratification proportions as small as 2-5%. These results show the importance of controlling for stratification when the trait and the polymorphism allele frequency differ between the races.