Each set of participants with bipolar disorder was compared against the set of 2938 controls for all of the 276 122 SNPs that met our stringent quality control filter (see above). According to the various diagnostic definitions, the number of cases varied from 102 to 1868. For each SNP we employed the Cochran–Armitage trend test of genotype distributions to test association with disease (i.e. we compared the group with bipolar disorder with the 2938 controls). Assuming no association with disease, the trend test statistic follows the χ2 distribution on 1 degree of freedom (as implemented within the PLINK analytic suite of programs).13 This method is a standard approach to analysis of genetic association data and is robust to departures of the data from Hardy–Weinberg equilibrium. To allow for any systematic inflation of the test statistics we adjusted the trend test statistic of each SNP by λ, where λ is the genomic control inflation factor, estimated to be the median of all 276 122 test statistics divided by 0.456.14 Because SNPs that are near to each other can show correlated association
