Our method is applicable to case-control studies (Online Methods). We demonstrate this by using the summary-level statistics of the DIAGRAM meta-analysis for T2D from a discovery set of 8,130 affected individuals (cases) and 38,987 controls. In our example analysis, we focused only on the CDKN2B region, where there was some previous evidence of multiple signals16. We analyzed the DIAGRAM meta-analysis data, with allele frequencies and LD structure estimated from the ARIC cohort, and replicated the findings by a joint analysis using the QIMR cohort as reference sample. Two SNPs, rs10965250 and rs10757282, which are only 700 bp apart, were retained in stepwise model selection as jointly associated SNPs with P < 5 × 10–12, using either the ARIC or QIMR cohort as a reference sample for LD (Table 4), consistent with the result from a previous analysis that the two SNPs define a haplotype association16. The risk alleles of these two SNPs are negatively correlated (r = –0.53 and –0.59 in the ARIC and QIMR cohorts, respectively); therefore, the secondary SNP (rs10757282; PM = 3.1 × 10–4) was masked by
