All 3 SNPs were in strong linkage disequilibrium (D′ = 1.0, r2 = 1.0 for all comparisons) and were in HWE when assessed in the total cohort (P > .05). When broken down by type of DM and DR status, participants with T2DM and no DR deviated from HWE (P = .009) for all 3 SNPs. Genotype frequencies of all SNPs in both DM subgroups are shown in Table 2. Given the strong linkage disequilibrium observed, all 3 SNPs gave identical association results. All 3 SNPs in EPO were associated with DR status in the combined DM group (T1DM and T2DM) and T2DM alone. The GG genotype of rs1617640, CC genotype of rs507392, and CC genotype of rs551238 were all significantly associated with the presence of DR after adjustment for age, sex, HbA1c level, duration of disease, and nephropathy in the combined DM group (P = .008, data not shown). In the subanalysis for type of DM, these genotypes were significantly associated with DR in the T2DM group (P = .006) (Table 3) but not the T1DM group, although the total number of subjects with T1DM was fewer, reducing the statistical power to detect an association in this group. In the multivariate analyses, further subclassification for the type of DR found all 3 SNPs to be associated with PDR, CSME, and blinding DR in the combined DM group (P = .030, P = .040, and P = .033, respectively) and T2DM alone (P = .020, P = .018, and P = .016, respectively). In addition to controlling for age, sex, HbA1c level, duration of disease, and nephropathy, further multivariate analyses were undertaken, also controlling for hypertension, hypercholesterolemia, BMI, and smoking. All 3 SNPs continued to be significantly associated with overall DR, PDR, and blinding DR in the combined DM group and T2DM alone (P < .05, data not shown). No associations were found in an allelic association model for any SNP and DR (or its subtypes) in the combined DM group or T1DM or T2DM alone (P > .05, data not shown).