Assessing the clinical utility of a genetic risk score constructed using 49 susceptibility alleles for type 2 diabetes in a Japanese population

Context: Genome-wide association studies (GWASs) have identified over 60 susceptibility loci for type 2 diabetes (T2D). Although the ability of previous genetic information (∼40 loci) to discriminate between susceptible and nonsusceptible individuals is limited, the added benefit of updated genetic information has not been evaluated. Objective: We assessed the clinical utility of GWAS-derived T2D susceptibility variants in a Japanese population. Design and Setting: We conducted a cross-sectional case-control study. Participants: T2D cases (n = 2613) and controls (n = 1786) with complete genotype data for 49 single-nucleotide polymorphisms (SNPs) were selected for analyses. Outcome Measures: We constructed genetic risk scores (GRSs) by summing the susceptibility alleles of 49 SNP loci for T2D (GRS-49) or 10 SNP loci with genome-wide significant association in previous Japanese studies (GRS-10) and examined the association of the GRSs with the disease by receiver operating characteristic analyses using a logistic regression model. Results: The GRS-49 was significantly associated with T2D (P = 8.75 × 10^-45). The area under the curve (AUC) for GRS-49 alone (model 1) and forage, sex, and body mass index (model 2) was 0.624 and 0.743, respectively. Addition of the GRS-49 to model 2 resulted in a small but significant increase in the AUC (AAUC = 0.03, P = 7.99 × 10^-15). Receiver operating characteristic AUC was greater for GRS-49 than for GRS-10 (0.624 vs 0.603, P =.019), whereas the odds ratio per risk allele was smaller for GRS-49 than for GRS-10 (GRS-49, 1.13, 95% confidence interval 1.11-1.15; GRS-10, 1.26, 95% confidence interval = 1.22-1.31, P = 7.31 × 10^-10). The GRS-49 was significantly associated with age at diagnosis in 1591 cases (β = -0.199, P =.0069) and with fasting plasma glucose in 804 controls (β = 0.009, P = 0.021). Conclusions: Updated genetic information slightly improves disease prediction ability but is not sufficiently robust for translation into clinical practice.