those outside the top decile, and the OR increased to approximately 4.50 when using top 2% of the PRS to identify high-risk individuals (batch 1: OR = 4.62, P value = 7.47E−31; batch 2: OR = 4.60, P value = 1.80E−69; batch 3: OR = 4.35, P value = 1.43E−12). The corresponding prevalence-adjusted PPVs ranged from 0.36 to 0.38, and the prevalence-adjusted NPVs were 0.87 across the three batches (Table 4; Additional File 2: Table S5). Overall, the trans-ancestry PRS was slightly more predictive in this East Asian sample compared with the eMERGE European samples, likely reflecting the contributions from the large BBJ T2D GWAS in the training dataset, the more homogeneous community-based TWB samples relative to the eMERGE study in which sample characteristics may vary across different health care systems, and differences in T2D case-control definitions (self-report vs. EMR-based phenotyping).Table 4Prediction accuracy of the trans-ancestry T2D PRS in the Taiwan Biobank (TWB)BatchLiability R2Covariates-adjusted AUCOR per SD(95% CI)Top 2% PRSOR(95% CI)P valueSensitivitySpecificityAdjusted PPV*Adjusted NPV*Batch 115.1%0.702.19(2.05, 2.33)4.62(3.56, 5.99)7.47E−310.070.980.370.87Batch 212.9%0.682.01(1.93, 2.10)4.60(3.88, 5.45)1.80E−690.070.980.380.87Batch 315.3%0.702.16(1.96, 2.38)4.35(2.89, 6.53)1.43E−120.060.980.360.87*Adjusted PPV and NPV are calculated using 13.7% for the Asian population
