Table 2.
Performance evaluation of PRSs derived from different approaches in validation datasets
| PRS method | Parametera | NSNP | JSCRC GWAS of EAS population | CORSA GWAS of EUR population | ||||
|---|---|---|---|---|---|---|---|---|
| AUCb | OR (95% CI)c | Pc | AUCb | OR (95% CI)c | Pc | |||
| GWAS-reported | EUR | 140 | 0.511/0.510 | 1.04 (0.95, 1.14) | 0.432 | 0.629/0.638 | 1.65 (1.51, 1.81) | 1.49E−28 |
| EAS | 37 | 0.577/0.580 | 1.33 (1.21, 1.46) | 2.01E−09 | 0.513/0.506 | 1.02 (0.94, 1.11) | 0.567 | |
| C+T | 5.00E−08 (0.001) | 38 | 0.569/0.573 | 1.29 (1.18, 1.42) | 6.73E−08 | 0.579/0.583 | 1.33 (1.23, 1.45) | 1.77E−11 |
| 5.00E−06 (0.001) | 88 | 0.569/0.575 | 1.30 (1.18, 1.43) | 3.30E−08 | 0.589/0.597 | 1.39 (1.28, 1.51) | 4.02E−14 | |
| 5.00E−04 (0.001) | 784 | 0.591/0.597 | 1.44 (1.31, 1.58) | 5.39E−14 | 0.559/0.567 | 1.27 (1.16, 1.38) | 3.51E−08 | |
| 0.05 (0.001) | 7128 | 0.611/0.618 | 1.52 (1.38, 1.68) | 1.52E−17 | 0.556/0.556 | 1.23 (1.13, 1.33) | 1.65E−06 | |
| 5.00E−08 (0.01) | 39 | 0.570/0.573 | 1.29 (1.18, 1.42) | 8.02E−08 | 0.572/0.574 | 1.30 (1.20, 1.42) | 5.96E−10 | |
| 5.00E−06 (0.01) | 92 | 0.571/0.577 | 1.30 (1.18, 1.42) | 4.54E−08 | 0.583/0.590 | 1.35 (1.24, 1.47) | 4.36E−12 | |
| 5.00E−04 (0.01) | 854 | 0.588/0.593 | 1.42 (1.30, 1.57) | 2.62E−13 | 0.558/0.564 | 1.25 (1.15, 1.36) | 1.04E−07 | |
| 0.05 (0.01) | 13,989 | 0.587/0.592 | 1.37 (1.25, 1.50) | 4.12E−11 | 0.555/0.553 | 1.21 (1.12, 1.32) | 4.89E−06 | |
| 5.00E−08 (0.1) | 48 | 0.573/0.577 | 1.31 (1.20, 1.44) | 1.02E−08 | 0.581/0.581 | 1.33 (1.22, 1.44) | 3.99E−11 | |
| 5.00E−06 (0.1) | 116 | 0.579/0.584 | 1.34 (1.22, 1.47) | 7.91E−10 | 0.592/0.597 | 1.39 (1.28, 1.51) | 3.42E−14 | |
| 5.00E−04 (0.1) | 992 | 0.597/0.602 | 1.46 (1.33, 1.61) | 6.02E−15 | 0.573/0.577 | 1.31 (1.20, 1.42) | 3.22E−10 | |
| 0.05 (0.1) | 27,032 | 0.604/0.608 | 1.52 (1.38, 1.68) | 7.05E−18 | 0.568/0.573 | 1.29 (1.19, 1.40) | 2.61E−09 | |
| LDpred | 1 | 883,144 | 0.611/0.616 | 1.55 (1.40, 1.70) | 8.25E−19 | 0.560/0.567 | 1.27 (1.17, 1.38) | 2.13E−08 |
| 0.3 | 883,144 | 0.612/0.617 | 1.56 (1.41, 1.71) | 3.15E−19 | 0.560/0.567 | 1.28 (1.18, 1.39) | 8.60E−09 | |
| 0.1 | 883,144 | 0.614/0.619 | 1.58 (1.43, 1.74) | 3.26E−20 | 0.567/0.574 | 1.31 (1.20, 1.42) | 4.61E−10 | |
| 0.03 | 883,144 | 0.621/0.626 | 1.64 (1.48, 1.80) | 6.87E−23 | 0.586/0.595 | 1.39 (1.27, 1.51) | 6.45E−14 | |
| 0.01 | 883,144 | 0.633/0.638 | 1.68 (1.52, 1.85) | 7.86E−25 | 0.602/0.608 | 1.47 (1.35, 1.60) | 2.04E−18 | |
| 0.003 | 883,144 | 0.495/0.491 | 0.98 (0.89, 1.07) | 0.627 | 0.514/0.513 | 1.02 (0.94, 1.11) | 0.663 | |
| 0.001 | 883,144 | 0.508/0.509 | 1.04 (0.95, 1.14) | 0.436 | 0.491/0.490 | 0.95 (0.88, 1.04) | 0.257 | |
| 3.00E−04 | 883,144 | 0.499/0.499 | 0.99 (0.91, 1.09) | 0.885 | 0.493/0.491 | 0.98 (0.91, 1.07) | 0.704 | |
| 1.00E−04 | 883,144 | 0.487/0.489 | 0.94 (0.86, 1.03) | 0.202 | 0.510/0.508 | 1.04 (0.96, 1.13) | 0.343 | |
| 3.00E−05 | 883,144 | 0.494/0.498 | 0.98 (0.89, 1.07) | 0.670 | 0.501/0.507 | 1.03 (0.95, 1.12) | 0.464 | |
| 1.00E−05 | 883,144 | 0.480/0.482 | 0.95 (0.87, 1.04) | 0.277 | 0.505/0.500 | 1.02 (0.94, 1.11) | 0.653 | |
| Lassosum | Optimal | 5984 | 0.606/0.610 | 1.51 (1.37, 1.66) | 4.53E−17 | 0.601/0.605 | 1.45 (1.33, 1.58) | 2.12E−17 |
| LDpred2 | Auto | 890,687 | 0.570/0.573 | 1.30 (1.19, 1.43) | 2.36E−08 | 0.557/0.563 | 1.24 (1.14, 1.35) | 3.19E−07 |
| PRS-CSx# | Auto | 1,145,689 | 0.639/0.646 | 1.73 (1.56, 1.91) | 7.19E−27 | 0.602/0.608 | 1.48 (1.36, 1.62) | 5.18E−19 |
EAS East Asian population, EUR European population, PRS polygenic risk score, C+T Clumping and P value thresholding, AUC area under the receiver operating characteristics curve, 95% CI 95% confidence interval, OR odds ratio, SD standard deviation, GWAS genome-wide association study, SNP single nucleotide polymorphism, CORSA Colorectal Cancer Study of Austria
aParameter for SNP selection: population for GWAS-reported variants; P value (LD r2) for C+T method; fraction for LDpred method; optimal parameter for lassosum method, auto parameter for LDpred2, and PRS-CSx methods
bCrude AUC/covariates-adjusted AUC
cOR (95% CI) per SD, derived from logistic model with the adjustment of sex, age, and principal components
#The optimal PRS was highlighted in bold