Table 4.
Influence of GWAS Sources on Breast Cancer PRS Performance in MGI
| GWAS Source (Effective Sample Size)a | Method Tuning Parameter | # SNPs | Pseudo-R2 | Brier Score | AAUC (95% CI) | Odds Ratio Continuous PRSb | Odds Ratio Top 1%c |
|---|---|---|---|---|---|---|---|
| MGI Cohort | |||||||
| Large GWAS Michailidou et al.11 (113,845) | Lassosum: s = 0.5, λ = 0.0043 | 118,388 | 0.0592 | 0.134 | 0.641 (0.625,0.656) | 1.70 (1.59,1.81) | 2.48 (1.63,3.77) |
| GWAS Catalog (N/A) | P&T: p ≤ 3.2 × 10−9 | 62 | 0.0346 | 0.136 | 0.607 (0.589,0.623) | 1.49 (1.40,1.58) | 3.06 (2.04,4.6) |
| UKB GWAS PHECODE (23,839) | Lassosum: s = 0.5, λ = 0.014 | 6,977 | 0.0340 | 0.137 | 0.609 (0.592,0.624) | 1.48 (1.39,1.57) | 1.94 (1.21,3.11) |
| UKB GWAS FINNGEN (18,376) | Lassosum: s = 0.5, λ = 0.018 | 2,267 | 0.0300 | 0.137 | 0.600 (0.584,0.616) | 1.44 (1.35,1.53) | 2.37 (1.54,3.65) |
| UKB GWAS ICD10 (15,792) | Lassosum: s = 0.5, λ = 0.018 | 4,047 | 0.0264 | 0.137 | 0.595 (0.579,0.610) | 1.40 (1.32,1.49) | 2.05 (1.30,3.24) |
| UKB GWAS PHESANT (15,282) | Fixed threshold: p ≤ 5 × 10−8 | 22 | 0.0204 | 0.138 | 0.579 (0.561,0.597) | 1.34 (1.27,1.43) | 2.32 (1.49,3.61) |
| UKB Cohort | |||||||
| Large GWAS Michailidou et al.11 (113,845) | Lassosum: s = 0.9, λ = 0.0043 | 286,144 | 0.0487 | 0.0807 | 0.643 (0.637,0.65) | 1.70 (1.65,1.75) | 4.02 (3.46,4.67) |
| GWAS Catalog (N/A) | P&T: p ≤ 2.5 × 10−8 | 79 | 0.0226 | 0.0819 | 0.598 (0.59,0.605) | 1.43 (1.39,1.46) | 2.68 (2.25,3.18) |
Italic values indicate best performing PRS according to the corresponding metrics for MGI or UKB.
a
Effective sample size: 4 / (1/#cases + 1/#controls); n/a: not available; references of studies contributing to GWAS Catalog PRS are listed in Table S4.
b
PRSs were scaled to mean = 0 and SD = 1.
c
Top 1% versus rest.