Table 1.

Performance of different PRS models in five-fold cross-validation of OCAC data.

Model	Number of SNPsa	Tuning parameter for best performance	AUC	OR per 1 SD of PRS	95% CI
(a) Models based on individual level genotype data
Lasso	1403	λ = 3.3	0.583	1.35	1.30–1.39
Elastic net	10,797	λ = 3.3, κ = −2.2	0.586	1.36	1.31–1.40
MCP	1403	λ = 3.3	0.583	1.35	1.30–1.39
(b) Models based on summary statistics
LDPred	5,291,719	ρ = 0.001	0.552	1.21	1.13–1.29
Stepwise	22	λ = 5.4	0.572	1.30	1.26–1.34
Select and Shrink (OCAC)	27,240	a = 2.75, b = 2, φ = 3e−6	0.593	1.39	1.34–1.44

AUC area under the receiver operating characteristic (ROC) curve AUC), OR odds ratio, SD standard deviation, PRS polygenic risk score, CI confidence interval, NA not applicable.

^aNumber of SNPs in PRS model run on full OCAC data set after selection of model parameters.