TABLE 5.

The 10-fold cross-validation c-index (×100) from five survival models (GRS, LASSO, Ridge, RSF, DNN) using different p-value cut-offs in the AREDS and AREDS2 data

	Number of predictors	GRSa	LASSO	Ridge	RSF	DNN	Time (minutes)
p <10−7	92	73.2 (1.6)	72.4 (1.7)	72.3 (1.7)	68.5 (1.4)	72.2 (1.8)	49
p <10−6	165	73.2 (1.6)	72.6 (1.5)	72.6 (1.5)	68.2 (1.3)	72.6 (1.6)	47
p <10−5	666	73.2 (1.6)	74.4 (1.3)	74.3 (1.3)	70.1 (1.8)	76.1 (1.2)	62
p <10−4	1500	73.2 (1.6)	75.2 (1.1)	74.8 (1.0)	71.1 (1.7)	76.5 (1.4)	77

Note: The last column shows the DNN’s computing time for running on the entire dataset once.

Abbreviations: AREDS, age-related eye disease studies; DNN, deep neural network; GRS, genetic risk score; RSF, random survival forest.

^aGRS is invariant to the choice of p-value cut-offs as it does not use individual SNPs but rather a composite score.