Table 7.

Concordance Index with 95% confidence intervals from 1000 bootstraps on the test set. The DACMVA framework for cross-modal variational auto-encoders with oversampling (CM-VAE-OV), adversarial training (CM-VAE-Adv), or both (CM-VAE-Adv-OV) is compared to the other baselines in varying data regimes.

Missing Percent	Multisurv	OV	TDImpute	TDImpute-OV	CM-VAE-OV	CM-VAE-Adv	CM-VAE-Adv-OV
4	0.681 ± 0.036	0.691 ± 0.034	0.688 ± 0.035	0.667 ± 0.033	0.700 ± 0.033	0.693 ± 0.032	0.703 ± 0.034
90	0.640 ± 0.036	0.666 ± 0.033	0.682 ± 0.033	0.675 ± 0.033	0.673 ± 0.036	0.685 ± 0.034	0.684 ± 0.033
95	0.598 ± 0.035	0.622 ± 0.038	0.630 ± 0.035	0.621 ± 0.035	0.639 ± 0.034	0.624 ± 0.036	0.636 ± 0.034