Table 4.

Performance comparison of RF and survival analysis with different combinations of features. The experiments were conducted with 10-fold cross-validation. The features used in MELD model comprise MELD score, hepatitis, HCC, DX1, age, gender, and BMI, whereas the features selected by RF are the top eight features identified by RF.

RF Model on Derivation Set
Features	AUC	Specificity	Sensitivity
Features selected by RF	0.787 ± 0.185	0.955 ± 0.187	0.653 ± 0.334
Features used by MELD	0.596 ± 0.315	0.707 ± 0.302	0.720 ± 0.395
Survival Analysis on Derivation Set
Features	Concordance Index	Likelihood ratio test	Wald test
Features selected by RF	0.85	p-value = 6e-08	p-value = 2e-05
Features used by MELD	0.695	p-value = 4e-04	p-value = 8e-05