Table 5

Multiple models testing prediction of long-term outcomes by serum ferritin combined with NFS to predict long-term outcomes

Method	Liver-related events (93/1179)	HCC (22/1177)	Cardiovascular events (123/1117)	Extrahepatic cancer (84/981)	Mortality (60/1189)
Cox regression	0.723±0.009*	0.885±0.002	0.682±0.005	0.604±0.001	0.65±0.02*
Lasso-penalised Cox regression	0.722±0.009*	0.884±0.003*	0.689±0.004	0.604±0.001	0.63±0.012*
Ridge-penalised Cox regression	0.722±0.009*	0.885±0.002	0.682±0.007	0.599±0.002	0.636±0.015*
Elastic net-penalised Cox regression	0.722±0.008*	0.884±0.002	0.689±0.004	0.603±0.001	0.63±0.011*

Median Harrell’s C-indices with corresponding standard errors of the median estimated on the test sets of the cross-validation. Cox proportional hazard models and its penalised variants (i.e., Lasso, Ridge and Elastic Net) were applied. The ratio number of events/total number for each outcome is reported on top into brackets.
*Reporting statistically higher C-indices with respect to the same model tested on the same cross-validation split but considering only confounders, according to the one-sided Wilcoxon signed rank test (p value <0.05).
†P value <0.1 on the same test.
HCC, hepatocellular carcinoma; NFS, NAFLD Fibrosis Score.