Analytical approach

The main baseline characteristics of the study population were shown according to the 3 risk classes of the liver fibrosis scores. Comparison were carried out using ANOVA test or Kruskal-Wallis test for continuous variables, as appropriate, and by χ2 test for categorical variables. Then, to examine the longitudinal association between liver fibrosis scores the risk of disability at 6 years and overall and cause-specific mortality, we performed a log-binomial regression and Cox proportional hazard regressions, respectively. Multivariable models were adjusted for the following potential confounding factors: age, sex, BMI, arterial hypertension, diabetes mellitus, COPD, CHF, total cholesterol, Charlson comorbidity index, physical activity, alcohol consumption, smoking, IL-6 and TNF-alpha. Interaction terms were entered in the models to evaluate whether risk estimates may change across categories of predicted presence of ultrasonographic liver steatosis (as estimated by FLI ≥60) and alcohol consumption (below or above 30 g/day for males and 20 g/day for females). In addition, sensitivity analyses were run including only subjects with FLI≥60 and with low alcohol consumption. The proportional hazard assumption of Cox regressions was tested through the inspection of Schoenfeld residuals.

We also examined the discriminative capacities of liver scores in predicting mortality and disability, calculating the c-statistics of the Cox models and the area under the receiver operating characteristic curve (AUC) of the log-binomial models. Sensitivity and specificity were also calculated. All analyses were performed using R 3.3.3 software for Mac (R Foundation).