Mortality risk in patients with non-alcoholic fatty liver disease by phenotype cluster
| Cluster | Absolute number of deaths | Total follow-up duration (years) | Incidence rate per 1000 person-years (95% CI) | Outcome: all-cause mortality (HR, 95% CI)* | ||
| Crude N=1651 Events=678 | Model 1 N=1651 Events=678 | Model 2 N=1622 Events=663 | ||||
| Average | 529 | 31 707.083 | 16.7 (15.3 to 18.2) | 1 | 1 | 1 |
| Lipid-liver | 71 | 3685.417 | 19.3 (15.1 to 24.3) | 1.17 (0.91 to 1.50) | 1.11 (0.86 to 1.42) | 1.11 (0.86 to 1.42) |
| Anthro-SBP-glucose | 78 | 1766.000 | 44.2 (34.9 to 55.1) | 2.94 (2.32 to 3.74) | 2.97 (2.34 to 3.78) | 2.88 (2.26 to 3.67) |
In the average group, there was one observation with missing data on follow-up duration.
*Cox proportional HRs; model 1 included sex. Model 2 included, in addition to sex, smoking status (current, no vs yes); years of education (numeric variable); and household income (<US$20 000 vs ≥US$20 000 in the last year). For the crude model, the global test for the proportional-hazard assumption was not significant (p=0.078), suggesting this assumption holds, which is also supported by the Schoenfeld residuals plot (online supplemental figure 2). The exposure variable for all models (crude, model 1 and model 2) was a three-level variable derived from the cluster analysis; in other words, the cluster analysis derived a new variable with three levels which we called phenotypes (average, lipid-liver and anthro-SBP-glucose) whose profiles were described in figure 1. Neither of the three models included the predictors used to derive the clusters or phenotypes. The predictors were used to derive and characterise the phenotypes only (not to adjust the models).