Predicting liver-related events in NAFLD: A predictive model.

Abstract

Management of NAFLD involves noninvasive prediction of fibrosis, which is a surrogate for patient outcomes. We aimed to develop and validate a model predictive of liver-related events (LREs) of decompensation and/or HCC and compare its accuracy with fibrosis models. Patients with NAFLD from Australia and Spain who were followed for up to 28 years formed derivation (n = 584) and validation (n = 477) cohorts. Competing risk regression and information criteria were used for model development. Accuracy was compared with fibrosis models using time-dependent AUC analysis. During follow-up, LREs occurred in 52 (9%) and 11 (2.3%) patients in derivation and validation cohorts, respectively. Age, type 2 diabetes, albumin, bilirubin, platelet count, and international normalized ratio were independent predictors of LRE and were combined into a model [NAFLD outcomes score (NOS)]. The NOS model calibrated well [calibration slope, 0.99 (derivation), 0.98 (validation)] with excellent overall performance [integrated Brier score, 0.07 (derivation) and 0.01 (validation)]. A cutoff ≥1.3 identified subjects at a higher risk of LRE, (sub-HR 24.6, p < 0.001, 5-year cumulative incidence 38% vs 1.0%, respectively). The predictive accuracy at 5 and 10 years was excellent in both derivation (time-dependent AUC,0.92 and 0.90, respectively) and validation cohorts (time-dependent AUC,0.80 and 0.82, respectively). The NOS was more accurate than the fibrosis-4 or NAFLD fibrosis score for predicting LREs at 5 and 10 years ( p < 0.001). The NOS model consists of readily available measures and has greater accuracy in predicting outcomes in patients with NAFLD than existing fibrosis models.