School of Public Health, Peking University, China (People's Republic)
Background: If not controlled effectively, a substantial portion of Non-alcoholic fatty liver disease (NAFLD) patients will progress to cirrhosis or hepatic cancer (HCC), which is irreversible and life-threatening but has no wonder drug. Development of risk stratification tools would enable clinicians to identify high-risk NAFLD patients and develop individualized (risk-based) strategies.
Objectives: To develop a simple model as a prescreening tool to identify NAFLD patients with higher risk of progression to cirrhosis or HCC.
Methods: This was an observational, prospective cohort study. We identified 6152 patients with NAFLD diagnosis from the UK biobank cohort, who were followed up from Jan 2006 to Sep 2021. The study population was then divided into derivation cohort (enrolled between Jan 1, 2006 and May 21, 2009) and validation cohort (enrolled between May 22, 2009 and Sep 30, 2010) with a proportion of 2:1 for temporal external validation. Demographics, lifestyle, laboratory tests, co-morbidities and medication were selected as predictors based on the literature and expert opinions. The Cox proportional hazards model was used with LASSO method to select predictors. Discrimination was evaluated using Harrell’s c-statistics, and calibration was evaluated by calibration curves. Decision curves were also plotted to assess the potential value of the model.
Results: This study included 4104 patients (1921 men [46.8%]; median [IQR] age, 58 [51,63] years) in the derivation cohort and 2048 patients (908 men [44.4%]; median [IQR] age, 59 [51,64] years) in the validation cohort. Cirrhosis or HCC occurred for 393 patients (9.6%), 187 patients (9.1%) in two cohorts respectively. The final model included 8 predictors: age, waistline, AST, GGT, albumin, PLT, LDL-c and diabetes mellitus. The prediction model showed good predictive performance in terms of discrimination (optimism-corrected Harrell’s c-index: 0.736, 95%CI: 0.709, 0.763) and calibration (predictions almost fell on the 45-degree diagonal). Decision curves showed higher net benefit of risk-based screening using the model compared to the “screen-all” approach.
Conclusions: We developed a simple model for NAFLD-related cirrhosis or HCC, which might facilitate the clinicians to stratify a NAFLD patient’s risk of future progression and to develop rational, cost-effective risk categories strategies.
