The impact of projected increases in obesity prevalence on incident liver disease in the UK: Insights from Bayesian-network modelling

Obesity is a complex multi-system disease and a growing public health challenge. Recent projections estimate that 71% of the adult population in England will be overweight or obese by 2040. Obesity is associated with non-alcoholic fatty liver disease and non-alcoholic steatohepatitis (NASH) and a complex set of variables that interact in myriad way. To address this complexity, we use Bayesian-network (BN) analysis to explore the impact of projected increases in obesity prevalence on magnetic resonance imaging (MRI) derived measures of liver steatosis and fibroinflammation in a large UK-based cohort. BNs explicitly model conditional dependencies as a directed acyclic graph, where network variables denote biomarkers and the directed edges connecting such variables the direction of causality. MRI derived measures of liver steatosis (proton density fat fraction [PDFF]) and fibroinflammation (corrected T1 [cT1]) were obtained from 27,002 participants in the UK Biobank. Liver data was combined with additional participant data to construct a BN. The probabilities of being normal weight, overweight, obese, and severely obese in our population were then fixed to 2040 projections. This resulted in an 8% (2160 participants), 5% (1350 participants) and 3% (800 participants) increase in the probability of severe steatosis (PDFF >10%), moderate steatosis (PDFF 5.6-10%) and NASH, respectively. BN analysis models complexity of liver disease and, in this case, illustrates a ĝbest case scenario' of the impact of projected increases in obesity on liver disease. This disproportionately impacts areas of low socio-economic status, and the cost associated with increasing rates of NASH.