Abstract
Background: Acute hepatic dysfunction in the critically ill population with pre-existing liver cirrhosis is associated with a high mortality. Several prediction models have been developed to risk stratify patients with liver disease. Methods: This UK dual-centre non-specialist hospital retrospective study (2015–2019) externally validated the Liver injury and Failure evaluation score (incorporating lactate, bilirubin and International Normalised Ratio), alongside two other general intensive care unit prediction models (Intensive Care National Audit and Research Centre and Acute Physiology and Chronic Health Evaluation II). Inclusion criteria matched a recent UK-wide study including at least one of biopsy proven cirrhosis, imaging suggestive of cirrhosis, hepatic encephalopathy or portal hypertension. Results: One hundred and ninety-nine admissions met inclusion criteria over the study period (n¼169), mean age 57(13). In-hospital mortality was 40% in this cohort compared to 18% of all intensive care unit individuals during the same period. Variceal bleeding was associated with a lower short-term (18% versus 47%, P<0.001, odds ratio 0.3 (95% confidence interval 0.1–0.5)) and longer-term mortality (log rank P¼0.015). In-patient mortality was higher in cases requiring renal replacement therapy (82% versus 29%, odds ratio 11.1 (95% confidence interval 4.6–26.9), P<0.001) or ventilation (47% versus 32%, odds ratio 1.9 (1.1–3.4), P¼0.03). For in-patient mortality, area under the receiver operating characteristic curves were Liver injury and Failure evaluation 0.69 (95% confidence interval 0.62–0.77), Intensive Care National Audit and Research Centre 0.80 (0.74–0.86) and Acute Physiology and Chronic Health Evaluation II 0.73 (0.65–0.81). Forty-one per cent of cases were alive at one-year follow-up. Area under the receiver operating characteristic curves for one-year survival were Liver injury and Failure evaluation 0.69 (0.61–0.77), Intensive Care National Audit and Research Centre 0.75 (0.67–0.82) and Acute Physiology and Chronic Health Evaluation II 0.69 (0.61–0.77). Conclusion: This first Liver injury and Failure evaluation score validation in a UK non-specialist hospital setting suggests this parsimonious, easy to calculate model may have utility in prediction of short-term and one-year mortality. As with previous studies variceal haemorrhage was associated with lower mortality.