Algorithms to Identify Alcoholic Hepatitis Hospitalizations in Patients with Cirrhosis.

backgroundAlcoholic hepatitis (AH) is a clinically diagnosed syndrome with high short-term mortality for which liver transplantation may be curative. A lack of validated algorithms to identify AH hospitalizations has hindered clinical epidemiology research.

methodsThis was a retrospective cohort study of patients with cirrhosis using Veterans Health Administration (VHA) data from 2008 to 2015. We randomly sampled hospitalizations based upon abnormal liver tests and administrative codes for acute hepatitis or alcohol-associated liver disease (ALD). Hospitalizations were manually adjudicated for AH per society guidelines. A priori algorithms were evaluated to compute positive predicted value (PPV) and positive likelihood ratio (LR+), and were tested in an external University of Pennsylvania Health System (UPHS) cohort.

resultsOf 368 hospitalizations, 142 (38.6%) were adjudicated as AH. AH patients were younger (55 vs. 58 years, p < 0.001), less likely to have prior cirrhosis decompensation (57% vs. 73.9%, p < 0.001), and had higher AST-to-ALT ratios (median 2.9 vs. 1.9 mg/dL, p < 0.001) and higher bilirubin levels (median 2.9 vs. 1.9 mg/dL, p < 0.001). Algorithms combining clinical laboratory criteria (AST > 85 U/L but < 450 U/L, AST-to-ALT ratio > 2, total bilirubin > 5 mg/dL) and administrative coding criteria yielded the highest PPV (96.4%, 95% CI 87.7-99.6) and the highest LR+ (43.0, 95% CI 10.6-173.5). Several algorithms demonstrated 100% PPV for definite AH in the UPHS external cohort.

conclusionWe have identified algorithms for AH hospitalizations with excellent PPV and LR+. These high-specificity algorithms may be used in VHA datasets to identify patients with high likelihood of AH, but should not be used to study AH incidence.