Gyr 1996.

Methods	Double‐blind, multi‐centre, parallel‐arm, placebo controlled RCT.
Participants	49 participants with cirrhosis and chronic overt hepatic encephalopathy (Grades I to III; Table 3). Mean age ± SD: flumazenil: 55.5 ± 9.4 years; placebo: 53.6 ± 10.3 years. Proportion of men: 69%. Aetiology of cirrhosis: alcohol 51%; hepatitis B/C 35%. Proportion testing positive for benzodiazepines at baseline (Table 6): 11% in flumazenil group; 5% in placebo group.
Interventions	Intervention comparison: intravenous boluses of flumazenil 0.4 mg, 0.8 mg, and 1 mg at 1‐minute intervals followed by a 3‐hour infusion of flumazenil 1 mg/hour versus placebo (saline). Total dose of flumazenil: 5.2 mg. Cointerventions: none reported.
Outcomes	Outcomes included in meta‐analyses: mortality, hepatic encephalopathy (Table 2), and serious adverse events (Table 7) assessed after a maximum of 4 weeks.
Neuropsychiatric assessment	Baseline and post infusion: Clinical assessment of mental status (Table 3) at baseline and every 30 minutes for 5 hours then every 1 hour until 12 hours post infusion; Continuous (20 minutes) electroencephalography immediately after the infusion and then at 2 hours 40 minutes, 3 hours, and 7 hours 40 minutes post infusion (Table 4).
Inclusion period (date)	Not reported.
Country	Switzerland (primary), France, Germany, Italy, Canada, the Netherlands, the UK, and Korea.
Notes	Included data: authors reported intention‐to‐treat analyses including all participants randomised and a per‐protocol analysis excluding protocol violators (25 participants). We included data on all participants in our analyses.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated list of random numbers.
Allocation concealment (selection bias)	Low risk	Sealed envelopes used in double‐blind administration of flumazenil and placebo.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinding of participants and personnel.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Blinding of outcome assessment.
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing outcome data.
Selective reporting (reporting bias)	Low risk	Trial described clinically relevant outcomes. We had no access to information about outcomes described in the original protocol or information in trial registries.
For‐profit funding	High risk	Support from Hoffmann‐La Roche Ltd.
Other bias	Low risk	No other biases.
Overall assessment	High risk	High risk of bias.