Phenobarbital and diazepam looked equivalent for inpatient alcohol withdrawal

Design

Multicentre target trial emulation using the Premier Healthcare Database with targeted maximum likelihood estimation

Population

24,643 adults with alcohol withdrawal syndrome; 11,919 on phenobarbital and 12,724 on diazepam at admission

Primary outcome

Hospital length of stay and invasive mechanical ventilation

Effect

Length of stay 4.1 vs 4.3 days (difference -0.21, 95% CI -0.42 to 0.00); ventilation 3.1% vs 2.7% (0.4%, -0.7% to 1.4%)

This comparative effectiveness study emulated a randomised trial using the Premier Healthcare Database, covering 754 US acute care hospitals from October 2022 to August 2024. It included 24,643 adults admitted with alcohol withdrawal as the primary diagnosis who started phenobarbital (48.4%) or diazepam (51.6%) on the day of admission. Earlier comparisons mostly used lorazepam, which is short-acting.

Hospital stay did not differ (4.1 vs 4.3 days; difference -0.21, 95% CI -0.42 to 0.00) and neither did invasive ventilation (3.1% vs 2.7%; difference 0.4%, -0.7% to 1.4%). Secondary outcomes were also similar, except that self-directed discharge was less frequent with phenobarbital (9.5% vs 12.6%; difference -3.1%, -4.9% to -1.3%).

The result suggests the two are close in effect, so choice can follow local protocol, severity, liver function, the availability of monitoring and the risk of drug interactions. As an emulated trial from billing data, residual confounding remains possible, and the database cannot show how doses were chosen.

• Either drug is a reasonable option in an inpatient protocol; choice can rest on local experience and monitoring
• Consider liver function when choosing a long-acting agent, since diazepam accumulates in liver disease
• Phenobarbital needs experienced dosing and monitoring, particularly for respiratory depression
• Use a validated withdrawal scale to guide treatment and reassess regularly
• Plan follow-up and treatment for alcohol use disorder before discharge

Why it matters

It compares phenobarbital with a long-acting benzodiazepine rather than lorazepam, which is the fairer comparison.

Don't overread it

Emulated trials from billing data cannot replace randomised trials, and they do not address outpatient or severe, ICU-level withdrawal.

The statistics, in plain English

The upper bound of the length-of-stay confidence interval is exactly zero, so the data are compatible with no difference or a difference of up to about half a day shorter with phenobarbital. A target trial emulation tries to mimic a randomised design with observational data, but only measured differences between groups can be adjusted for.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.