Haloperidol showed probable benefit in both hypoactive and hyperactive ICU delirium, with wide uncertainty

Design

Post hoc Bayesian analysis of the AID-ICU randomised placebo-controlled trial

Population

987 adult ICU patients with delirium (540 hypoactive, 447 hyperactive)

Primary outcome

90-day days alive out of hospital and mortality by delirium subtype

Effect

Mortality difference in risk differences -3.7 percentage points (95% CrI -15.3 to 7.8); days alive out of hospital difference 2.9 days (95% CrI -4.8 to 10.6)

The AID-ICU trial randomised adult ICU patients with delirium to haloperidol or placebo. This post hoc Bayesian analysis asked whether the effect differed by motoric subtype at randomisation. Of 987 participants, 540 (54.7%) had hypoactive and 447 (45.3%) had hyperactive delirium.

Haloperidol was associated with more days alive out of hospital at 90 days in both subtypes, with a 76.5% probability that the effect was larger in hyperactive delirium (difference in mean differences 2.9 days; 95% credible interval -4.8 to 10.6). Mortality was lower in both subtypes with a 73% probability of larger benefit in hyperactive delirium (difference in risk differences -3.7 percentage points, 95% credible interval -15.3 to 7.8). Days alive without delirium or coma and without ventilation showed a similar pattern. There were no relevant differences in serious adverse reactions or rescue medication.

The credible intervals span no difference and meaningful benefit, and this is a post hoc subgroup analysis in an ICU population, not older ward patients. The authors call for confirmation. It is hypothesis-generating and does not change delirium care for older adults outside the ICU, where non-drug measures and medicine review stay first.

• Haloperidol had a moderate to high probability of benefit in both hypoactive and hyperactive ICU delirium.
• The probability of larger benefit in hyperactive delirium was 73% to 77%, with wide credible intervals.
• No relevant difference in serious adverse reactions or rescue drugs.
• This was an ICU population; do not extrapolate to ward or nursing home delirium.
• Non-drug measures and medicine review remain the first step.

Why it matters

It informs whether motoric subtype should influence antipsychotic use, a question clinicians face often.

Don't overread it

A post hoc Bayesian subgroup analysis cannot show that haloperidol works better in either subtype or that it should be used in non-ICU delirium.

The statistics, in plain English

A Bayesian probability of 73% for larger benefit is not the same as statistical significance; it says that, given the data and the model, benefit looked more likely than not, but a credible interval running from -15.3 to 7.8 points means no difference or even harm remain plausible. Post hoc subgroups are more fragile than the trial's main result.

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.