Publication

• Title: Haloperidol for the treatment of delirium in ICU patients
• Acronym: AID-ICU
• Year: 2022
• Journal published in: The New England Journal of Medicine
• Citation: Andersen-Ranberg NC, Poulsen LM, Perner A, Wetterslev J, Estrup S, Hästbacka J, et al. Haloperidol for the treatment of delirium in ICU patients. N Engl J Med. 2022;387:2425-35.

Context & Rationale

• Background

  • ICU delirium is common and is associated with worse short- and long-term outcomes (mortality, length of stay, prolonged organ support, cognitive/functional sequelae).
  • Haloperidol has been widely used to “treat delirium” in ICU practice despite uncertainty about patient-centred benefit and known risks (QTc prolongation/arrhythmia, extrapyramidal effects).
  • Before AID-ICU, randomised evidence in ICU delirium treatment was limited and inconsistent; key prior large ICU RCT evidence suggested no improvement in delirium/coma–free days with antipsychotics (haloperidol/ziprasidone) in critical illness.
  • Systematic reviews and overviews of reviews before this trial generally concluded there was no clear efficacy signal for pharmacological delirium treatment in ICU and highlighted heterogeneity and outcome limitations.
• Research Question/Hypothesis

  • In adult ICU patients with delirium, does intravenous haloperidol (vs placebo) increase the number of days alive and out of hospital within 90 days?
• Why This Matters

  • Haloperidol is a high-frequency ICU intervention for delirium with potential for preventable harm if ineffective.
  • Delirium is a syndrome with uncertain pharmacological targets; robust evidence is required to justify routine antipsychotic exposure in critically ill patients.
  • A pragmatic, blinded, multicentre RCT with patient-centred outcomes can directly inform practice and guideline recommendations.

Design & Methods

• Research Question: Whether intravenous haloperidol, compared with placebo, increases days alive and out of hospital by day 90 among adult ICU patients with delirium.
• Study Type: Investigator-initiated, pragmatic, international, multicentre, randomised, double-blind, placebo-controlled, parallel-group trial with stratification by site and delirium motor subtype; conducted in general ICUs in Denmark, Finland, Italy, Spain, and the United Kingdom (enrolment February 2018 to July 2022).
• Population:
  • Adults (≥18 years) acutely admitted to ICU with delirium diagnosed by a validated screening tool (CAM-ICU or ICDSC).
  • Randomised after identification of delirium; median time from ICU admission to randomisation was 3.9 days (IQR 1.8–9.7) with haloperidol and 4.1 days (IQR 1.6–8.9) with placebo.
  • Key protocol-level exclusions included pregnancy/positive human chorionic gonadotropin; contraindication to haloperidol; habitual antipsychotic use; permanently incompetent patients; delirium assessment not possible; withdrawal from active therapy/brain death. ¹
  • Common trial-screening exclusions reported during enrolment included delirium tremens, involuntary admission status, withdrawal from active therapy, previous antipsychotic use, contraindication to haloperidol, inability to assess for delirium, and lack of consent (country-specific regulations).
• Intervention:
  • Intravenous haloperidol 2.5 mg (0.5 mL) three times daily while delirium was present.
  • Additional 2.5 mg “as-needed” doses were permitted up to a maximum total daily dose of 20 mg of trial medication.
  • Protocolised pausing/resuming: trial medication was paused from the next day after two consecutive negative delirium assessments on the same day (two negative CAM-ICU assessments or two ICDSC scores <4); restarted if delirium recurred.
  • Treatment continued until ICU discharge or death, up to 90 days (and restarted if readmitted to ICU within 90 days).
• Comparison:
  • Placebo (isotonic saline) administered on the same fixed schedule and “as-needed” algorithm as the intervention.
  • Rescue/escape medications (if agitation/behavioural risk persisted after maximal as-needed dosing) could include propofol, a benzodiazepine, or an alpha-agonist infusion at clinician discretion.
  • Other antipsychotic drugs were discouraged during the intervention period; open-label antipsychotic use was tracked.
• Blinding: Double-blind (participants, bedside clinicians, investigators, and statisticians); identical trial drug and placebo preparations with emergency unblinding mechanisms.
• Statistics: A total sample size of 1000 patients was planned to detect an 8% difference in mean days alive and out of hospital within 90 days (two-sided alpha 0.05; 90% power), using cohort-derived assumptions linking the mean difference to lower in-hospital mortality and shorter hospital stay; primary analysis was adjusted for stratification variables, with a prespecified approach for outcomes truncated by death; analysis was primarily intention-to-treat in the enrolled/consented population, with prespecified sensitivity/per-protocol analyses in a detailed statistical analysis plan. ²
• Follow-Up Period: 90 days for the primary endpoint (days alive and out of hospital); longer-term follow-up was prespecified and subsequently reported.

Key Results

This trial was not stopped early. A prespecified interim safety analysis was conducted after 500 participants had been followed to 90 days, without stopping the trial.

• The primary outcome (days alive and out of hospital by day 90) was not significantly different: 35.8 (95% CI 32.9 to 38.6) with haloperidol vs 32.9 (95% CI 29.9 to 35.8) with placebo; adjusted mean difference 2.9 days (95% CI −1.2 to 7.0; P=0.22).
• Mortality by 90 days was lower with haloperidol (36.3% vs 43.3%; adjusted RR 0.84; 95% CI 0.72 to 0.98), but this did not translate into a statistically significant improvement in the primary composite.
• Serious adverse reactions were uncommon and similar overall (2.2% vs 1.9%), with slightly more QTc-prolongation discontinuations in the haloperidol group (2.4% vs 1.4%).

Internal Validity

• Randomisation and Allocation:
  • Centralised randomisation with allocation concealment; stratified by site and delirium motor subtype (hyperactive vs hypoactive).
  • Blinded preparation and identical placebo supported concealment at bedside.
• Drop out or exclusions (post-randomisation):
  • 1000 patients were randomised; 987 were included in the primary analysis population due to exclusion of patients who did not receive trial medication and/or had consent not provided for participation (501 haloperidol; 486 placebo).
  • Primary outcome data were missing for 24 patients (10 haloperidol; 14 placebo) due to withdrawal of consent for data use beyond the intervention period; death status was missing for 1 placebo patient (mortality denominator 485).
  • These exclusions introduce a modified intention-to-treat structure; the absolute proportion affected was small but is directionally relevant for bias assessment.
• Performance/Detection Bias:
  • Double blinding reduced performance and ascertainment bias.
  • Primary outcome components (mortality and hospital days) were objective and less vulnerable to subjective assessment bias.
  • Delirium assessments (CAM-ICU/ICDSC) are operator-dependent and site-dependent; however, delirium-related outcomes were secondary and not the basis of the primary endpoint.
• Protocol Adherence:
  • Median duration of trial medication: 3.6 days (IQR 2.1–6.1) with haloperidol vs 3.3 days (IQR 2.0–5.7) with placebo.
  • Median daily trial drug dose (scheduled): 8.3 mg/day (IQR 6.4–10.8) vs 9.0 mg/day (IQR 7.0–11.3).
  • Any as-needed doses were common and balanced: 72.7% vs 73.7% of patients; median daily total dose including as-needed was 8.9 mg/day (IQR 7.0–11.5) vs 9.3 mg/day (IQR 7.2–11.7).
  • Open-label antipsychotic exposure in ICU was similar: 13.2% vs 13.0%.
• Baseline Characteristics:
  • Groups were well balanced for key prognostic variables (age median 70 vs 71 years; predicted 90-day mortality 34.7% ± 15.4 vs 34.6% ± 15.4; mechanical ventilation at randomisation 63.9% vs 62.8%).
  • Delirium phenotype distribution was comparable (hyperactive delirium 45.3% vs 44.4%).
  • Low prevalence of documented neurodegenerative disease (0.4% vs 0.6%) suggests a selected delirium population relative to many real-world ICUs.
• Heterogeneity:
  • Prespecified subgroup analyses across patient and delirium characteristics did not identify clinically important heterogeneity; interaction statistics were not reported in the main table (subgroup forest plot values not reported).
• Timing:
  • Randomisation occurred a median of ~4 days after ICU admission (3.9 vs 4.1 days), which may represent a relatively established delirium trajectory rather than very early delirium onset.
  • Delirium assessments were frequent (median 2.0 per day in both groups), supporting protocolised “pause/resume” delivery.
• Dose:
  • Trial dosing allowed up to 20 mg/day total; actual median daily totals were ~9 mg/day with substantial between-patient variability.
  • Whether this constitutes an “optimal” delirium-targeting dose is uncertain; clinically, it approximates common ICU practice and maximises pragmatic relevance.
• Separation of the Variable of Interest:
  • Any rescue medication use: 57.5% (288/501) with haloperidol vs 62.1% (302/486) with placebo (adjusted RR 0.93; 99% CI 0.82 to 1.06).
  • Days with rescue medication per patient: mean 2.9 (99% CI 2.3 to 3.5) vs 2.9 (99% CI 2.3 to 3.4); adjusted mean difference 0.1 (99% CI −0.7 to 0.9).
  • Open-label antipsychotics: 13.2% vs 13.0% (suggesting minimal differential contamination).
  • QTc prolongation leading to discontinuation: 2.4% vs 1.4% (a measurable safety-related separation).
• Key Delivery Aspects:
  • Delirium presence and motor subtype were used in randomisation and treatment pausing/resuming, supporting protocol fidelity.
  • However, bedside referral and the relatively low screen-to-randomise ratio suggest local selection processes that could influence enrolled case-mix.
• Crossover:
  • Open-label antipsychotic exposure occurred in ~13% of both groups; this likely attenuates effect size estimates but did not differentially favour one arm.
• Outcome Assessment:
  • Primary endpoint components are objective (death and hospitalisation days); the combined endpoint “days alive and out of hospital” is patient-centred but not delirium-specific.
  • Delirium/coma-free days were secondary and limited to ICU-assessed delirium/coma, without post-ICU delirium measurement.
• Statistical Rigor:
  • Primary analysis used an adjusted approach and accounted for truncation by death; secondary outcomes used 99% confidence intervals, consistent with prespecified multiplicity control.
  • Missing primary outcome data were low (24/1000), but consent-withdrawal mechanisms introduce potential non-random missingness.

Conclusion on Internal Validity: Overall, internal validity appears strong to moderate given robust blinding and objective primary outcome ascertainment with good protocol adherence, but is tempered by a modified intention-to-treat population (post-randomisation exclusions/consent withdrawals) and signals of selection at enrolment.

External Validity

• Population Representativeness:
  • Older critically ill population (median age ~70 years) in mixed general ICUs across five European countries.
  • Low prevalence of documented neurodegenerative disease (≤0.6%) and multiple exclusion pathways (contraindications, chronic antipsychotic use, inability to assess delirium, and local consent constraints) may under-represent ICUs with high dementia burden or frequent baseline cognitive impairment.
  • Median randomisation timing (~4 days after ICU admission) may not represent earliest delirium onset in all settings.
• Applicability:
  • Findings are most applicable to resource-rich ICUs using structured delirium screening (CAM-ICU/ICDSC), with capacity for ECG monitoring and protocolised drug delivery.
  • Generalisation to settings where haloperidol is used primarily for acute agitation/safety crises (rather than delirium syndrome treatment) is limited, as rescue medication use and restraint practice may differ.
  • Generalisation to ICU populations with high rates of chronic antipsychotic exposure, Parkinsonism/Lewy body disease risk, or delirium with prominent substance withdrawal is constrained by exclusions and case-mix.

Conclusion on External Validity: External validity is moderate: the enrolled population and pragmatic ICU delivery support applicability in similar European ICUs, but selection processes and low baseline cognitive comorbidity limit generalisability to broader delirium populations.

Strengths & Limitations

• Strengths:
  • Large (n=1000), multicentre, international, double-blind, placebo-controlled design with pragmatic delivery.
  • Patient-centred primary endpoint incorporating survival and healthcare utilisation (days alive and out of hospital).
  • Clear protocolised dosing algorithm with prepublication of protocol and detailed statistical analysis plan. ^{1 2}
  • Objective ascertainment for mortality and hospitalisation components; balanced co-interventions and low overall serious adverse reaction rates.
• Limitations:
  • Modified intention-to-treat population due to post-randomisation exclusions and consent withdrawals; primary outcome missing for 24 patients.
  • Primary endpoint is not delirium-specific; delirium symptom burden and time-to-resolution were not primary drivers of inference.
  • Potential selection at enrolment (clinician referral/low screen-to-randomise ratio) and concentration of enrolment in a minority of sites may reduce representativeness.
  • Treatment began a median of ~4 days after ICU admission, potentially after delirium was established and less modifiable.

Interpretation & Why It Matters

• Routine haloperidol for ICU delirium

  Haloperidol did not significantly increase days alive and out of hospital at 90 days (adjusted mean difference 2.9 days; 95% CI −1.2 to 7.0), challenging routine syndrome-level treatment strategies where the goal is improved patient-centred recovery rather than short-term sedation.
• Mortality signal vs primary neutrality

  Lower 90-day mortality (36.3% vs 43.3%; adjusted RR 0.84; 95% CI 0.72 to 0.98) occurred alongside a non-significant composite primary endpoint and a numerically longer hospital stay, underscoring the interpretive tension between components of composite, death-truncated outcomes.
• Trial design implications

  Future delirium trials may need (i) earlier enrolment, (ii) delirium phenotype stratification (motor subtype/cognitive vulnerability), and (iii) delirium-centred outcomes paired with long-term functional recovery measures, to test mechanistic benefit rather than only downstream utilisation endpoints.

Controversies & Subsequent Evidence

• Outcome selection and mechanistic inference:
  • The primary endpoint (days alive and out of hospital) is clinically meaningful but not delirium-specific; the accompanying editorial highlighted that the trial did not establish clear delirium-symptom benefit and that similar rescue medication use questions symptomatic efficacy in routine care contexts. ³
• Selection and site contribution concerns:
  • The editorial noted the relatively low number screened-to-enrolled compared with some prior ICU delirium RCTs and that recruitment was concentrated in a subset of centres, raising the possibility of clinician preselection and reduced representativeness. ³
• Mortality signal and long-term follow-up:
  • A prespecified 1-year follow-up analysis reported lower 1-year mortality with haloperidol (43.6% vs 50.1%; adjusted RR 0.87; 95% CI 0.76 to 1.00; P=0.045; hazard ratio 0.81; 95% CI 0.67 to 0.97; P=0.022).
  • In the same follow-up, health-related quality of life at 1 year (EQ-5D-5L index) was not significantly different: overall median 0.3 vs 0.0 with adjusted mean difference 0.04 (95% CI −0.03 to 0.11; P=0.091), and among survivors median 0.85 vs 0.81 with adjusted mean difference 0.00 (95% CI −0.03 to 0.04; P=0.916).
  • EQ-5D-5L data were missing in 16.7%, and sensitivity analyses for missingness were performed; interpretive uncertainty remains regarding mechanisms for the mortality signal in the absence of demonstrable HRQoL improvement.
  • These findings sharpen the central debate: whether the mortality reduction reflects a true, reproducible biological/behavioural effect or residual confounding/selection effects despite randomisation, especially given primary endpoint neutrality.
  • ⁴
• How AID-ICU fits with the broader delirium-treatment evidence base:
  • Earlier overviews of pharmacological interventions for ICU delirium concluded evidence of benefit was uncertain and heterogeneous; AID-ICU adds the largest blinded treatment dataset with a patient-centred primary outcome, but does not resolve whether antipsychotics improve delirium-specific recovery trajectories. ⁵

Summary

• AID-ICU randomised 1000 ICU patients with delirium to intravenous haloperidol vs placebo, using a pragmatic fixed-dose-plus-PRN regimen (max 20 mg/day) with protocolised pausing/resuming.
• The primary outcome was not significantly different: days alive and out of hospital at 90 days were 35.8 vs 32.9 (adjusted mean difference 2.9 days; 95% CI −1.2 to 7.0; P=0.22).
• 90-day mortality was lower with haloperidol (36.3% vs 43.3%; adjusted RR 0.84; 95% CI 0.72 to 0.98), but the composite primary endpoint remained neutral and hospital stay was numerically longer.
• Serious adverse reactions were uncommon and similar overall (2.2% vs 1.9%), with slightly more QTc-related discontinuations on haloperidol (2.4% vs 1.4%).
• Interpretation hinges on whether the mortality signal is causal and reproducible, given limited evidence of delirium-specific symptomatic improvement and potential enrolment selection effects.

Overall Takeaway

AID-ICU is a landmark pragmatic, double-blind, multicentre ICU delirium treatment trial because it tested a widely used antipsychotic strategy against placebo using a patient-centred endpoint and found no statistically significant improvement in days alive and out of hospital at 90 days. The observed mortality signal (at 90 days and in subsequent 1-year follow-up) remains the key interpretive tension, demanding cautious integration with prior evidence and careful attention to enrolment selection, delirium phenotypes, and outcome choice in future trials.

Bibliography

• Andersen-Ranberg NC, Poulsen LM, Perner A, et al. Agents Intervening against Delirium in the Intensive Care Unit (AID-ICU) — protocol for a randomised placebo-controlled trial of haloperidol in patients with delirium in the ICU. Acta Anaesthesiol Scand. 2019;63:1426-33.
• Andersen-Ranberg N, Poulsen LM, Perner A, et al. The Agents Intervening against Delirium in the Intensive Care Unit trial (AID-ICU trial): a detailed statistical analysis plan. Acta Anaesthesiol Scand. 2020;64:1357-64.
• Marcantonio ER. Haloperidol for Treatment of ICU Delirium — Progress or Setback? N Engl J Med. 2022;387:2464-5.
• Mortensen CB, et al. Long-term outcomes with haloperidol versus placebo in acutely admitted adult ICU patients with delirium. Intensive Care Med. 2024;Not reported.
• Barbateskovic M, Kraig Jensen S, et al. Pharmacological interventions for delirium in intensive care patients: an overview of reviews. Acta Anaesthesiol Scand. 2020;64:254-66.