One drug came through 158 delirium-prevention trials with the evidence intact
A network meta-analysis of 41,084 surgical patients over 60 ranked 52 drug interventions. Dexmedetomidine held up after high-risk trials were excluded; almost nothing changed length of stay or mortality.
Delirium after surgery is common, distressing and associated with worse recovery, and the search for a drug that prevents it has generated a very large and very messy literature. A network meta-analysis published in The BMJ pulls that literature together and asks which candidates survive contact with the trials least likely to be biased.
The scale of the evidence
The review searched Embase, Medline and the Cochrane Library to 4 March 2024 for randomised trials of one or more drugs given to prevent delirium after surgery requiring general or regional anaesthesia, in participants at least 60 years old, with delirium measured by a validated tool [s1]. Trials of surgery under local anaesthesia only, of patients ventilated preoperatively, and of treatments for existing delirium were excluded [s1].
That yielded 158 trials with 41,084 participants comparing 52 drug interventions [s1]. Seventeen trials were rated at high risk of bias [s1]. Screening, extraction and quality assessment were done in duplicate by assessors masked to each other's decisions, using Cochrane risk of bias version 2 and CINeMA, and the synthesis was a Bayesian arm-based network meta-analysis [s1].
The overall risk of delirium after surgery across the dataset was 14.5%, or 5,957 people [s1].
What worked
Restricting to trials not at high risk of bias, the most effective interventions at preventing delirium were [s1]:
- Dexmedetomidine, odds ratio 0.46 (95% credible interval 0.36 to 0.57)
- Corticosteroids, 0.53 (0.31 to 0.87)
- Melatonin receptor agonists, 0.54 (0.34 to 0.85)
- Parecoxib, 0.34 (0.16 to 0.74)
- Olanzapine, 0.27 (0.07 to 0.94)
- Intranasal insulin, 0.13 (0.04 to 0.34)
Read the credible intervals rather than the point estimates. Olanzapine's runs from 0.07 to 0.94 — consistent with a very large effect and with a marginal one. Intranasal insulin's lower bound of 0.04 reflects a small evidence base rather than an extraordinary drug.
The authors' own conclusion separates one intervention from the rest: dexmedetomidine is effective in preventing postoperative delirium, and the finding remains after excluding studies at high risk of bias [s1]. Corticosteroids, melatonin receptor agonists, parecoxib, intranasal insulin and olanzapine have potential benefit, but the evidence is of moderate to very low quality [s1].
What did not follow
Preventing an episode of delirium is worth doing in itself. Whether it changes what happens afterwards is a separate question, and the answer here is largely no.
Most interventions had no effect on length of stay, mortality, cognition or quality of life [s1]. Only corticosteroids reduced the severity of delirium, by a mean difference of −2.42 points (95% credible interval −4.72 to −0.12) on the Memorial Delirium Assessment Scale [s1] — a small effect whose interval nearly touches zero.
On harms, hypotension and bradycardia were more common with dexmedetomidine, though postoperative nausea and vomiting were reduced [s1]. Postoperative infection rates were not increased by corticosteroids [s1], which is the specific concern that has limited their use in this setting.
That harm profile matters for the drug the review endorses. Dexmedetomidine's cardiovascular effects are predictable rather than idiosyncratic, and they occur in exactly the older, frailer surgical population most at risk of delirium in the first place.
A problem the authors name
The review ends with an unusual complaint about the field it reviews: evidence synthesis in this area is complicated by inadequate trial registration practices and incomplete adoption of core outcome sets [s1].
Both parts of that sentence have consequences. Without prospective registration, a reviewer cannot tell whether the reported outcome was the planned one. Without core outcome sets, 158 trials measure overlapping but non-identical things, and a network meta-analysis has to assume comparability it cannot verify.
The registration is PROSPERO CRD42023488337 [s1]. A linked BMJ editorial published the same day addresses the prevention question directly [s2].
Where prevention meets detection
A retrospective cohort study published three weeks earlier makes a point that sits alongside the drug question rather than competing with it. Across 3,781 patients aged at least 70 undergoing elective non-cardiac, non-cranial surgery under general anaesthesia at a single tertiary academic hospital between 2020 and 2022, delirium was assessed with the 4AT and supplemented by chart review [s3].
Delirium was diagnosed in the post-anaesthesia care unit in 9.1% of patients (341 of 3,781) [s3]. Those patients were older — median 77.1 years (IQR 73.7 to 82.7) versus 75.7 (72.9 to 80.6) — and more often frail (52% versus 40%), cognitively impaired (36% versus 18%), and undergoing high-risk (52% versus 27%) and longer surgery, at a median 196 minutes (133 to 275) versus 142 (95 to 204) [s3].
The authors conclude that delirium detected in the recovery unit is a strong early marker of risk for later ward delirium and other complications, and that routine screening there enables early diagnosis and risk stratification [s3]. They also state the obvious untested step: whether intervening during the recovery-unit stay improves the eventual trajectory [s3].
This is a single-centre retrospective study, so the association it reports cannot be read as a demonstration that screening changes outcomes.
What to watch
The practical shape of the evidence is now reasonably clear. One sedative has a defensible effect on the incidence of delirium with a known cardiovascular cost [s1]; nothing reliably shifts the downstream outcomes that matter most [s1]; and a cheap screening step identifies the highest-risk patients before they leave the recovery unit [s3]. The trial that has not been done is the one combining them.
This article describes research findings and is not guidance about any medicine or procedure.
Sources
- [s1] Effectiveness of drug interventions to prevent delirium after surgery for older adults: systematic review and network meta-analysis of randomised controlled trials. BMJ, 12 February 2026. https://doi.org/10.1136/bmj-2025-085539
- [s2] Preventing postoperative delirium. BMJ, 12 February 2026. https://doi.org/10.1136/bmj.s247
- [s3] Association between postoperative delirium in the postanaesthesia care unit and subsequent ward delirium: a retrospective cohort study. European Journal of Anaesthesiology, 22 January 2026. https://doi.org/10.1097/EJA.0000000000002351
Sources
- Effectiveness of drug interventions to prevent delirium after surgery for older adults: systematic review and network meta-analysis of randomised controlled trials — BMJ , February 12, 2026
- Preventing postoperative delirium — BMJ , February 12, 2026
- Association between postoperative delirium in the postanaesthesia care unit and subsequent ward delirium: a retrospective cohort study — European Journal of Anaesthesiology , January 22, 2026
More on
Sleep medicine's own guideline suggests clinicians not use melatonin for insomnia
The pooled effect is about seven minutes off the time it takes to fall asleep. The American Academy of Sleep Medicine recommends against it — on evidence it grades as weak in both directions.
For jet lag, the evidence on melatonin is unusually strong. Timing is the catch.
A Cochrane review found melatonin reduced jet lag in nine of ten trials. Sleep medicine's own guideline rates it a first-tier option for jet lag disorder. Take it at the wrong hour and it can make things worse.
Women enrolled unevenly in trials, then dosed as if the difference didn't exist
A systematic review of 27 studies finds women reach higher drug concentrations at the same psychotropic dose. A separate review of 103 atrial fibrillation trials finds 43% under-enrolled women against disease prevalence.
Melatonin is the top cause of accidental medicine exposure in under-sevens, a review finds
A systematic review located 19 articles on melatonin in children aged 0 to 6. Trials covered 167 children, all with neurologic conditions. There was no efficacy data for typically developing children.