A drug that didn't move mortality, couldn't work alone in most patients, and still changed how an entire generation thinks about sedation depth
4,000 mechanically ventilated ICU patients, open-label, dexmedetomidine as the primary sedative vs. usual care.
SPICE III (Sedation Practice in Intensive Care Evaluation) randomized 3,904 critically ill adults expected to need invasive ventilation for more than 24 hours across multiple countries. One arm received early dexmedetomidine as the sole or primary sedative; the other received usual care, meaning propofol, midazolam, or a combination, at clinician discretion.[1] This built directly on SPICE I, which had already shown that deep sedation in the first 48 hours predicts higher mortality, longer ventilation, and more delirium, and that dexmedetomidine was mostly being used as a secondary adjunct days into a course rather than as a starting agent.[2] SPICE III asked what happens if you start with it instead.
90-day all-cause mortality: 29.1% vs. 29.1%. P = .98.
This is as close to a null result as a trial produces. Using dexmedetomidine as the primary early sedative did not lower 90-day mortality compared with usual care, full stop, in the overall intention-to-treat population.[1] There was no difference by sepsis status, by country, by cause of death, or by discharge destination in the prespecified subgroup analyses.
64.7% of the dexmedetomidine arm needed supplemental propofol to hit target sedation.
Dexmedetomidine achieved the target RASS in only 11.5% of usual-care patients receiving it as a rescue agent, versus 97.8% attainment when it was the assigned primary agent, but that number is deceptive: reaching target with dexmedetomidine as primary agent very often meant reaching it with propofol added on top. Supplemental propofol was used in 64.7% of the dexmedetomidine group versus 60.1% of usual care, essentially the same propofol exposure rate regardless of which arm a patient was randomized to.[1,3] Dexmedetomidine, used alone, frequently isn't strong enough to sedate a genuinely agitated ventilated patient.
+1 day free of coma or delirium. +1 ventilator-free day. Both at 28 days.
| Secondary outcome (28 days) | Dexmedetomidine | Usual care | Difference |
|---|---|---|---|
| Median days free of coma/delirium | 24 | 23 | +1.0 (95% CI 0.5–1.5) |
| Median ventilator-free days | 23 | 22 | +1.0 (95% CI 0.4–1.6) |
Both differences were statistically significant but modest in absolute terms, and neither was adjusted for multiple comparisons, which the trial's own investigators and independent reviewers have flagged as a reason to treat them as hypothesis-generating rather than definitive.[4] "Ventilator-free days" is scored as days alive and off the ventilator by day 28, so a death before day 28 counts as zero, which is worth remembering when comparing this metric across trials.
NNH of 43 for a serious adverse event. This is the number worth remembering.
| Adverse event | Dexmedetomidine | Usual care | NNH |
|---|---|---|---|
| Any adverse event | 9.6% | 1.8% | — |
| Any serious adverse event | 2.7% | 0.4% | 43 |
| Serious bradycardia | 0.7% | 0.05% | 154 |
| Serious hypotension | 0.5% | 0.05% | 222 |
| Prolonged sinus arrest | 0.82% | 0.15% | 149 |
All differences P < .01. A number needed to harm of 43 for a serious adverse event, with bradycardia, hypotension, and sinus arrest specifically elevated, is a real signal, not statistical noise from a large sample size. This is the number that should sit next to the "+1 day" headlines whenever dexmedetomidine's benefits get cited.
Younger patients did worse on dexmedetomidine. Older patients did better. The overall null result was hiding two opposite effects.
A pre-specified secondary Bayesian heterogeneity-of-treatment-effect analysis of the same cohort split the population at the median age of roughly 63.7–67 years and found a genuine interaction.[5,6]
Dexmedetomidine associated with lower 90-day mortality: OR 0.83 (95% credible interval 0.68–1.00), with 97.7% Bayesian probability of a mortality benefit across illness-severity categories.
Dexmedetomidine associated with higher 90-day mortality: OR 1.26 (95% credible interval 1.02–1.56), with 98.5% probability of increased mortality, concentrated particularly in younger, non-operative patients.
This is exactly the kind of finding that a single pooled "no difference" headline erases. A companion dose-analysis of the same trial found that in younger patients, escalating propofol dose on top of a steady dexmedetomidine infusion was associated with reduced adjusted mortality (HR 0.59, 95% CI 0.43–0.78), while no comparable mortality signal existed from escalating the dexmedetomidine dose itself in that group.[7] These are secondary, hypothesis-generating analyses, not a second RCT, and should be read that way; but they argue against reading SPICE III as a flat "dexmedetomidine doesn't matter" result.
The primary outcome was flat. Sedation strategy and depth almost certainly matter more than which specific agent is running.
Nearly two-thirds of patients started on dexmedetomidine alone needed it anyway. Starting dexmedetomidine as a monotherapy strategy in a genuinely agitated patient is likely to under-treat them during the gap.
NNH of 43 for a serious adverse event is not a rounding error. Bradycardia and sinus arrest are the ones that end up called overhead.
The subgroup signal (benefit in older patients, harm signal in younger, non-operative patients) is secondary and Bayesian, not primary and frequentist, but it's a large enough effect size to keep in mind rather than dismiss.
All four headline figures in the source post, mortality, supplemental propofol rate, the +1 day secondary outcomes, and the NNH of 43, matched the NEJM primary source and its supplementary appendix exactly. Nothing needed correcting. The one thing worth adding, since it changes how the "no benefit" headline should be read clinically, is the age-based heterogeneity of treatment effect from the trial's own secondary Bayesian analysis: dexmedetomidine trended toward harm in patients under 65 and toward benefit in patients over 65, an effect that was invisible in the pooled primary result.