Design: randomised, parallel-group, double-blind, placebo-controlled trial at 21 hospitals in Denmark
Population: 779 adults with in-hospital cardiac arrest who had received at least one dose of adrenaline; median age 73
Primary outcome: sustained return of spontaneous circulation
Effect: 39% vs 37%; RR 1.05 (95% CI 0.88–1.24), p=0.62

Sodium bicarbonate is given during cardiac arrest in many hospitals on physiological reasoning rather than evidence. This Danish trial across 21 hospitals randomised adults having an in-hospital cardiac arrest who had received at least one dose of adrenaline to up to 100 mmol of sodium bicarbonate or placebo, double-blind. Of 913 randomised, 779 entered the primary analysis.

Sustained return of spontaneous circulation occurred in 39% with bicarbonate and 37% with placebo (RR 1.05, 95% CI 0.88–1.24, p=0.62). Survival at 30 days was 12% versus 9.1% and favourable neurological survival 8.1% versus 5.4% — both with confidence intervals crossing 1.0. Alkalosis and hypernatremia were more common after bicarbonate.

The secondary numbers are the ones worth being careful about. A 30-day survival of 12% against 9.1% looks like something, and the confidence interval (0.84 to 1.88) says the trial cannot distinguish it from chance. The honest summary is that routine bicarbonate has no demonstrated benefit and a demonstrated metabolic cost, and that the specific situations where it is defensible — known hyperkalaemia, tricyclic overdose — were never what this trial was about.

Remove bicarbonate from the default arrest sequence; it is not a routine drug.
Keep it for its specific indications — hyperkalaemia and sodium channel blocker toxicity — where the rationale is not acid-base correction.
Check a gas and a potassium during a prolonged arrest rather than treating an assumed acidosis empirically.
Audit what your arrest trolley prompts the team to reach for; habit, not protocol, is usually what drives this.

Why it matters

It removes a drug that many arrest teams still give reflexively, and names a harm for doing so.

The statistics, in plain English

A risk ratio of 1.05 with an interval from 0.88 to 1.24 is a genuinely flat result: the trial had enough events on the primary endpoint to say the effect, if any, is small. The survival results are different — 45 deaths against 37 is a small number of events, so the interval is wide and the trial is underpowered for them. Wide intervals on a secondary endpoint do not mean a hidden benefit; they mean the question was not answered.