Design: randomised, double-blind, placebo-controlled trial
Population: 1,028 adults presenting to the emergency department with acute nontraumatic headache, all given 1 g oral paracetamol before randomisation
Primary outcome: treatment success, defined as a 30% or greater reduction in numerical rating scale score at 30 minutes
Effect: 78.9% vs 65.1% (difference 13.8%, 95% CI 8-19); rescue analgesia 7.1% vs 15.3% (difference -8.2%, -12 to -4.3); adverse events 15.4% vs 11.1%

A double-blind trial randomised 1,028 adults presenting to the emergency department with acute nontraumatic headache. All received 1 g of oral paracetamol first, then either 2 g of intravenous magnesium sulfate over 30 minutes or matching saline placebo. Treatment success — a 30% or greater fall in the numerical rating scale at 30 minutes — was reached by 78.9% on magnesium against 65.1% on placebo.

The authors then do something unusual and admirable: they undercut their own headline. Every timed difference in the numerical rating scale fell below 1.3 points, the accepted threshold for a change a patient would notice. The composite success rate moved substantially; the actual pain barely did.

Two secondary outcomes point the other way and are harder to dismiss. Rescue analgesia was needed by 7.1% on magnesium against 15.3% on placebo, and satisfaction was higher. A halving of rescue analgesia is a real operational and patient-level difference that does not depend on a scale threshold.

Against that, adverse events were more frequent — 15.4% against 11.1%, mild but not nothing, in a patient population that is often discharged within the hour.

Consider intravenous magnesium as an adjunct where rescue analgesia is the problem you are trying to avoid, not where you expect a dramatic fall in pain score
Give the oral paracetamol first; this trial tested magnesium added to it, not instead of it
Warn about mild infusion-related effects; they were about a third more common than with placebo
Exclude the dangerous causes of headache before treating; this trial enrolled nontraumatic headache, which is not the same as benign headache
Record the pain score before and at 30 minutes, or you cannot tell whether it worked

Don't overread it

The success rate rose, but every pain-score difference was below the threshold for a change patients can detect — statistical significance here is not clinical importance.

The statistics, in plain English

The 13.8 percentage point difference in success (95% CI 8 to 19) is statistically solid, but success was defined as a 30% fall in the pain score, and a dichotomous threshold can shift a lot when the underlying distribution moves only slightly — which is exactly what the sub-threshold numerical rating scale differences show. This is a good example of why a responder analysis should always be read alongside the continuous outcome. The rescue-analgesia difference, -8.2 percentage points (95% CI -12 to -4.3), is not threshold-dependent and is the more trustworthy signal of benefit.