- Design
- Systematic review and Bayesian random-effects meta-analysis of randomised trials
- Population
- 110 trials, 22,608 children and adults treated with cefepime or another beta-lactam
- Primary outcome
- All-cause mortality, about 30 days
- Effect
- 6.6% vs 6.2%, OR 1.10 (95% CrI 0.98–1.24), 94.4% probability of harm; peer-reviewed trials OR 1.17 (1.02–1.34)
This systematic review pooled 110 randomised trials, with 22,608 children and adults, that compared cefepime with another beta-lactam, alone or with the same second drug. It used a Bayesian random-effects model for all-cause mortality at about 30 days.
Mortality was 6.6% with cefepime and 6.2% with comparators (OR 1.10), a 94.4% probability of harm. Restricted to 73 published peer-reviewed trials, the odds ratio was 1.17 and the probability of harm 98.6%. The direction was consistent across comparators, indications and doses of 2 g or more twice daily.
This revives a concern first raised almost two decades ago. It does not remove cefepime from use: the absolute difference is small, and cefepime remains useful for febrile neutropenia and for organisms that carry AmpC beta-lactamase. But where an equally suitable alternative exists, the balance now tilts away from it, and neurotoxicity in renal impairment is another reason to dose carefully.
- Where piperacillin-tazobactam, ceftazidime or a carbapenem is equally suitable, consider it over cefepime
- Adjust cefepime dose for renal function and watch for confusion, myoclonus or seizures
- Review local febrile neutropenia protocols in light of this signal
- Do not switch patients who are responding well solely because of this analysis
Why it matters
A widely used antibiotic now carries a consistent, if small, mortality signal that guidance has not yet addressed.
Don't overread it
The absolute difference is small and the trials span decades of different practice; it does not show why mortality might be higher.
The statistics, in plain English
A Bayesian analysis gives the probability that the effect is harmful rather than a P value. OR 1.10 (95% credible interval 0.98–1.24) with 94.4% probability of harm means the data lean clearly towards harm but do not exclude no effect. The absolute difference is about 0.4 percentage points — roughly one extra death per 250 patients, if real.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for infectious diseases, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free