Seventy-four studies (77 articles) were included (3,321 participants). These included 15 prospective studies (955 participants, range 15 to 196), including three blinded RCTs (327 participants) and two unblinded RCTs (89 participants) and 10 observational studies (539 participants). Fifty-nine studies were retrospective (2,366 participants).
In the three blinded RCTS, at three months the proportion of participants in the intervention groups with a 50% or more reduction in seizure frequency from baseline was 23%, 31% and 57%. Two of these RCTs reported a greater reduction in seizure frequency in the high stimulation than in the sham stimulation group (25% versus 6% and 28% versus 15%).
In the two non-blinded RCTs across all stimulation paradigms 29% to 45% of participants had at least 50% fewer seizures than at baseline; mean seizure reduction ranged from 26% to 30%.
In the 10 prospective observational studies 21% to 50% of participants had at least 50% fewer seizures than at baseline and mean seizure reduction ranged from 17% to 55%.
When all 74 studies were pooled, the odds of achieving (versus not achieving) at least 50% fewer seizures significantly favoured vagus nerve stimulation (OR 1.83, 95% CI 1.80 to 1.86). Data on seizure frequency were available for 1,789 participants among whom seizures were reduced by a mean of 44.6% (standard error 0.5%). Seizure outcomes by Engel class were available for 2,634 participants among whom 50.6% achieved at least 50% fewer seizures after vagus nerve stimulation; fewer than 5% achieved complete seizure freedom and 25.4% had no benefit.
In subgroup analyses, more than one year's follow-up, young age, generalised epilepsy, traumatic epilepsy and tuberous sclerosis were associated with significantly higher seizure reduction. Funnel tests did not suggest significant publication bias. Only three studies reported adverse events (hoarseness was the most common with 37% to 62%).