Skip navigation

Anticonvulsant NNTs

Over the last few years several new antiepileptic (or anticonvulsant) drugs have been registered. A systematic review [1] has collected information which allowed us to calculate NNTs and NNHs in one particular clinical circumstance.


A comprehensive search was used, including MEDLINE, hand searching, and use of an established database. Trials were included if they:
  • recruited patients only with partial epilepsy
  • they were randomised trials of add-on therapy of antiepileptics
  • the treatment period was at least 8 weeks
  • seizures were reported as an outcome


The chosen efficacy outcome was the number of patients with at least 50% reduction in seizure frequency compared with baseline. Five adverse effects - ataxia, dizziness, fatigue, nausea and somnolence - were considered common and important, and information was extracted if available.


NNTs for the drugs were calculated by Bandolier for any dose of drug where there were either three randomised trials or at least 100 treated patients. The NNTs are shown in Table 1 and the Figure as an NNT league table.

Table 1: NNTs for new antiepileptic drugs in add on treatment for refractory epilepsy

Drug Dose (mg) Total number of patients NNT (95%CI)
Gabapentin 900 256 NSD
Gabapentin 1200 514 8.8 (5.7 to 20)
Zonisamide 500 291 7.4 (4.5 to 20)
Lamotrigine 500 454 7.2 (5.0 to 13)
Tiagabine 32 651 6.5 (4.9 to 9.6)
Vigabatrin 300 384 3.3 (2.6 to 4.6)
Topiramate 600 246 3.0 (2.3 to 4.5)
Topiramate 1000 303 2.9 (2.2 to 3.9)
Outcome is patients with at least 50% reduction in seizure frequency. NSD indicates no significant difference from placebo.
Two of the drugs, topiramate and vigabatrin are clearly better (have lower NNTs) than tiagabine, lamotrigine, zonisamide and gabapentin (though the doses for which data were available may not be the doses commonly used). NNHs (Table 2) for the five adverse effects and treatment-related study withdrawal show no profound differences between the drugs, though with the eye of faith there is a tendency towards more effective drugs having lower (worse) NNHs.

Table 2: NNHs for new antiepileptic drugs in add on treatment for refractory epilepsy

  Gabapentin Zonisamide Lamotrigine Tiagabine Vigabatrin Topiramate
Ataxia 12 5 7 NSD NSD 12
Dizziness 9 8 5 8 14 6
Fatigue 14 8 NSD 18 10 8
Nausea NSD NSD 12 NSD NSD 20
Somnolence 9 ND 22 NSD 15 5
Adverse effect withdrawal NSD 11 25 13 12 9
NSD indicates no significant difference. ND indicates no data

Comment and déja vù

The authors make the point, quite rightly, that the trials they have chosen, and therefore the results, may be highly specific to the use of the antiepileptic drugs used as add-on therapy in adults with partial epilepsy. They may not be the same in people with generalised epilepsies, or in childhood epilepsies.

The authors themselves hesitate to draw differences between the effectiveness of the drugs. This may because they use odds ratios as their outcome. Odds ratios probably give the wrong answers in situations like this where the rate of events is high [2], but even if relative risks are used there is overlap of confidence intervals. The large confidence intervals for NNTs reflects the relatively small number of patients in the analyses. But most people can draw a conclusion from a simple picture like Figure 2, which is unlikely to be very wrong even if more trials are done.

Finally, some readers will have a sense of déja vù if they read the Epilepsia paper [1]. A similar review was published a year earlier by the same authors [3]. The new one contains one more trial, and has the adverse effect analysis, but, curiously, does not reference the previous paper.


  1. AG Marson, ZA Kadir, JL Hutton, DW Chadwick. The new antiepileptic drugs: a systematic review of their efficacy and tolerability. Epilepsia 1997 38: 859-80.
  2. DL Sackett, JJ Deeks, DG Altman. Down with odds ratios! Evidence-Based Medicine 1996 Sept/Oct 1: 164-6.
  3. AG Marson, ZA Kadir, DW Chadwick. New antiepileptic drugs: a systematic review of their efficacy and tolerability. BMJ 1996 313: 1169-74
previous or next story in this issue