Anticonvulsant drugs for management of pain: a systematic reviewBMJ 1995; 311 doi: http://dx.doi.org/10.1136/bmj.311.7012.1047 (Published 21 October 1995) Cite this as: BMJ 1995;311:1047
- Henry McQuay, clinical reader in pain reliefa,
- Dawn Carroll, senior research nursea,
- Alejandro R Jadad, research fellowa,
- Philip Wiffen, principal pharmacistb,
- Andrew Moore, consultant biochemistb
- aOxford Pain Relief Unit, Churchill Hospital, Oxford OX3 7LJ
- b Pharmacy Department, Churchill Hospital
- Correspondence to: Dr McQuay.
- Accepted 26 June 1995
Objective: To determine effectiveness and adverse effects of anticonvulsant drugs in management of pain.
Design: Systematic review of randomised controlled trials of anticonvulsants for acute, chronic, or cancer pain identified by using Medline, by hand searching, by searching reference lists, and by contacting investigators.
Subjects: Between 1966 and February 1994, 37 reports were found; 20 reports, of four anticonvulsants, were eligible.
Main outcome measures: Numbers needed to treat were calculated for effectiveness, adverse effects, and drug related withdrawal from study.
Results: The only placebo controlled study in acute pain found no analgesic effect of sodium valproate. For treating trigeminal neuralgia, carbamazepine had a combined number needed to treat of 2.6 for effectiveness, 3.4 for adverse effects, and 24 for severe effects (withdrawal from study). For treating diabetic neuropathy, anticonvulsants had a combined number needed to treat of 2.5 for effectiveness, 3.1 for adverse effects, and 20 for severe effects. For migraine prophylaxis, anticonvulsants had a combined number needed to treat of 1.6 for effectiveness, 2.4 for adverse effects, and 39 for severe effects. Phenytoin had no effect on the irritable bowel syndrome, and carbamazepine had little effect on pain after stroke. Clonazepam was effective in one study for temporomandibular joint dysfunction. No study compared one anticonvulsant with another.
Conclusions: Anticonvulsants were effective for trigeminal neuralgia and diabetic neuropathy and for migraine prophylaxis. Minor adverse effects occurred as often as benefit.
To evaluate the effectiveness of these drugs, we conducted a systematic review of all randomised controlled trials reported between 1966 and February 1994
Only 20 trials were eligible for inclusion, and three pain syndromes were subject to more than one trial: trigeminal neuralgia, diabetic neuropathy, and migraine prophylaxis.
Overall number needed to treat to produce benefit (improved pain scores) was about 2.5, for minor adverse effects it was about 3, and for severe adverse effects it was 20-30
Anticonvulsants do have an analgesic effect, although at similar risk of minor adverse effects
Anticonvulsant drugs have been used in pain management since the 1960s, soon after they were first used to revolutionise the management of epilepsy. The clinical impression is that they are useful for neuropathic pain, especially when the pain is lancinating or burning.1 Although these disorders are not common (the incidence of trigeminal neuralgia is 4/100000 a year2), they can be very disabling. Carbamazepine is one of few effective interventions for trigeminal neuralgia and is usually the drug of choice.3 In Britain carbamazepine is licensed for paroxysmal pain of trigeminal neuralgia (up to 1600 mg daily). Phenytoin is also licensed for trigeminal neuralgia if carbamazepine is ineffective or if a patient cannot tolerate effective doses. When anticonvulsants are used as adjuvant drugs in other pain syndromes valproate is often preferred to carbamazepine because it may be better tolerated.4 Anticonvulsants are also prescribed in combination with antidepressants, as in the treatment of post-herpetic neuralgia.5 In Britain no anticonvulsant is licensed for treating any pain other than trigeminal neuralgia.
Serious side effects have been reported with anticonvulsant drugs, including deaths from haematological reactions.6 The commonest adverse effects are impaired mental and motor function, which may limit clinical use, particularly in elderly people.6 7 8
The purpose of this review was to evaluate the effectiveness of anticonvulsant drugs as analgesics in order to provide evidence based recommendations for clinical practice and to identify an agenda for clinical research. We used the “number needed to treat” method9 to produce clinically interpretable measures of benefit, minor harm, and major harm.
SELECTION OF REPORTS
We included reports if they were randomised controlled trials of the analgesic effects of anticonvulsant drugs. We excluded studies if they were not randomised; were studies of experimental pain, case reports, or clinical observations; or were studies of anticonvulsants used to treat pain produced by other drugs.
Reports were identified by several methods. We conducted a Medline search (SilverPlatter 3.0, 3.1, and 3.11) from 1966 to February 1994 with a search strategy designed to identify the maximum number of randomised or double blind reports by means of a combination of text words, “wild cards,” and MeSH terms.10 This search strategy was narrowed to include specific anticonvulsant drugs. We hand searched 40 medical journals, chosen from the 50 with the highest number of reports in Medline, and nine specialist journals that were either not on that list or not indexed.11 Our search covered volumes published between 1950 and 1990. We identified additional reports from the reference list of the retrieved papers. A letter was sent to the first author for further information on their published report (method of randomisation, double blinding, outcome measures, and dropouts) and to ask if they knew of any other studies which met our inclusion criteria, done either by them or by other investigators.
Eligibility was determined by reading each report identified by the search. We scored the reports independently for quality using a three item scale11 and then met to agree a consensus score for each report. Studies that were described as randomised were given one point, and a further point was given if the method of randomisation was described and was appropriate (such as use of random number tables). If randomisation was inappropriate (such as alternate allocation) one point was deducted and the report was excluded. Studies that were described as double blind were given one point. A further point was given if blinding was described and was appropriate (such as matched placebos), and one point was deducted if blinding was inappropriate. Reports that described the number and reasons for withdrawals were given one point. The maximum score was 5 and the minimum for an included report was 1.
From each report we gathered information about the pain condition and the number of patients studied; the anticonvulsant drug used and the treatment regimen; study design (placebo control or control with active treatment); duration of the study and follow up; outcome measures and results; and minor and major (drug related withdrawal from study) adverse effects. Reports often used several different measures of pain. The prior definition of a clinically relevant outcome was greater than 50% pain relief. We extracted information about improvement in binary form for analysis, and we used a hierarchy of measures: the number of patients free of pain at the end of the study; complete, excellent, or very good response; or the number of patients who improved. No weighting was used between these different indices.
Odds ratios and 95% confidence intervals were calculated with a fixed effects model. We calculated numbers needed to treat and 95% confidence intervals9 for effectiveness, for adverse effects and for drug related withdrawal from study, both for the individual reports and for combined single treatment or control arms.
We identified 37 reports (all published), 34 from Medline and three from reference lists. Of these, 17 were excluded and one report12 was a duplicate publication (table I). The remaining 20 randomised controlled trials were eligible for our study. Four anticonvulsant drugs were used: carbamazepine in 10 of the trials, phenytoin in five, clonazepam in three, and sodium valproate in two. The pain conditions investigated were chronic non-malignant pain (in 17 trials), cancer pain (one trial), postoperative pain (one trial), and acute herpes zoster (one trial). Tables II and III give details of the eligible placebo controlled trials and active treatment controlled trials respectively. The median quality score for the placebo controlled studies was 3 (range 2-5) and for the active treatment controlled studies was 2 (range 1-4).
We requested data from 19 authors. Five replied, but only one (Leijon) was able to supply information relevant to this review.
The only placebo controlled study of treating acute pain compared sodium valproate 15 mg/kg with the non-steroidal anti-inflammatory drug ketoprofen (2 mg/kg) and placebo, all given intravenously over 20 minutes.38 Ketoprofen produced a significant fall in pain intensity compared with placebo, but valproate did not (table II). In a comparison of carbamazepine and prednisolone for managing acute herpes zoster, the 20 patients given prednisolone reported less pain and faster skin healing (3.7 weeks v 5.3 weeks) than the 20 given carbamazepine 400 mg/day, and 13 of the patients given carbamazepine still had pain after two months compared with three of those given prednisolone (table III).42
CHRONIC PAIN Trigeminal neuralgia
Of the 12 placebo controlled studies of treating chronic pain, three were for trigeminal neuralgia, all with carbamazepine (table II).22 23 24 25 26 27 28 29 30 31 In a crossover trial 19 out of 27 patients had a complete or very good response after five days' treatment with carbamazepine (titrated to a maximum dose of 1 g/day), compared with none after placebo.30 Again with a crossover design and dose titration (to a maximum dose of 2.4 g/day), 15 of the 20 patients given initial carbamazepine had a good or excellent response after 14 days' treatment compared with six of the 24 patients given initial placebo.31 The extent to which the pain was relieved may be gauged from the third study.29 With doses ranging from 400 mg/day to 800 mg/day given for periods of two weeks, the mean fall in maximum pain intensity was 58% with carbamazepine compared with 26% with placebo. The effectiveness odds ratios of two of the three studies, and the combined ratio, showed carbamazepine to be more effective than placebo. The number needed to treat for effectiveness compared with placebo was 2.6, that for minor adverse effects was 3.4, and that for drug related withdrawal from the study was 24 (table IV).
Three active treatment controlled studies compared carbamazepine with tizanidine ((alpha)2 adrenergic agonist),43 tocainide (antiarrhthymic drug),44 and pimozide (antipsychotic drug)45 (table III). Carbamazepine produced better results than tizanidine; there was no significant difference in the tocainide study; and pimozide produced better results than carbamazepine.
Two placebo controlled studies of treating diabetic neuropathy (one with carbamazepine32 and one with phenytoin36) found that treatment with anticonvulsant resulted in 30-50% more patients improving after two weeks than did placebo. A third study comparing treatment with phenytoin for 23 weeks with placebo found no difference in mean pain intensity.37 For the two studies with dichotomous data, the combined effectiveness odds ratio showed a significant effect for anticonvulsant compared with placebo. The number needed to treat for effectiveness compared with placebo was 2.5, that for adverse effects was 3.1, and that for drug related withdrawal from the study was 20 (table IV).
There were no eligible active treatment control studies of diabetic neuropathy.
Of three placebo controlled studies of treating migraine prophylaxis, with three different anticonvulsants, two showed greater effect with the anticonvulsant than with placebo (table II). Six weeks of treatment with carbamazepine 3 tablets/day led to improvement in 38 out of 45 patients compared with 13 of the 48 given placebo.33 Sodium valproate 800 mg/day for eight weeks produced significant reduction in the number of migraines, in their duration, and in the pain intensity; it was effective in 25 out of 29 patients.39 The third study was of 1 or 2 mg of clonazepam daily for 60 days and found no significant difference between clonazepam and placebo.40 For the two studies with dichotomous data, the combined effectiveness odds ratio showed a significant effect for anticonvulsant compared with placebo; the number needed to treat for effectiveness compared with placebo was 1.6, that for adverse effects was 2.4, and that for drug related withdrawal from the study was 39 (table IV).
There were no eligible active treatment control studies of migraine prophylaxis.
Other pain syndromes
Placebo controlled studies (table II)—Phenytoin 300 mg/day for six weeks had no effect in the one study of the irritable bowel syndrome.35 Four weeks of carbamazepine treatment at a final dose of 800 mg/day was judged to have improved central pain after stroke in five out of 14 patients, compared with 10 out of 15 patients given amitriptyline 75 mg and one out of 15 patients given placebo.34 In a 60 day study of clonazepam (mean daily dose 0.375 mg) for temporomandibular joint dysfunction, analysis at 30 days showed significantly lower pain intensity scores with the anticonvulsant compared with placebo.41
Active treatment controlled studies (table III)—A 24 week comparison of phenytoin and intramuscular gold showed that gold gave significantly better relief of pain and morning stiffness from rheumatoid arthritis.46 Phenytoin 200 mg/day was compared with buprenorphine alone and a combination of buprenorphine and phenytoin (100 mg/day) for treating cancer pain; all three regimens produced good or moderate relief in more than 60% of patients.47 In a comparison of a combination of carbamazepine and clomipramine with transcutaneous electrical nerve stimulation for treating post-herpetic neuralgia, drug treatment produced improvement in nine out of 16 patients while nerve stimulation produced improvement in three out of 13 patients.48
ADVERSE EFFECTS AND DRUG RELATED WITHDRAWAL FROM STUDY
In the placebo controlled studies there were 16 drug related withdrawals from anticonvulsant treatment compared with two from placebo (table IV). Where adverse effects were reported the incidence was 25-50% in each study. Drowsiness, dizziness, and disturbance in gait were the common problems.
Although randomised controlled trials are the optimum method of assessing health care technologies and interventions,49 buttressed by double blinding when outcome measures are subjective,50 51 52 many interventions are time honoured rather than supported by trials. On whom does the burden of proof then fall?53 The aim of our study was to review the effectiveness and safety of anticonvulsant drugs in the management of pain, restricting the review to reports of randomised controlled trials.
We used the number needed to treat approach because most of the data was in dichotomous form, which lends itself to this analysis, and because the number needed to treat is more readily clinically interpretable than, for example, effect sizes. The number needed to treat was calculated for minor and major adverse effects as well as for effectiveness because adverse effects are important for clinical decision making. This approach may be useful for reviews of other long established interventions. The older the report, the more likely it was to present simple binary data, such as improved versus not improved. More recent reports which restricted data presentation to mean data for treatment and control were not accessible to the method.
EFFECTIVENESS OF ANTICONVULSANT DRUGS
Anticonvulsants were ineffective in the one report of postoperative pain38 and in the one of acute herpes zoster.42 There is no logic in using anticonvulsants to manage acute nociceptive pain when there are other (effective) remedies.
The overall pattern of a number needed to treat for effectiveness of about 2.5, for adverse effects of about 3, and for drug related withdrawal from the study of 20-40 was surprisingly similar for the three pain syndromes that were subject to more than one trial (table IV).
Trigeminal neuralgia—Medical students are often taught that a positive response to carbamazepine is diagnostic for trigeminal neuralgia. However, if only one out of every two patients responds to the treatment, this statement needs to be qualified. One caveat is that the study populations may have included patients who had other interventions, such as nerve blocks, and the number needed to treat for effectiveness may be more impressive when trigeminal neuralgia is treated with carbamazepine in the initial stages. The statement that “approximately 70% of patients will have significant pain relief”3 would seem to be about right.
Diabetic neuropathy is perceived as a model for other neuropathic pain syndromes, and results from diabetic neuropathy are often extrapolated to other syndromes. The studies reviewed here gave conflicting results, with a negative result in the longest study (46 weeks) balanced by two studies with positive results. The number needed to treat for effectiveness was the same as that for adverse effects. The usual clinical decision is between antidepressants and anticonvulsants as first line treatment. A direct comparison of antidepressant and anticonvulsant activities is available from the study of pain after stroke, in which the number needed to treat for effectiveness was 1.7 for amitriptyline and was 3.4 for carbamazepine, with the same number needed to treat for adverse effects and study withdrawal.34
Migraine prophylaxis—The three placebo controlled randomised controlled trials showed anticonvulsants to be effective, but recent advances in migraine management may reduce the impact of these results.
This review shows that there is a need for high quality studies of the relative effectiveness of different anticonvulsants in treating chronic pain syndromes and for comparisons of antidepressants with anticonvulsants. The usefulness of such primary studies would be greatly increased by improvements in the quality of reporting. Investigators presenting data as means for treatment and control should also consider the (simple) presentation of binary data, for example the number of patients with more than 50% pain relief.
We thank Clare Abbott and Jo Riordan for helping to locate the reports. We are also very grateful to the authors of the original reports who took the trouble to reply to our correspondence and those who provided us with original data: S Vilming, F Lechin, G Leijon, D Greenbaum, and S Yajnik.
Funding Supported by a grant from Oxford Regional Health Authority and Pain Research Funds.
Conflict of interest None.