EVIDENCE-BASED ANSWER
Tricyclic antidepressants, anticonvulsants, and capsaicin reduce the pain of diabetic neuropathy; limited data suggests that lidocaine patches may also be efficacious. Both tricyclic antidepressants and anticonvulsants are superior to placebo in relieving painful diabetic neuropathy. Compared with placebo, patients taking tricyclic antidepressants report reduced pain (number needed to treat [NNT] for at least 50% reduction= 3.5) (strength of recommendation [SOR]: A). Similarly, patients taking anticonvulsants report reduced pain (NNT for at least 50% reduction in pain=2.7) (SOR: A).
Limited evidence suggests that selective serotonin reuptake inhibitors (SSRIs) are no more efficacious than placebo (SOR: C). Both anti-depressants and anticonvulsants have a high rate of minor adverse effects (number needed to harm [NNH]=2.7 for both). Tricyclic antidepressants have an NNH of 17 for side effects severe enough that patients withdrew from the study.
Compared with placebo, topical capsaicin also reduces pain (NNT=4) (SOR: A); however, there are no systematically collected data on side effects for capsaicin. A single case series demonstrates that lidocaine patches are efficacious for neuropathic pain, though expensive (SOR: B). Almost no trials comparing different classes of treatments have been performed.
Evidence Summary
A recent well-done meta-analysis1 summarized available randomized placebo-controlled trials of antidepressants (including tricyclics and SSRIs) and anticonvulsants (including phenytoin, carbamazepine, and gabapentin). Almost all trials compare individual agents against placebo, and there have been no head-to-head trials that address functional outcomes, quality of life, patient satisfaction, or cost. Most trials do not describe diagnostic criteria, consider causes of pain other than diabetes or address diabetic control, which is known to predict frequency of neuropathy. Finally, very few trials include typical primary care patients in a primary care setting or control for important confounding variables such as over-the-counter medications or comorbid illnesses.
Within the constraints of this literature, place-bos have a substantial impact, with an aggregate 32% of patients receiving placebo reporting at least 50% reduction in pain. A total of 16 trials have addressed the efficacy of antidepressants for diabetic neuropathy. Compared with placebo, tricyclic antidepressants have an aggregate NNT of 3.5 (95% confidence interval [CI], 2.6–4.7) for patients reporting at least 50% reduction of pain, along with an NNH of 2.7 (95% CI, 2.1–3.9) for minor adverse effects (typically the muscarinic effects of dry mouth, constipation, and blurred vision) and 17 (95% CI, 10–43) for side effects severe enough to cause withdrawal from a trial. Dosages were in the low to middle range of those used to treat depression; there was no significant difference in efficacy between trials less than 3 weeks and those greater than 3 weeks. No evidence supports differences among different tricyclic agents, and limited evidence suggests that SSRIs are no more efficacious than placebo.
A total of 4 randomized placebo-controlled trials (1 each for phenytoin [Dilantin], carbamazepine [Tegretol], gabapentin [Neurontin], and valproate [Depakote]) have extractable data about the efficacy of anticonvulsants for the pain of diabetic neuropathy. As a class, the NNT for patients reporting at least a 50% reduction in pain was 2.7 (95% CI, 2.2–3.8); the NNH for minor adverse effects (typically transient central nervous system effect such as dizziness, somnolence, or disturbance in gait) was 2.7 (95% CI, 2.2–3.4).
These summary estimates do not include the valproate trial,2 which was reported after the meta-analysis was completed; the report did not allow calculation of NNT, but the findings were consistent with these results. Phenytoin dosage was 300 mg/d; carbamazepine dosage was titrated to 200–600 mg/d, gabapentin from 300–3600 mg/d, and valproate 1200 mg/d. Patients taking anticonvulsants did not have a higher rate of withdrawal compared to those taking placebo. Limited evidence suggests no significant differences among anticonvulsants; there is insufficient evidence to determine optimal dosage of any of these agents.