Eighteen RCTs (n=1,427) were included in the meta-analysis, five of which had multiple treatment groups. The sample size varied from 14 to 294. The median of follow-up rates of patients receiving antidepressants was 71 per cent and for patients receiving placebos was 78 per cent.
Five studies were judged as high quality using the Jadad score. Seven studies were judged as high quality using the van Tulder score. Four studies were judged as high quality using both measures.
When the studies were pooled, antidepressants were significantly associated with a reduction of pain compared with control group (SMD -0.43, 95% confidence interval, CI: -0.55, -0.30, p<0.001; 22 treatment arms), a reduction of fatigue (SMD -0.13, 95%CI: -0.26, -0.01, p=0.04; 14 treatment arms), a reduction of depressed mood (SMD -0.26, 95% CI: -0.39, -0.12, p<0.001; 10 treatment arms), and an improvement of sleep (SMD -0.32, 95%CI -0.46, -0.18, p<0.001; 13 treatment arms) and HRQOL (SMD -0.31, 95%CI -0.42, -0.20, p<0.001; 12 treatment arms). Based on the evaluation of the effect size, the effect was clinically negligible for fatigue and small for the other outcomes.
For each antidepressant class, when the studies were pooled a large effect in pain reduction was significantly associated with TCAs (SMD, -1.64, 95%CI: -2.57, -0.71, p<0.001; six trials), a medium effect for MAOIs (SMD -0.54, 95% CI: -1.02, -0.07, p=0.03; three trials), and a small effect for SSRIs (SMD -0.39, 95%CI: -0.77, -0.01, p=0.04; six trials) and SNRIs (SMD -0.36, 95%CI: -0.46, -0.25, p<0.001; three trials).
Statistically significant heterogeneity was only observed in the outcome of pain (P=0.01, I2 =44.3%). Sensitivity analyses of drug classes did not materially affect the results.
No evidence of publication bias was found according to the visual scanning of forest plots and the fail-safe N analysis; the fail-safe N ranged from 168 to 924. The authors did not report the results of publication bias assessed by the funnel plots.