Twenty-two studies with a total of 5,647 participants were included. Only 3 studies were blinded randomised controlled trials.
The aggregate suicide rate, based on a weighted random-effects model, was 0.159%/year (95% CI: 0.133, 0.204) during lithium treatment and 0.875%/year (95% CI: 0.630, 1.119) off-lithium. This indicated an overall 5.50-fold lower crude suicide risk associated with lithium treatment.
Twelve studies, which included patients with and without lithium treatment and non-zero suicide rates without treatment, yielded overall suicide rates of 1.02%/year (95% CI: 0.80, 1.29) without treatment versus 0.15%/year (95% CI: 0.10, 0.23) with lithium treatment, a significant crude difference of nearly 7-fold (P<0.001). However, there was significant heterogeneity (Q=23.4, P=0.016). Therefore, the random-effects model was used to estimate the log risk ratio for the on- versus off-lithium contrast. The pooled log rate ratio was 2.18 (95% CI: 1.42, 2.95), indicating a risk ratio of 8.85 (95% CI: 4.14, 19.1), which was highly significantly different from the null risk ratio of 1.0 (P<0.0001). The removal of any single study did not significantly affect the pooled risk ratio estimate.
Bivariate analyses of suicide rates found no significant relationships between suicide rates, either with or without lithium treatment, and the number of participants in the study, study quality score, blind study design, discontinuation of lithium, diagnoses, exposure time or publication year. These conclusions were sustained in multivariate logistic regression analyses.
The off-lithium suicide rates did not differ significantly between the 8 studies that involved treatment discontinuation (1.009%/year, 95% CI: 0.754, 1.323) and the 5 studies that did not (1.052%/year, 95% CI: 0.660, 1.593) (P=0.91). This comparison, and the higher suicide rate in studies that involved observations only in lithium-treated patients, suggested that it was unlikely that lithium-treated patients had a lower suicide risk independent of treatment, or that treatment discontinuation accounted for the higher suicide risk without lithium.
The quality ratings of the studies averaged 46.8% (range: 14.3 to 100). Publication bias was assessed using a funnel plot and was non significant (t=2.08, P=0.082).