The literature search identified a total of 3717 citations. After initial review by at least two authors (SD, BH, DF), 999 trials were deemed potentially eligible. Of these, we excluded 375 for the following reasons: duplicate publication (n = 125), not a randomised controlled trial (n = 118), SSRI control only (n = 62), no SSRI arm (n = 13), subgroup analysis (n = 20), foreign language other than French or English (n = 13), short trial duration (n = 12), incomplete or inaccessible data (n = 10), and population of suicidal patients (n = 2). We identified an additional 78 trials meeting eligibility criteria by the electronic search of the Cochrane Collaboration register of controlled trials (Cochrane depression, anxiety, and neurosis group) and the manual review of the bibliographies of three published systematic reviews and of all eligible trials ().13-15
The 702 trials comprised 411 comparisons between SSRIs and placebo, 220 comparisons between SSRIs and tricyclic antidepressants, and 159 comparisons between SSRIs and active therapies other than placebos or tricyclic antidepressants. As some trials had more than one comparison arm, the total number of comparisons exceeds the number of published trials. Of the 159 comparisons, the most common comparative treatments were moclobemide (21 trials), psychotherapy (20), maprotiline (18), and mianserin (16).
Identification and inclusion of trials
Four hundred and thirty seven trials (62.3%) randomised less than 100 patients. Forty nine trials (7.0%) went on for longer than six months. Of the 493 trials that reported dropout information, 227 (46.0%) had a dropout rate of more than 25%. One hundred and four trials (14.8%) reported adverse events beyond thresholds of 3%, 5%, or 10% of enrolled patients or occurring in more than a defined number of patients, without any detail of adverse events related to suicidality.
Of the 702 trials including 87 650 patients, 414 (59.0%) were conducted in patients with a diagnosis other than major depressive conditions. Sixty eight per cent of trials (n = 475) included more than 50% women, and 91% (n = 638) of trials were conducted in participants with an average age of less than 60 years.
A total of 345 trials representing 36 445 patients reported the number of suicide attempts (143 in total) and were included in the analysis. Of the 345 trials reporting suicide attempts as adverse events, 64 reported at least one suicide attempt. In comparing trial characteristics between trials that reported suicide attempts and those that did not, the only significant difference was that larger trials tended not to report (χ2 test, df = 2, P = 0.001). The overall rate of suicide attempts was 3.9 (95% confidence interval 3.3 to 4.6) per 1000 patients treated in clinical trials. When we used study duration as exposure time, we found an incidence of 18.2 suicide attempts per 1000 patient years. For the trials conducted in patients with depression, the overall rate of suicide attempts was 4.9 (95% confidence interval 4.2 to 5.6 per 1000 patients). provides the reported numbers of fatal and non-fatal suicide attempts.
Fatal and non-fatal suicide attempts in the analysed trials
We found a significant increase in the odds of suicide attempts (odds ratio 2.28, 1.14 to 4.55, number needed to treat to harm 684; P = 0.02) for patients receiving SSRIs compared with placebo (). Given reduced sample sizes, our ability to detect significant differences within subgroups was limited. However, all odds ratios exceeded 1.0 except for trials whose participants had a mean age of over 60 (). In comparing non-fatal suicide attempts, a significant difference overall remained (2.70, 1.22 to 5.97; P = 0.01). In comparing fatal suicide attempts, we did not detect any differences between SSRIs and placebo (0.95, 0.24 to 3.78).
Fatal and non-fatal suicide attempts in SSRI trials and placebo trials
In the pooled analysis of SSRIs compared with tricyclic antidepressants, we did not detect differences in the odds of suicide attempts (0.88, 0.54 to 1.42; ). We found no clinically or statistically important differences in any subgroup analyses. The odds ratio of non-fatal suicide attempts was 0.85 (0.51 to 1.43) and the odds ratio of fatal suicide attempts for SSRIs compared with tricyclic antidepressants was 7.27 (1.26 to 42.03).
Fatal and non-fatal suicide attempts in SSRI trials and trials studying tricyclic antidepressants
We found an increase in the odds of suicide attempts when comparing SSRIs with therapeutic interventions other than tricyclic antidepressants (1.94, 1.06 to 3.57, number needed to treat to harm 239; ). Again with smaller sample sizes, we found no subgroup specific differences that reached significance. All odds ratios exceeded 1.0, except for trials in which the proportion of women exceeded 75%. The odds ratio for fatal suicide attempts was 0.59 (0.16 to 2.24) and that for non-fatal suicide attempts 2.25 (1.16 to 4.35).
Fatal and non-fatal suicide attempts in SSRI trials and trials studying other therapies