Figure

Further information

Table






Further information

By fitting the model in table 2[t2] using a Bayesian Conditional Independence Model in WinBUGS, we can incorporate uncertainty in the various components to give 95% credibility intervals for the excess suicides. Allowing for imprecision in the proportion of males in those on antidepressants (0.33, 95% credible interval (0.30 to 0.36)¹), and in the proportion of suicides that are male (0.75 95% credible interval (0.72 to 0.78)), assuming Poisson distributions on the numbers of male and female suicides, and using the pooled relative risk of antidepressants from the paediatric trials (table 1[t1]), the excess male suicides are estimated at 233 (95% credible interval - 121 to 427), and for females at 155 (95% credible interval - 81 to 287). The excess total suicides are 388 (95% credible interval -202 to 704). These estimates are very dependent on the assumed value of the relative risk. Table A gives the estimates at various different assumed values for the relative risk – the credible intervals for the estimated number of suicides from these models are narrower than for those given above as we are no longer modelling uncertainty in the value for the relative risk. If the true relative risk is 1.5 we estimate this may have resulted in 327 excess suicides in 2002 compared to 1991 (30 per year between 1991 and 2002).

From only four drugs, it is hard to assess heterogeneity: the Bayesian analysis gives an I² of 17% (95% credible interval 2% to 84%), which is compatible with negligible heterogeneity or substantial heterogeneity. Furthermore, paediatric clinical trial data for venlafaxine (a non-SSRI) indicate that there is a large increased risk of "suicide related events" in depressed children randomised to this drug (OR 13.8 (95% CI 1.8 to 103.6), with I² of 35% (95% credible interval 2% to 84%)).²⁶ If these data had been included in our pooled estimate of risk, the pooled odds ratio would have been 2.29 (95% credible interval 1.02 to 6.24). Using this estimate our model predicts there may have been 550 excess suicides (50 per year). All the above estimates of excess suicides could conceivably be masked within underlying trends in population suicide rates. As the relative risks in relation to each drug (Table 1[t1]) were in fact derived from up to three trials (paroxetine) for each SSRI our credible intervals will slightly under-estimate levels of uncertainty.

Our estimate of excess suicides is also dependent on the assumed rate of suicide in people treated with antidepressants in primary care in England. The rate we used (85 per 100,000¹⁶) is lower than reported than in Khan’s analyses of antidepressant clinical trial data (rate 590 per 100,000 per year for people taking SSRIs; 450 per 100,000 for those taking placebo and 760 per 100,000 for TCA users.¹⁵ These estimates are over five times higher than Jick et al’s¹⁶ and if used in our model would result in estimates of 175 (35 ´ 5) excess suicides per year between 1991 and 2002, or 1,925 additional suicides in 2002 (>40% of suicides) compared to 1991. We believe this level of increase is highly unlikely, in a recent study of 135 suicides amongst people aged <35 years only 28% were taking antidepressants at the time of death.

1. Spiegelhalter DJ, Thomas A, Best NG. WinBUGS version 1.2 User manual. Cambridge: MRC Biostatistics Unit, 1999.
2. Kelly S, Bunting J. Trends in suicide in England and Wales, 1982-96. Population Trends 1998;92:29-41
3. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003;327:557-560.
4. ApplebyL, Cooper J, Amos T, Faragher B. Psychological autopsy study of suicides by people aged under 35. Br J Psych 1999;175:168-174.

Posted as supplied by author

Table A Effects of varying estimates of the relative risk associated with antidepressant prescribing on estimated number of excess suicides in 2002 compared to 1991

Assumed relative risk	Excess male suicides in 2002 compared to 1991 (95% CI)	Excess female suicides in 2002 compared to 1991 (95% CI)	Total excess suicides
1.5 (assumed known)	196 (172 to 221)	131 (111 to 152)	327
2.0 (assumed known)	294 (258 to 331)	196 (167 to 227)	490
2.3 (estimated with uncertainty from meta-analysis with venlafaxine)	330 (14 to 502)	220 (9 to 339)	550