|
Abstract
|
OBJECTIVE: To test the validity of the conclusion of a recent meta‐analysis1 of the possible association between fluoxetine and suicidal acts.
DESIGN: Monte‐Carlo simulations of the number of suicidal acts in clinical trials of the sizes reported in the published meta‐analysis.1 Main outcome measures: Simulated incidence of suicidal acts using binomial distributions with parameters equal to or higher than those reported in earlier papers. A pooled risk difference and the associated 95% confidence interval were calculated for each set of simulated data.
RESULTS: Assuming a true incidence of 0.3% of suicidal acts in patients receiving fluoxetine and 0.2% in patients receiving placebo, none of 30 sets of simulations showed a significant difference as judged by the pooled risk differences and the corresponding 95% confidence intervals. Even when the risk of suicidal acts was increased to 0.6%, only 38 of 100 simulations produced significant differences in the expected direction. Of the other 62 sets, 56 produced a non‐significant positive risk difference and six produced a non‐significant negative risk difference.
CONCLUSION: The Monte‐Carlo simulations show that the meta‐analysis which claimed to demonstrate that the available data do not show that fluoxetine is associated with an increased risk of suicidal acts, was of low power. Much larger studies are needed before one can say categorically that fluoxetine is not associated with a higher incidence of suicidal acts than placebo. Conversely, the data to date (as the recent meta‐analysis rightly claim) also fail to demonstrate that the risk of suicidal acts is higher with fluoxetine than with placebo. Copyright © 1993 John Wiley & Sons, Ltd.
Language: en |