Paroxetine Association with Suicide Attempts Overstated
Jack Modell, GlaxoSmithKline
3 October 2005
In reference to the article by Aursnes et al., we wish to make several points that call into question the methodology employed and the conclusions drawn about the association between paroxetine and suicide attempts and the overall benefit-risk profile.
First, the Aursnes analysis was based solely on early data submitted to regulatory authorities in 1989. The studies analyzed included only 916 paroxetine-treated patients. Today we have data from more than 13,000 adult patients treated with paroxetine in well-controlled clinical trials. The European Regulatory Authority (CHMP) recently reviewed the much more extensive and complete paroxetine safety and efficacy data currently available (including the cases in the Aursnes analysis) and concluded that the benefit-risk ratio remains favorable for all approved indications in adults.
Second, as the authors suggest, risk may be closely related to the time observed in a study. Therefore, we would expect the number of suicide attempts to be proportional to exposure (191 years of exposure to paroxetine versus 73 to placebo in this dataset). Out of a total of eight suicide attempts, we would have expected 6 to come from the paroxetine group. As evidenced by the data cited, there was not a significant difference between actual and expected number of events based on exposure (Fisher's Exact Test one-sided p = 0.30). It is unclear how the authors came to the conclusion that “the data strongly suggest that use of SSRIs are connected with increased intensity per year of suicidal attempts.”
Third, we disagree with the absolute number of suicide attempts referenced by the authors. One case (“037” from ref. 93 in Table 1 of the paper) is clearly described in the original study report as “not considered a true suicide attempt by the investigator.” A second case (“04 02 056” from ref 84 in Table 1), in fact, did not occur in the study quoted but instead in a long-term extension trial of crossover design, and hence may have easily been confounded. Suicide attempts, in general, are an inaccurate measure of suicidality as they include events more properly deemed “self harm.”
Finally, the Bayesian analysis was flawed in that it compared the authors’ hypothesis to one in which paroxetine reduces the chances of a suicide attempt. Rather, the hypothesis should have been compared to one in which the rate is identical for paroxetine and placebo groups. It can be shown that this flaw, combined with a preponderance of data in the paroxetine group (nearly three times the exposure), led to bias in the posterior probability. Also, the use of an informative prior is equivalent to stacking the deck before analyzing the data. The bias introduced by these two separate facets of the analysis led to a posterior probability that does not fairly represent the data.
Based on clear weaknesses in the Aursnes analysis of suicide attempts, we believe that the authors’ conclusions regarding the appropriate use of paroxetine in adults lack support. The overall benefit-risk profile of paroxetine, informed by thousands more patient cases than reviewed by Aursnes, remains favorable. This is especially true in light of the significant morbidity and potential mortality accompanying untreated depression.
Respectfully Submitted,
James Roger, PhD
Director, Research Statistics Unit
GlaxoSmithKline
Alastair Benbow, MB, BS, MRCP, FFPM
Vice President and European Medical Director
GlaxoSmithKline
Frank Rockhold, PhD
Senior Vice President, Biomedical Data Sciences
GlaxoSmithKline
Jack Modell, MD
Vice President, Clinical Psychiatry North America GlaxoSmithKline
Padraig Wright, MB, MRCP, MD
Vice President, Clinical Psychiatry Europe
GlaxoSmithKline
Competing interests
GlaxoSmithKline is the manufacturer of Seroxat, Paxil, and Paxil CR (paroxetine HCl)
Paroxetine Association with Suicide Attempts Overstated
3 October 2005
In reference to the article by Aursnes et al., we wish to make several points that call into question the methodology employed and the conclusions drawn about the association between paroxetine and suicide attempts and the overall benefit-risk profile.
First, the Aursnes analysis was based solely on early data submitted to regulatory authorities in 1989. The studies analyzed included only 916 paroxetine-treated patients. Today we have data from more than 13,000 adult patients treated with paroxetine in well-controlled clinical trials. The European Regulatory Authority (CHMP) recently reviewed the much more extensive and complete paroxetine safety and efficacy data currently available (including the cases in the Aursnes analysis) and concluded that the benefit-risk ratio remains favorable for all approved indications in adults.
Second, as the authors suggest, risk may be closely related to the time observed in a study. Therefore, we would expect the number of suicide attempts to be proportional to exposure (191 years of exposure to paroxetine versus 73 to placebo in this dataset). Out of a total of eight suicide attempts, we would have expected 6 to come from the paroxetine group. As evidenced by the data cited, there was not a significant difference between actual and expected number of events based on exposure (Fisher's Exact Test one-sided p = 0.30). It is unclear how the authors came to the conclusion that “the data strongly suggest that use of SSRIs are connected with increased intensity per year of suicidal attempts.”
Third, we disagree with the absolute number of suicide attempts referenced by the authors. One case (“037” from ref. 93 in Table 1 of the paper) is clearly described in the original study report as “not considered a true suicide attempt by the investigator.” A second case (“04 02 056” from ref 84 in Table 1), in fact, did not occur in the study quoted but instead in a long-term extension trial of crossover design, and hence may have easily been confounded. Suicide attempts, in general, are an inaccurate measure of suicidality as they include events more properly deemed “self harm.”
Finally, the Bayesian analysis was flawed in that it compared the authors’ hypothesis to one in which paroxetine reduces the chances of a suicide attempt. Rather, the hypothesis should have been compared to one in which the rate is identical for paroxetine and placebo groups. It can be shown that this flaw, combined with a preponderance of data in the paroxetine group (nearly three times the exposure), led to bias in the posterior probability. Also, the use of an informative prior is equivalent to stacking the deck before analyzing the data. The bias introduced by these two separate facets of the analysis led to a posterior probability that does not fairly represent the data.
Based on clear weaknesses in the Aursnes analysis of suicide attempts, we believe that the authors’ conclusions regarding the appropriate use of paroxetine in adults lack support. The overall benefit-risk profile of paroxetine, informed by thousands more patient cases than reviewed by Aursnes, remains favorable. This is especially true in light of the significant morbidity and potential mortality accompanying untreated depression.
Respectfully Submitted,
James Roger, PhD
Director, Research Statistics Unit
GlaxoSmithKline
Alastair Benbow, MB, BS, MRCP, FFPM
Vice President and European Medical Director
GlaxoSmithKline
Frank Rockhold, PhD
Senior Vice President, Biomedical Data Sciences
GlaxoSmithKline
Jack Modell, MD
Vice President, Clinical Psychiatry North America GlaxoSmithKline
Padraig Wright, MB, MRCP, MD
Vice President, Clinical Psychiatry Europe
GlaxoSmithKline
Competing interests
GlaxoSmithKline is the manufacturer of Seroxat, Paxil, and Paxil CR (paroxetine HCl)