"...An Analysis of Suicidality in SSRI Pediatric Trials
This sequence of events led me, when visiting Glaxo-SmithKline's archives in Philadelphia in the course of In Re Paxil litigation, to look at indices of withdrawal from Paxil/Seroxat trials in children, as a warning letter to doctors in the UK from GSK in June 2003 had indicated that during the withdrawal phase of treatment children were at increased risk of becoming suicidal.
I reviewed protocols 329, 377, 701, 453, 658, 704, on which FDA have based their assessments of this drug, as well as protocols 715 and 716, assessing among other things the narrative summaries on all patients where these were present as well as details of adverse events such as investigator's judgments as to the relatedness of these events to drug intake and the relationship between the time of any dose change and the adverse event.
On the basis of this review, as well as comparable reviews of other data from adult trials, and combined with statements in the published literature, it is possible to make the following points about the data sent to FDA. These points, I believe, justify the approach taken by MHRA towards the data, namely that it is not worth worrying too much about getting the "signal" from these trials fixed as accurately as possible as the methodologies are so poor that retrospective tinkering of the type currently proposed by FDA does not seem warranted.
First, the narrative summaries that FDA propose to send to the Columbia psychiatric group for blind review of suicidal content are not blindly constructed. In these trials, there would appear to be a systematic bias on the part of the clinical investigators to deny the role of active drug in the causation of problems; which may in fact be supplemented by a tendency to blame placebo for problems with suicidality where these occur in association with each other. Against this background, the non-blind construction of these summaries along with the many psychosocial events that narrative summaries in this domain typically include will commonly give enough doubt to enable a re-categorization of suicidal acts to suicidal gestures or other categorizations.
Second, narrative summaries in general are typically only present for the final suicidal episode or adverse event that leads to discontinuation from a study, but in SSRI trials some children have had more than one suicidal episode. Against a background of investigator reluctance to credit suicidality to the active drug, it is highly likely that the majority of unrecorded suicidal episodes/acts will have occurred in the active treatment group rather than on placebo. In other words, the true picture as regards the number of suicidal acts may be considerably worse than is currently represented in the data.
Third, FDA appear to have ignored the issue of any relationship between dose change and suicidal events, even though the letter from Glaxo SmithKline to healthcare professionals in the UK now states clearly that dose lowering of Paxil/Seroxat may lead to suicidality.
Fourth, it seems clear from conversations with Drs. Temple and Katz that there is data FDA have not seen Glaxo SmithKline's protocols 511 and 716, for instance. These can be expected to yield a further number of narrative summaries, as can other studies such as 715.
Finally, many observers with clinical trial experience will guess that it is highly likely that a number of children who dropped out of these pediatric studies which were organized in South America, South Africa and elsewhere will have been lost to follow-up, so that FDA statements that no children enrolled in these trials committed suicide may well be in error.
I have analyzed the data from pediatric trials for suicidality and hostility using Paxil/Seroxat protocols 329, 377, 453, 676, 701, 704 and 716, supplemented by data from GSK's Canadian website, FDA medical reviews of Prozac trials for depression and OCD, published data on Zoloft/Lustral as well as an expert report prepared for Pfizer in 1997, and one of two studies on Cipramil taken from the MHRA website. A second study gives a more favorable picture for Cipramil, but I did not become aware of this study until after the analysis was finished. The results from that study would not substantially affect the results outlined here.
This analysis does not include data on venlafaxine, mirtazapine, nefazodone or buproprion. Wyeth has independently indicated that venlafaxine should not be used in children because it causes suicidality and hostility. Nefazodone has been removed for adults from European, Canadian and Australian markets.
I have broken down the studies into a group of depressed and a group of anxious studies, which involve children being treated for obsessive-compulsive disorder (OCD) or social phobia. Because I am subject to confidentiality orders, the data cannot be broken down by individual trials.
From a pool of 931 depressed patients taking the above SSRIs versus 811 depressed patients taking placebo, there were 52 suicidal acts on SSRI versus 18 on placebo. This is a 5.6% rate versus a 2.2% rate or a relative risk of 2.51. The data was analysed using a Mantel Haenszel procedure. The default procedure here gives a point estimate of the common odds ratio of 2.51, (95% C.I., 1.46, 4.34, p = 0.000899).
In a pool of 638 anxious patients taking SSRIs versus 562 anxious patients taking placebo, there were 10 suicidal acts in the SSRI group versus 1 in the placebo group, a 1.6% rate versus a 0.18% rate. When the data was analysed using a Mantel Haenszel procedure, the point estimate for the common odds ratio 11.31 (95% C.I. 1.34, 95.64, p = 0.0156).
This data is consistent with independent contributions from both the illness and the treatment. Depression carries with it a greater risk of suicidal acts than do the anxiety disorders, but in the case of the anxiety disorders the risk from treatment is no less than in the case of depression.
When these data sets are combined in 1569 patients put on SSRIs there were 62 episodes of suicidality versus 19 episodes in 1373 patients put on placebo. This is a 4% rate in the SSRI group versus a 1.4% rate in the placebo group, or a relative risk of 2.9 times greater on SSRIs. Using a Mantel-Haenszel procedure, the point estimate for the common odds ratio is 2.91 (95% C.I. 1.73, 4.91, p = 0.000041). These figures parallel the figures from adult trials submitted to the FDA as part of the license applications for recent antidepressants.
In analyzing the adult data, I started from FDA medical reviews of recently licensed antidepressants for adults. The critical methodological point in these adult studies is that it is now apparent that Lilly when submitting their data on Prozac to FDA filed as placebo suicides and suicidal acts, acts that had not happened on placebo, but had happened during the run-in phase of the trial, or that had happened months after the clinical trial was over (See Appendix 2). Despite FDA recognition that these procedures are inappropriate, Glaxo SmithKline and Pfizer have also filed under the heading of placebo suicidal acts that did not happen in the randomized phase of their respective trials (See Appendices 3 & 4).
Once this is taken into account and the figures adjusted accordingly (see Table 1), the results for SSRIs versus placebo using an exact Mantel-Haenszel procedure, with a one-tailed test for significance, the odds ratio of a suicide on these new antidepressants as a group compared to placebo is 4.40 (95% Confidence Interval is 1.32 - infinity; p = 0.0125). The odds ratio for a suicidal act on these antidepressants compared to placebo is 2.39 (95% Confidence Interval 1.655 - infinity; p < 0.0001). The odds ratio for a completed suicide on an SSRI antidepressant (including venlafaxine) compared to placebo is 2.46 (95% Confidence Interval 0.707 - infinity; p = 0.16), with an odds ratio for a suicidal act on SSRIs compared to placebo of 2.22 (95% Confidence Interval 1.47 - infinity; p < 0.001). These data have been the subject of two peer review publications at this point.
There is a striking overlap between the results in trials from adults and pediatric trials. While the rate of suicidal acts is higher in pediatric trials of depression, the relationship between active treatment and placebo is the same in both adult and pediatric groups. It should also be noted that the suicidality issue in these pediatric studies is not a matter affecting the 6-12 year old age group, showing a decline thereafter through the teenage years that could be extrapolated into adulthood. On the contrary, as Glaxo SmithKline make clear, the issue affects teenagers much more than preteens.
Finally, Glaxo SmithKline's website also contains data on acts of hostility from 4 different protocols. I have been able to review the narrative summaries and data from these and one further protocol. These protocols combined yield a total of 524 patients on Paxil/Seroxat versus 526 on plabebo, with 31 hostile episodes on Paxil/Seroxat versus 2 on placebo. Using a Fisher exact test for count data, this gives a point estimate of the common odds ratio of 15.54 (95% C.I. 3.92, 134.91, p = 0.000001).
These results are in line with the analyses of the data conducted by Andrew Mosholder of the FDA and by the MHRA, but excludes a number of drugs these authors included. This analysis represents a much purer set of SSRI drugs, and more data on SSRI drugs than has been available to other reviewers..."