AHRP Press Conference
Bethesda, September 14, 2004
David Healy MD

As of 1988 the combined worldwide clinical trial literature pointed to an excess of suicides on SSRI agents compared to placebo. This has remained the case ever since. This excess is demonstrable even though these trials were not set up to look at the issue of suicidal acts, and even though a majority of suicidal acts that have occurred are not recorded in these trials.

In fact, there is considerable distortion of the published literature. In 1989, when SmithKline Beecham submitted clinical trial data on Paxil, there was up to an 8-fold excess of suicidal acts on Paxil compared to placebo. Following on the emergence of a controversy regarding Prozac and suicide, Martin Brecher, the FDA officer reviewing the Paxil file, contacted SmithKline indicating to the company that FDA thought this was largely a public relations issue, but that it might be helpful if the company submitted another set of suicide figures. The original and subsequent figures can be seen in slides 3 to 4.

When resubmitting their figures, SmithKline took two approaches to the data. First, they included under the placebo column, suicidal acts that had not occurred on placebo in the randomised phase of the trial. These had occurred in the run-in phase of the trial - a phase lasting for a week or two during which patients have had any prior drug treatment stopped. Implying that these suicidal acts had occurred during the randomised phase of the trial is in breach of FDA regulations. FDA reviewers noted what had been done but there appears to have been no objections. This is a maneuver both Lilly and Pfizer have also undertaken. Second the companies have also included suicidal acts after the completion of the trial under the heading of placebo, and claimed as a result that the frequency of suicidal acts on active drug is no different to that on placebo. See slides 1 & 2.

When these manipulations are reversed, the figures for suicides and suicidal acts in the regulatory trials undertaken by companies lodged with FDA come out as outlined in slide 5. Combined these trials demonstrate an excess of both suicides and suicidal acts on SSRIs and other new antidepressants of almost precisely the same degree as the excess of suicidal acts found in pediatric trials - see slides 6,7 & 8.

Not content simply with the resulting parity of suicidal acts between Paxil and placebo, SmithKline and Stuart Montgomery published the data in European Neuropsychopharmacology in 1995, a journal whose editor was Dr Montgomery, and in this it was claimed that Paxil was five times less likely to lead to a suicidal act than placebo - slide 9.

This transformation was achieved by means of a "space-shuttle" fallacy, a maneuver also adopted by both Lilly and Pfizer. Patients who do poorly on SSRIs drop-out early in treatment and their exposure to the drug is therefore limited compared to patients who do well, who may be kept on treatment for up to a year. Analysing by exposure therefore weights the results towards patients doing well. A good analogy is analysing deaths per mile of space shuttle travel - slide 10. The few deaths occurring on entries to and exits from the atmosphere are minimised by the millions of miles travelled by successful shuttle, so that shuttle travel becomes safer per mile travel than walking around your own house. Treatment with SSRIs is entirely comparable in that the hazards lie on entry into and exit from treatment.

But in addition to this, Dr Montgomery had been the psychiatric expert on the British Committee of Safety for Medicines, a position he apparently filled without declaring the consultations he had with pharmaceutical companies filing their drugs for registration in Britain. Dr Montgomery was also brought by FDA to sit on the 1991 PDAC hearing into the issue of suicide on Prozac, where he spoke up vigorously against the notion that Prozac posed risks.

A further key feature of the SSRI controversies can be seen in slide 6. Based on their statements at the PDAC hearing on September 13th and 14th FDA officials would probably say that because the confidence interval for suicides on SSRIs overlaps 1.0, that these results are not statistically significant. That there is no signal. This is statistically wrong. As these trials were neither powered nor designed to look at suicides, significance testing of this sort cannot be applied to the results. The correct interpretation of the results is that our best estimate of the risk of suicide on SSRIs is that it is 2.6 times greater on SSRIs than on placebo, and that the scientific data is in fact consistent with the risk being greater than 8 times more on SSRI than on placebo.

A final point of note is the Columbia project. This took close to a year to undertake. It did not change the results FDA had to hand a year previously. During this time therefore it would appear all but certain that children will have died of suicide who need not have died. Was it worth risking these lives?

Consider the material that went to Columbia. Slides 11 -14 are taken from GSK's website. The material is not very dissimilar to the material from the case records. These are narrative summaries of the suicidal acts and instances of suicidality in these trials and the brief of the Columbia University group was to determine which of these constitute genuine instances of suicidality.

Narrative summaries of adverse events in trials are constructed primarily when trial subjects drop out of the trials. They typically range from half a page to one and half pages.

1. There is no methodology that underpins these summaries and the quality of what is written can vary hugely. These summaries will often have been written by quite junior personnel, or if written by senior personnel will be written by figures who have had little or no contact with the trial subject during the course of the trial or at the time of the adverse event. They may also be written by company personnel, where company means either for instance Glaxo SmithKline personnel or Clinical Research Organisation (such as Quintiles) personnel.
   
   
2. The summaries also were not written with the investigator or other personnel having the trial subject - in this case the child and/or their carers - present for further questioning. These summaries will often have been written some time after the event and in some cases after all contact has been lost with the child or their carers so that the summariser will not have known for certain what has happened the child.
   
   
3. If not written by company personnel, the summaries can often be written with a representative of the company organising the clinical trials there in the room.
   
   
4. The summaries will appear in English from investigators whose primary language in many instances will not be English.
   
   
5. The summaries may not be blind. Prior to constructing narrative summaries, clinicians make a series of determinations about the event in response to questions about the severity of the event, the ongoing status of the event and the relatedness of the event to intake of the study drug.
   
   In the case of the Paxil summaries it is a matter of public record that clinicians were significantly less likely to conclude that these suicidal events were related to study drug when the drug was Paxil than when the drug was placebo. In other words when children became suicidal on placebo, the clinicians decided that this event was definitely or probably related to placebo intake but when children became suicidal on Paxil they decided these events were probably not or definitely not related to Paxil intake. See slide 14, where even though the investigator must know the child is on placebo, the event is still attributed to placebo.
   
   It is likely that in some cases, investigators will have made judgements that for instance a problem like increased anxiety that started perhaps a week or two into the trial was related to drug intake, and then in addition perhaps that a further problem like a tremor emerging a week or two after further treatment or perhaps after a dose escalation was related to treatment but these same investigators will have then gone on to deny any relationship to drug intake in the case of a suicidal event occurring days later.
   
   Having made this judgement of non-relatedness, the clinicians then go on to either construct the narrative summary or pass the details to company personnel who write the narrative summaries.
   
   
6. As narrative summaries are constructed primarily when subjects drop out of trials, one consequence is that at this point a review of the events may lead to a first appreciation, or at least first recording of the fact that in the case of a particular child there had been prior suicidal acts. There will in general be no narrative summaries for these prior suicidal acts. Thus the set of narrative summaries that FDA passed to Columbia are incomplete. This was recognised by FDA at the February 2nd PDAC hearings.
   
   
7. It also appears that there have been placebo controlled trials of children who are depressed - viz protocol 716 in the case of Paxil - that contain a large number of narrative summaries that FDA had not been given as of February 2nd and which therefore were not passed to Columbia at least in the first instance.

Probably in part out of a recognition of some of the above factors, other regulatory authorities have not engaged in the exercise of getting outsiders to scrutinise these narrative summaries. Indeed the PDAC hearing on February 2nd, while not deterring FDA from going ahead with this analysis, at several points categorised it as a pointless exercise - "searching for a needle in a haystack".

FDA's decision to proceed with an analysis appeared to depend in great part on the fact that there are a very small number of these narrative summaries - on both placebo and active drug side - that do appear to outline trivial events. The majority of events however look quite severe. In the case of some of these trivial events, Columbia as I understand were asked to decide if this should be regarded as a suicidal event. The implication is that some such events will be dropped from the dataset. However, even this may be a mistake. If the investigator has indicated a coding in the suicidality group, transmission of this event through a third party medium may well lead to a loss of important context compromising the ability of a later group, such as that from Columbia, up to 10 years later, to decide what has happened.

In the case of the suicidal acts from pediatric trials it would be helpful to analyse the timing of the events in relation to dose transitions. It is important in this regard not to accept company statements that the mean length of time from the onset of study medication to the suicidal event is greater than might be expected if the drug were causing the problem. Alternatively companies will have indicators as to whether the event occurred within a week of a dose transition. A proper analysis would look at all dose transitions and establish the mean and median durations between dose transitions and the events of concern. This it seems was not requested and not done.

Given these caveats regarding the exercise FDA undertook, during which children's lives were put at stake and in all probability lost, it seems reasonable to question why this exercise was undertaken?