ANTIDEPRESSANTS AND SUICIDE
BRIEFING PAPER, 20 JUNE 2003
Current trials provide evidence of an antidepressant treatment effect rather than evidence that antidepressants work. This is of importance in any calculation of risk-benefit trade-offs and related warnings.
Current trials provide some evidence for a suicide reducing effect of SSRIs in some cases, but this benefit only seems likely to be realised in the context of warnings and the monitoring that would go with such warnings.
There is a long clinical tradition of recognising antidepressant-induced suicidality.
Traditional and company methods of assessing cause and effect leave little doubt that SSRIs can induce suicidality.
Companies have failed to report in full their clinical trial data on suicidal acts.
Current clinical trial data on the relative risk of suicidal acts on antidepressants in either children or adults points to comparable elevations in the risk of suicide induction on active agent compared with placebo an approximately 2.5 times greater risk, not including any risk that might stem from withdrawal effects.
Current clinical trial data produces rates for suicidal act induction on antidepressants that overlaps with the evidence from epidemiological studies in primary care.
There are major disparities between warnings given about benzodiazepines and SSRIs and these disparities have no evidence-based justification.
There is simply no basis for thinking that patients with milder nervous disorders being treated in primary care are at any significant risk of suicide and therefore the notion that they are at risk as they emerge from their depression is meaningless unless they are at risk from the treatment, in which case the wording of the SPc and PIL should make this clear. This position applies also to the many patients being treated with these drugs who are not depressed, and should apply to healthy volunteers also.
The issue of suicidality in children and adolescents on Seroxat recently has led to coverage that has stressed possible differences in the risk benefit profile between children and adults for SSRIs like Seroxat. Seroxat apparently has been linked to suicidality in children that is not offset by any benefits hence concerns about usage. This balancing of risks against benefits will imply to many that Seroxat and other SSRIs have been proven to work in adults, and that these positive effects change the issues as regards warnings.
There are a number of points at play in such a formulation that this paper will attempt to tackle. One of these is the notion that SSRI agents have been proven to work for adults. There are grounds to dispute that SSRI have been proven to work in adults even to the same extent that they have been proven to cause suicidality. Indeed arguably SSRIs almost in principle cannot be proven to work this is not to say that the SSRI group of drugs should not have been licensed. But these issues are important when it comes to considering off-setting risks against in an adult population. Second, even if there are off-setting benefits, it is not clear that such benefits should stand in the way of reasonable warnings. Some case might be made for a trade-off if there were clear data indicating that benefits would be lost by warnings. The data as it stands however points in the opposite direction.
The Benefit in Risk-Benefit Determinations
The majority of current SSRI antidepressants come with evidence that the drug outperforms a placebo in some domain of measurement in a minority of trials in which the drug has been tested. This provides evidence that a treatment has an effect in some trials, which can be taken as a proof in principle of a treatment effect. But this proof in principle is not evidence of treatment effectiveness and almost certainly is not evidence that the treatment works in general.
For example, two trials of citalopram showed a difference from placebo whereas three did not leading to this quote from Paul Leber, head of the CNS division of the FDA through to 1998:
"there is clear evidence from more than one adequate and well-controlled clinical investigation that citalopram exerts an antidepressant effect. The size of that effect, and more importantly the clinical value of that effect is not something that can be validly measured at least not in the kind of experiments conducted. Accordingly substantial evidence in the present case, as it has in all other evaluations of antidepressant effectiveness, speaks to the proof in principle of a products effectiveness.
"The Clinical Efficacy Trials subsection within the Clinical Pharmacology section not only describes the clinical trials providing evidence of citaloprams antidepressant effects, but make mention of adequate and well controlled clinical studies that failed to do so. I am mindful, based on prior discussions of the issue that the Office Director is inclined toward the view that the provision of such information is of no practical value to either the patient or prescriber. I disagree. I believe it is useful for the prescriber, patient, 3rd party payer to know, without having to gain access to official FDA review documents, that citaloprams antidepressants effects were not detected in every controlled clinical trial intended to demonstrate those effects. I am aware that clinical studies often fail to document the efficacy of effective drugs, but I doubt that the public, or even the majority of the medical community, is aware of this fact. I am persuaded that they not only have a right to know but that they should know. Moreover I believe that labelling that selectively describes positive studies and excludes mention of negative ones can be viewed as potentially "false and misleading".
In the application of Zoloft/Lustral for a license, arguably only one of six studies stood out as clearly indicating a superiority of Lustral/Zoloft over placebo. A further study employing a discontinuation design had positive results that may well have stemmed in part from withdrawal from Lustral/Zoloft. [Apparently this discontinuation design played a big part in the CSMs decision to license Lustral according to Pfizer consultants/CSM members]. Lustral/Zoloft however did less well than amitriptyline when it was compared to that and it failed in two hospital depression studies.
This led to this quote from Leber:
"how do we interpret.. two positive results in the context of several more studies that fail to demonstrate that effect? I am not sure I have an answer to that but I am not sure that the law requires me to have an answer to that -- fortunately or unfortunately. That would mean, in a sense, that the sponsor could just do studies until the cows come home until he gets two of them that are statistically significant by chance alone, walks them out and says he has met the criteria".
The notion that a minority of trials showing a treatment effect can be taken as evidence of a proof in principle of therapeutic effectiveness is important if you consider that regulatory statutes require that a drug be proven to be effective. The basis on which current antidepressants are licensed therefore is very slender. It is not on the basis that drugs have been proven to be effective but rather that there is proof in principle that they might be effective.
Altogether, across recent trials submitted for regulatory purposes, active compounds have failed to distinguish from placebo at an approximately 45-50% rate. This has led to a number of recently analyses questioning whether antidepressants in fact do work.
Concerns that the evidence base is slender do not stop with the frequent failure of antidepressants to differentiate from placebo. Even in those trials where they do separate from placebo, the treatment effect is something inferred from appropriate changes on rating scale measurements from one domain only - basically changes the treating physician wants to see. The patients point of view has also been assessed using quality of life scales, but the results of 90% of over 100 trials, in which such instruments have been used, have been left unpublished. Where results have been published the SSRIs do not appear to do things that patients appreciate as useful.
In essence therefore these drugs can be shown to do something from one point of view in particular. This makes it possible to claim that there is proof in principle of a drug effect.
This proof in principle is sufficient in my opinion to warrant a license for treatment. But it is not evidence of effectiveness. The position as regards effectiveness would be more secure if it was clear that ongoing treatment with that drug does not lead to emergent problems on discontinuing treatment. If a treatment for which there are slender grounds for claiming that it is effective is then further compromised by evidence of a withdrawal syndrome there can be real public concern about whether such a drug should be marketed without appropriate warnings.
In contrast if a treatment truly worked, the problem would go away. Even a mental illness goes away in the face of effective treatment. GPI used to account for 5-10% of presentations to asylums but penicillin "works" for this, and there are no cases now. In contrast, since the launch of the SSRIs we have more and more depression. The effects of the SSRIs are therefore more like those of the H-2 blockers than they are like the effects of antibiotics for ulcers they appear to have led to increases rather than reductions in the size of the population apparently affected in a manner that is inconsistent with claiming that these drugs work.
These issues are particularly clear when it comes to SSRIs and children. In line with other trials of older antidepressants in children, it has been difficult to show that these drugs do much. However using washout designs, it is possible to demonstrate that treatment can have an effect in some children. This fits in with clinical impressions. But what is clear from the difficulties in demonstrating a treatment effect is that while it might be possible to demonstrate an effect, and such a demonstration can help support the use of these drugs clinically, the evidence base does not warrant any generalisation to claims that antidepressants work for children.
In general the trials undertaken on SSRIs in both adults and children have been conducted in highly artificial samples of convenience that do not provide grounds for assuming treatment effects will either generalise to the population at large or in fact assuming that there is likely to be efficacy in any substantial number of patients.
These results however do not prove the drugs dont provide benefits for or even "work" for some patients. The pattern of results of rather minimal benefits compared to placebo in trial samples may stem from clearly favourable responses in a proportion of patients and clearly adverse responses in others. Indeed this explanation is much more consonant with clinical experience than the alternative that the drugs have a relatively minimal effect in a bare majority of cases.
In such cases treatment may be warranted but there is a considerable onus on effective warnings and close monitoring. These principles underpinned the British Association of Psychopharmacology consensus statement on the use to psychotropic drugs in the treatment of children and adolescents.
It is against this background of mixed benefits that the risks of suicidality and withdrawal need to be assessed.
Finally, the MHRA assessment of an effect of Seroxat on suicidal act induction in children and teenagers is an example of an assay system in which there has been a demonstration of a proof in principle that Seroxat, and by extension all other SSRIs, can produce suicidality. Even if other assay systems do not come to the same conclusion, just as failed trials of antidepressants did not stop the licensing of these products, so also inconclusive assays of the suicidality issue should not count as an argument against warnings.
Determination of Cause and Effect
In pharmacology, the standard methods for demonstrating cause and effect involve an application of principles derived from Kochs postulates. The key elements are the appearance of the phenomenon with some reasonable temporal association between that appearance and drug intake. The phenomenon ideally should clear up on drug discontinuation, and reappear on re-exposure. This pattern involves Challenge-Dechallenge-Rechallenge (CDR) relationships.
A second particularly strong indicator of a relationship lies with evidence of a parallel between the dose of the drug and the frequency of the adverse event. A third pointer lies in the demonstration of some understanding of the mechanism underpinning the event that permits an intervention to ameliorate the event.
In the case of the SSRIs and suicide, as early as 1992, there was replicated evidence of CDR, and dose response relationships between these drugs and suicide as well as evidence for an ability to intervene on the basis of antidotes to a proposed mechanism of action. A clear causal link had been established that has been replicated on many occasions for instance in Pfizers dose finding studies of Zoloft/Lustral in children there is clear evidence of dose response relationships between Zoloft/Lustral and suicidal acts, with C-D responses leading to causal attributions to Zoloft/Lustral-induced suicidality.
All that remained to be done after 1992, arguably, was to determine the frequency with which this problem might be happening.
The notion that test-retest methods as well as dose response relationships form the basis for determining causality has been endorsed by most senior figures in the field who have considered these issues. It is enshrined in the federal judicial manual. Its presumably underpinned the CSMs determination that benzodiazepines can lead to suicidal or homicidal behaviour and such methods are used routinely by pharmaceutical companies including Pfizer, Lilly and GlaxoSmithkline in assigning causality to a particular adverse effect.
A large number of documents now in the public domain show clinical monitors in Lilly, Pfizer and GSK linking suicidality to their drug. These linkages in fact extend back to the early 1980s at least and to studies in healthy volunteers.
One such example comes from Pfizer in 1999, when asked to investigate the effect of Zoloft/Lustral on suicidality by the Irish Medicines Board. Pfizer offered a document in which there were 252 suicidal events in a clinical trials programme, with a number of actual suicidal acts in triple figures. In 10% of cases, the investigator had linked the suicidal event to Zoloft/Lustral intake. In an even larger number of cases, Pfizer reviewers had linked the suicidal event to intake of Zoloft/Lustral, so that overall 21% of cases were linked to drug intake. These linkages in the case of Pfizer reviewers can only have been made on the basis of assessing classic determinants of causality viz. CDR, dose-response and temporal relationships between drug intake and the adverse event.
(Quite aside from causal determinations, based on these figures, there is every reason to believe that all companies have sufficient clinical trial data on file to analyse the linkage between their drug and suicidality, taking into account the range of underlying conditions treated, the dose of drug used, and the duration of exposure. Such data however remain unavailable to the academic community.)
RCTs and Causality
In contrast to this approach to causal determination, it is not possible to find an authoritative source for the notion that RCTs are the way to determine cause and effect in this domain. RCTs, as a subset of epidemiology studies, provide evidence of associations rather than evidence of cause and effect.
This view seems to be supported by recent regulatory developments. The FDA for example will not permit Lilly to make a causal connection between Prozac and suicidal acts on the basis of RCT data that indicates suicidal acts and deliberate overdose occur in their clinical trial database at a significantly greater rate than on placebo.
RCTs do not establish a causal link but rather point to the frequency with which an association may happen. In some cases RCTs and epidemiology in general may be the only method to use as in situations where CDR and dose response relationships cannot be established readily.
RCTs and epidemiology become particularly problematic when a treatment both produces benefits on a target effect and in other cases produces the emergence of this target effect. This gives rise to a situation parallel to that in which vaccination for instance may reduce the frequency of brain damage but yet cause brain damage in its own right. In such circumstances a relative risk of less than 1.0 may even be grounds for serious concern.
Effectively RCTs are irrelevant to the issue of SSRIs and suicide unless the suicide inducing effect is so great that it dwarfs any suicide sparing effect. This in fact appears to be the case, and in this case RCT data provides the basis not for claiming cause and effect but for providing warnings.
An alternative scenario in which an RCT might play a more central role in the determination of cause and effect would be if a study was specifically designed to investigate a particular adverse event. In fact no such study to investigate links between SSRIs and suicide has been undertaken. But this is not because such a study would be unfeasible or unethical as suggested recently in the Lancet. This editorial pointed to the difficulties investigating a feature that may be both part of the disease and an effect of a drug. But in fact clinicians in practice and clinical trials can distinguish between the sleeplessness that is part of depression and that may follow SSRIs, as well as the anorexia that is part of depression and may follow SSRIs and the sexual dysfunction that is part of depression and may follow SSRIs.
It is similarly possible to distinguish the suicidality that may be part of depression and may follow SSRIs. Lilly in conjunction with the FDA drew up a blind randomised rechallenge protocol to investigate this issue in 1990 complete with scales sensitive to the emergence of suicidal ideation. Investigators had been recruited and pills encased in blister packs by 1991. The trial never took place.
The failure to undertake this trial faces anyone who would adhere to the notion that test-retest and other methods can only provide a signal and that RCTs are what determine causality with a problem. You either say that when these methods provide a signal, there should be a trial. In which case, where is the trial? Or else you concede cause and effect. There was a very clear signal present, sufficient to warrant the execution of a specific RCT but companies have deliberately opted not to undertake such a study.
Not only have companies failed to undertake such a study, but they have failed even to include the specifically created suicidal ideation scale in subsequent trials this could have easily been incorporated into trials by Lilly and other SSRI companies and would probably on its own have settled the issue without the need to run a specially constructed trial. Instead there has been the ludicrous example of companies arguing on the basis of changes in item 3 of the HAM-D when this scale was rated by clinicians unaware of the possibility of adverse outcomes on treatment who will not have asked the questions that might have elicited the phenomenon in question.
Instead what companies in effect have done is to use the lack of evidence of a trial to argue that the signal provided by test-retest methods etc is of no value, and that the judgments of their own monitors and investigators count for nothing.
Quite aside from any consideration of determining cause and effect, what the data on frequency of suicidality stemming from RCTs and epidemiology can do is to indicate whether there is a case for warnings. It could be argued that if a particular adverse effect was shown to have a causal link to a treatment but that event happened very infrequently it might be inappropriate to warn about the adverse event as patients who might benefit from treatment would be discouraged from seeking treatment. This might, in the case of the antidepressants and suicidal acts, lead to a greater frequency of for example suicides.
Just this public health argument was put forward by the FDA in the 1991 hearings on Prozac. It amounts to a policy of covert vaccination or what was termed by Plato The Noble Lie. It may have some ethical justification if there was evidence that treatment on balance more commonly produced good outcomes than bad outcomes. The clinical trial evidence reviewed below makes such a position difficult to sustain. This data on the frequency with which treatment produces adverse outcomes is such that the case for warnings would seem unanswerable.
There is some evidence that SSRIs can reduce suicidality. But it is only in the event of proper education, warnings and monitoring that the potential benefits of treatment on suicidality, which do exist, are likely to be realised with these drugs.
Suicide Reduction on SSRIs
Concerns about the effects of SSRIs on suicidal behaviours first surfaced with zimelidine. In clinical trials of this drug there was a greater frequency of suicidal acts on zimelidine than there was on placebo. This led to an analysis of the zimelidine clinical trial database which indicated that while there might have been a greater frequency of suicidal acts on zimelidine, that some of the patients who were more suicidal to begin with appeared to do somewhat better on zimelidine than on comparator. This appeared to indicate that the greater frequency of suicidal acts on zimelidine was an artefact; that this drug did not trigger suicidality. This analysis of the zimelidine figures was undertaken by Stuart Montgomery.
When fluvoxamine was launched, the clinical trial programme for this drug also showed a greater frequency of suicidal acts on fluvoxamine compared to comparator and placebo. Comparable analysis of the fluvoxamine database was undertaken to that that had been carried out for zimelidine and this also showed that many patients who are more suicidal to begin with appear to respond to fluvoxamine. Again the regulators, primarily the German regulators, were persuaded that this indicated that the greater frequency of suicidal acts on fluvoxamine was an artefact.
Stuart Montgomery it would seem advised Lilly to undertake a comparable analysis when the controversy arose surrounding the risks of suicide induction on Prozac. Lilly undertook an analysis of this kind on a selected set of their clinical trials and it was this that gave rise to the Beasley et al paper in the BMJ. In this paper there was an excess of suicidal acts on Prozac compared to placebo, but compared to placebo there were greater reductions in the scores on item 3 of the Hamilton Rating Scale for Depression.
Stuart Montgomery and representatives of SmithKline Beecham have undertaken a similar analysis on SmithKlines database for Seroxat/Paxil. Indeed, analyses of this kind became standard in the course of 1990 with both SmithKline and Pfizer handling their data this way and FDA reviewers using the apparent suicide reduction effect to allay concerns about possible suicide induction.
There are a number of consequences of this strategy. First, in so far as it is accepted that suicide reduction takes place in certain subjects, and I accept this, then the figures for the relative risk of suicidal acts on SSRIs compared to placebo, outlined below, underestimate by a corresponding factor the true extent to which SSRIs are inducing suicide in some.
Second, this suicidality reducing effect does validly underpin company claims that their drugs can reduce the risk of suicide. However, while the drugs may reduce the risk of suicide in some, it seems clear from clinical trials overall that in fact any benefits in this area that are being realised are likely to be outweighed in situations where patients are not warned. In the absence of warnings, the RCT evidence indicates that the potential benefits on suicide reduction are not likely to be achieved and that in fact these drugs increase the risk of suicide.
Antidepressants and Suicide Induction: Early Thinking
Up to approximately 1955, the dominant theories linking depression to suicide stressed that the points at which suicide risk were greatest were on entry into and exit from a depressive episode. This theory was put forward most clearly by Staehelin from Basel. It applied to severe depressive disorders of the kind that might need treatment with ECT for example. Far from ECT leading to suicide on the emergence from depression, however, there is convincing RCT evidence that ECT has a potent anti-suicidal effect.
The role of antidepressants in suicidality that might develop on emergence from a depressive episode needs to be interpreted against the background of agitation induced by reserpine. Reserpine was one of the first psychotropic drugs. While primarily used as an antipsychotic, in fact the only placebo controlled prospective trial of reserpine shows it to be an antidepressant with as great a treatment effect as Prozac or other SSRIs on a very similar population to those that had been recruited to Prozac trials.
Reserpine was also an antihypertensive and it was from populations of patients with no nervous problems whatsoever who were being treated for hypertension that starting in 1955 there were reports of a troublesome side-effect, which came to be called akathisia were first described. This was linked to suicide.
Akathisia as will be clear from the following quotes is probably better described in many cases as agitation:
"increased tenseness, restlessness, insomnia and a feeling of being very uncomfortable" ,
"the first few doses frequently made them anxious and apprehensive... they reported increased feelings of strangeness, verbalized by statements such as I dont feel like myself .. or Im afraid of some of the unusual impulses that I have" .
Sarwer-Foner and Ogle describe the case of CJ who on the first day of treatment reacted with marked anxiety and weeping and on the second day "felt so terrible with such marked panic at night that the medication was cancelled".
The entire field of psychopharmacology to a person believes that Reserpine causes suicide, in great part because the suicides that happened on Reserpine happened in normal individuals without mental problems being treated for hypertension.
The phenomenon outlined above came to be called akathisia but might better be termed agitation. Arguably a misleading impression emerged that Reserpine caused depression. This stemmed partly from the fact that the states that Reserpine induced led to suicides. This agitated state however responded to discontinuation of treatment and could be alleviated by for example a stimulant. Psychiatric observers of the phenomenon in the late 1950s did not classify this new problem as a depressive disorder.
Two years later, in the first large scale study of Imipramine undertaken by Kielholz and Battegay in Basel, despite operating against a background of Staehelins ideas, when two of the patients in the trial of Imipramine committed suicide Battegay and Kielholz argued that Imipramine had played a part in the suicide.
The following year Mann and MacPherson in one of the first publications from North America on the serotonin reuptake inhibitor (SRI) Imipramine, and in language reminiscent of what was happening with reserpine, noted that in two of their patients treatment had to be discontinued immediately after two days medication in two patients who stated that
"they felt completely overwhelmed by the drug with very marked dizziness, severe palpitations, jerking movements and markedly increased tension and agitation on the usual initial dosage scale".
The similarity between these reports on both imipramine and reserpine and those reported following Panorama is striking. There would seem almost no chance of copycatting here in that the reserpine and imipramine quotes cited here are not well known.
By the early 1960s a new view had emerged which was that the point at which an individual emerged from a depressive disorder while on treatment was a period of risk. However, this risk was not simply one that was linked as before to the resolving depressive disorder, risk was now also linked to the treatment. It was argued by Kielholz that for example monoamine oxidase inhibitors (MAOIs) posed a greater threat than tricyclic antidepressants. This would not be the case if the risk stemmed purely from the depression as MAOIs were thought to be less effective than tricyclic antidepressants. The greater risk from MAOIs was thought to stem from the fact that they were more stimulant.
The dominant trend in European thought therefore did accept as is contained in current warnings for SSRIs in the UK that the point of emergence from a depression is a particularly risky period but this dominant view included a recognition that this risk stemmed in part from the treatment itself.
For example, in a draft article written by Professor Ulrik Malt in 1994, following the emergence in a clinical trial of suicidality on Zoloft/Lustral: Malt put it as follows: "Since the introduction of the tricyclic antidepressants, it has been known by clinicians that TCA could increase suicidality in the first week. For this reason a close supervision of depressed patients given TCA was recommended". (However the final published BMJ article had no mention of this or the suicidal acts on Zoloft/Lustral compared with no such acts on placebo).
In the meantime, having remained un-investigated from the better part of 40 years, drug-induced akathisia is now firmly linked to suicide; it is enshrined in DSM-IV as causing suicide. The CSM furthermore note agitation a common side effect of SSRIs. And a recent letter from GSK under the heading of Does Seroxat cause side effects states that akathisia will now be included in the PIL.
Such a situation lays the basis for a potential class-wide warning on antidepressant and suicide risk.
What has been left almost completely unconsidered in this debate hitherto, however, has been the impact of drug treatment on individuals who are not suicidal to begin with. This is an issue that the recent Panorama programme has brought into focus.
Until the advent of the SSRIs, those being treated with antidepressants were at a significant suicide risk. In the course of the 1990s, cases of Valium have been converted into cases of Prozac, but behind these labels and marketing campaigns are people who were never thought to be at a significant risk of suicide. In addition a range of altogether milder problems in ever-younger people have become targets for drug treatment.
The epidemiology section below offers estimates of the likely rate of suicide in cases now being diagnosed as depressed in UK primary care along with best estimates for the rate of suicide in those on SSRIs.
Suicide on SSRIs: The First Regulatory Submissions
The story begins circa 1985/1986, when Eli Lilly had trouble getting Prozac licensed in Germany. One of the hazards flagged up by German regulators was an association between Prozac intake and suicidal acts. The German regulators had prior experience of two SSRIs, zimelidine and fluvoxamine, both of which seemed to be associated with increased rates of suicidal acts. In the case of both zimelidine and fluvoxamine, experts had briefed regulators that this might be simply a coincidental finding as when the data sets were analysed further it seemed that those who were most suicidal to begin with did best with the SSRI.
After considerable internal company assessment of the issues, Lilly resubmitted a portfolio of data to the German regulators and comparable data to other regulators in the US and the UK. The German regulatory submission in 1986 gave the data on suicide as in Table 1 below.
% Suicides & Suicidal Acts
From Table 1 it can be seen that there is an excess of suicidal acts on Prozac whether these data are calculated in terms of the absolute numbers of patients or in terms of patient exposure years. Through to the mid 1980s the preferred method of data submission to regulatory authorities was in terms of absolute patient numbers.
Table 1 however is misleading. Of the 15 patients described here as committing suicidal acts and falling in the non-Prozac group, scrutiny of the clinical trial material in Brickler exhibit 1 reveals that 4 of these occurred during the placebo run in (placebo washout) phase of various clinical trials. A further 4 appeared to have occurred at some point up to a year after the trials were over. These patients were recruited by following patients up over the course of a year and if any engaged in a suicidal act at some point during that year, even if they had been on Prozac beforehand, provided they had been discontinued from Prozac for 6 weeks beforehand, these were classified under comparator.
Removing these two patient groups leads to the figures in Table 2, which gives much clearer evidence of an increased rate of suicidal acts on Prozac compared to other whether the data are calculated in terms of either patient exposure years or absolute numbers of patients.
% Suicides & Suicidal Acts
There are further possibilities here. One is to accept Lillys approach and include these extra patients in the comparator group but to analyse the resulting data appropriately. In this case, the patient number denominator needs to change so that it includes all patients who have entered the study as all go through a washout stage and all effectively survive through to a years follow-up. If this is done we end up with the figures in Table 3.
% Suicides & Suicidal Acts
This in fact is methodologically inappropriate but borrows from exactly the methods used by companies to hide the problem.
Two quibbles can be raised about the figures here and for other agents below. First, ideally like should be compared with like so that only figures from placebo controlled randomised trials should be aggregated in this way. When this is done, there is still a clear excess of suicides on Prozac in this instance and other SSRIs in other cases.
It is also argued that any open or non-randomised data included here will be data from difficult patients which therefore lead to an artificial increase in the suicidal act figures for the SSRI. In fact, data from open or non-randomised patients when aggregated across SSRIs yields a lower rate of suicidal acts. This stands to reason in that patients in compassionate or other protocols are likely to be so because of a demonstrated beneficial response to Prozac or other SSRI.
A second argument can be put forward here. These are the data submitted by the companies, who faced with a problem have gone to extraordinary lengths to conceal that problem viz deliberately miscoding data. It seems hardly likely in the circumstances that they will have done anything to make a precarious position look any worse.
Methodological Issues: The Shuttle Fallacy
From the mid to late 1980s, data on aspects of treatment began to be presented to regulators in terms of patient exposure years (PEY). The calculation of an outcome in terms of patient exposure years or in terms of survival curves makes sense in the case of an illness such as breast cancer where one looks at the length of time to recurrence. In these circumstances there is no particular reason not to use patient exposure years given the clear cut nature of the endpoint and in particular if the treatment protocols are not being used to demonstrate dose response effects to treatment.
However, if the issue at stake is a drug induced adverse effect, in which dose response relationships would clearly be involved if there is a causal relationship between the drug and the adverse effect, and if the full details of variations in dose during the course of the treatment protocol are not known, then calculating the incidence of adverse effects in terms of patient exposure years is inappropriate.
It is also inappropriate to use patient exposure years where the end point is not clear-cut, or is one that characteristically appears early in treatment, or is one that can be avoided by remedial measures.
In contrast to the appearance of breast cancer, which cannot be avoided, the evolution of a clinical picture to a suicidal act in patients taking a psychotropic drug is open to outside influence. Patients commonly first get agitated. Patients who are agitated can be recoded as failures to respond to treatment rather than as patients suffering from an adverse effect of treatment. Whether the origin of the agitation is linked to treatment or a failure to respond to treatment, the patient can be discontinued from the treatment protocol.
In the case of SSRI agents and possible suicidality, the only reasonable end point that would justify the use of patient exposure years or a survival analysis is the appearance of agitation. However, agitation itself can be a feature of the illness and hence even this is not unambiguous in terms of how it might be coded by companies. In the case of trials on Zoloft/Lustral, whatever the origin of the agitation, trials that I have complete details on show a drop-out rate for agitation of 4.75% on Zoloft/Lustral versus 0.65% on placebo. There are comparable findings for other SSRIs and as noted the CSM/MHRA lists agitation as an adverse event caused by SSRIs.
A further possibility is the adoption of the use in clinical trials of a rating scale sensitive to the emergence of suicidal ideation. As outlined above, such a scale was prepared by Lilly in 1990 but appears not to have been adopted in any of their clinical trials.
There is a further difficulty in terms of patient exposure years calculations in this series of antidepressant trials, which is that many of the trial protocols have involved patients who are crossed over to drugs like Prozac and it will not be clear in many instances just when the calculation of exposure time starts. It should ideally start after commencement on Prozac rather than after commencement in the trial. The relation between any dose escalation of treatment and suicidal effects in such protocols is also uncertain.
Against this methodological background, given that many of the patients who drop out are likely to be the ones sensitive to the drug induced problem, the use of patient exposure year protocols and calculations will breach randomisation by leading to the artificial selection of a group of patients who are good responders to the drugs.
The use of PEY in this context can be likened to calculating the risks of death on the space shuttle in terms of miles travelled. Clearly if a shuttle gets safely into orbit, the fact that it then travels millions of miles means that the number of deaths per million miles is probably less than the number of deaths per million miles walking around the inside of ones house. But no-one would think this was a reasonable way to estimate risk in this case. Going onto or halting a psychotropic drug is quite comparable to the take-off and re-entry component of space shuttle travel, whereas patients stabilised on treatment are in a position comparable to those of an astronaut orbiting safely.
In the mid to late 1980s, regulators in the United Kingdom in particular, but the US also to a lesser extent, faced a crisis regarding the benzodiazepine group of drugs. These drugs, the minor tranquillisers, had been the dominant drugs on the psychotropic market up till then. But in the early 1980s, dependence on benzodiazepines was described and the benzodiazepine group of drugs fell under a cloud.
In 1988, the MCA issued guidance that the benzodiazepines can make people suicidal and can also make them dependent. As regards making patients suicidal the basis for this statement comes from classic CDR reports, as well as evidence of dose response effects, and temporal relations between drug intake and outcome. These are the standard methods of demonstrating causality in this clinical domain.
At present, there is therefore a striking inconsistency in the way evidence is being handled as regards the benzodiazepines and the SSRIs.
The Emergence of a Public Controversy
In February 1990, a paper appeared (Teicher et al 1990) demonstrating on a CDR basis, with some evidence of dose response effects that Prozac can lead to the emergence of suicidal ideation in patients being treated with it. A series of other studies followed providing further evidence of this type.
Similar reports were filed with all companies producing SSRIs in the succeeding years, either spontaneously or in the course of clinical trials. In the course of following-up of these reports, company monitors indicated a causal relationship between each of the major SSRIs and suicidal acts based on CDR, dose-response indicators, and temporal relationships.
In a large number of cases company monitors over-rode the assessments of treating physicians to indicated that the SSRI had caused the problem, even when the treating physician had indicated a belief that the adverse effect was not drug related.
The Licensing of Sertraline & Paroxetine
By 1991, Pfizer had applied to get Zoloft/Lustral licensed for the treatment of depression, and SmithKline had applied to get Seroxat/Paxil licensed for the treatment of depression. The data from the clinical trials undertaken by these companies had been lodged with the regulators. The data on Zoloft/Lustral drawn from FDA reviews show the following frequencies of suicides and suicidal acts Table 4.
% Suicides & Suicidal Acts
And for Seroxat/Paxil drawn from FDA reviews see Table 5.
% Suicides & Suicidal Acts
However again as with Prozac there are discrepancies between these tables and the underlying raw data. In the case of both Zoloft/Lustral and Seroxat/Paxil, placebo run in / washout suicides and suicidal acts have been coded as placebo suicidal events. Retabulating the data in a manner that separates run in from true placebo on the basis simply of scrutiny of FDA reviews yields the data in table 6.
% Suicides & Suicidal Acts
In the case of Seroxat/Paxil there is some dispute as to what the figures originally presented were. There is evidence for a figure of 1 suicidal act only on placebo. The medical review of Seroxat/Paxil lodged on the FDA site indicates that there were no more than 3 suicidal acts on placebo in contrast to the 8 published by GSK and used as a basis for calculation by the FDA.
It would appear that following an original submission of data with as few as 1 suicidal act at the end of 1989, the controversy re Prozac emerged and SmithKline were invited in a somewhat unorthodox fashion to consider submitting a different set of figures.
The FDA and other regulators were in possession of these data at the time that the psychopharmacological drug advisory committee (PDAC) meeting on Prozac took place in 1991 to consider whether there was any evidence that Prozac could trigger suicidality. These regulatory hearings did not hear about any evidence on Zoloft/Lustral or Seroxat/Paxil. The hearing came to the conclusion that the case against Prozac had not been proven.
One factor invoked in the Prozac PDAC hearings was a public health argument, according to which it might be possible to warn that Prozac can cause problems for some but those warnings might deter others who would benefit from Prozac from seeking treatment such that the overall effect of a warning might be an increased rate of suicide and suicidal acts.
This raises intriguing questions about whether in effect in the SSRI story there has been a process of covert vaccination. Some patients who will be harmed by the drug have been deliberately kept uninformed so that others might benefit.
Whatever about the ethical aspects of such an approach, a lot hinges on demonstrations that overall Prozac and other SSRIs reduce rates of suicide and suicidal acts in patients. The above figures make it impossible to argue that in the main SSRIs reduce suicidal events.
In fact, the original problem thrown up by Teicher et al suggested that perhaps only a small subset of patients would be adversely affected by SSRI induced akathisia/suicidality. This led to a methodological problem, namely was it possible to detect any signal from an adverse effect of this type against the background of a larger number of patients benefiting from Prozac and other SSRIs. This problem appeared to necessitate a CDR placebo controlled randomised trial design. Such a design was drawn up by Lilly in collaboration with the FDA but was never instituted.
The figures in Tables 1 6, however, demonstrate that these methodological difficulties do not in fact arise. This is because while the SSRIs as a group may make some patients less suicidal, they do not appear to have a protective effect in the main but actually pose a risk in the main. This risk comes through in an absolute increase in the number of suicidal acts on drugs compared to placebo. This absolute increase holds true for all SSRIs on which data have been submitted to regulators. Data for citalopram and venlafaxine are included below in Table 7. (There are no details on whether all of the suicidal acts coded for placebo actually occurred on placebo). Comparable problems had been noted with zimelidine and fluvoxamine.
% Suicides & Suicidal Acts
Analysing the data from all these agents combined (see Table 10; Appendix B), using an exact Mantel-Haenszel procedure, with a one-tailed test for significance (assuming the figure for suicidal acts on placebo in Seroxat/Paxil trials to be 3 and not 1; and using the NDA figures for Prozac rather than all figures as given above) yields the following odds ratios:
The odds ratio for a completed suicide on an SSRI compared to placebo is 2.46 (95% Confidence Interval 0.707 infinity; p = 0.16).
The odds ratio for a completed suicidal act on SSRIs compared to placebo is 2.22 (95% Confidence Interval 1.47 infinity; p < 0.001).
When this analysis is extended to include figures for nefazodone and Mirtazapine drawn from FDA medical reviews, the following outcomes are obtained:
The odds ratio of a suicide on antidepressants compared to placebo is 4.40 (95% Confidence Interval is 1.32 infinity; p = 0.0125).
The odds ratio for a suicidal act on antidepressants compared to placebo is 2.39 (95% Confidence Interval 1.655 infinity; p < 0.0001).
These figures are entirely in line with the figures cited by the MHRA for odds ratios for suicidal acts on Seroxat/Paxil in children of 1.5 3.2.
As noted above, there is a further factor that needs to be taken into account, namely that in so far as any of these antidepressants were suicide reducing in some adults in these trials, and there is considerable evidence for this, the true odds ratio for suicide induction is correspondingly higher than the figures cited here.
There is yet another factor to take into account, which is the role that any withdrawal syndrome might play in contributing to suicidality (see paper on SSRIs and dependence). Neither of these two possibilities appear to have been taken into account by anyone handling the problem to date.
Finally it can be noted that the figures for rates of suicide that stem from these trials are of the order of 168/100,000 patients on SSRIs compared with 64/100,000 patients on placebo. The significance of these figures will become clear when seen in the context of the epidemiology of suicide in primary care mood disorders.
Epidemiology traditionally involves the study of representative samples of the population, and requires a specification of the methods used to make the sample representative. A series of what have been termed epidemiological studies have been put on the table by companies and the MHRA in this debate.
The first is a one-column letter involving no suicides. The second is a selective retrospective post-marketing chart review, involving no suicides, which analyzed by the ACNP, the FDA and others show a 3-fold increased relative risk of emergent suicidality for Prozac versus other antidepressants .
A third was conducted by Warshaw and Keller on anxious patients, in which the only suicide occurred in a patient taking Prozac. Of the 654 patients in this study only 192 got Prozac. This, therefore, was not a study designed to test Prozacs capacity to induce suicidality.
A fourth study on 632 patients, conceived 20 years before Prozac was launched and instituted 10 years before launch, had only 182 patients who had got Prozac at any point. This was clearly not a study designed to establish whether Prozac might induce suicidality. Originally 1000 patients had been recruited, so that by the time the impact of Prozac could be considered 368 patients had dropped out many of whom can be expected to have dropped because of intolerance to antidepressants, but the authors of this study, who include Eli Lilly personnel and a number of noted consultants to pharmaceutical companies pay no heed to this important methodological point.
In general, under the definition of epidemiology offered above, none of these studies qualify as epidemiology. Two other studies are offered by companies and by the MHRA as evidence that SSRIs do not cause suicide induction the Jick study and a series of studies undertaken by the Drug Safety Research Unit in Southampton.
A set of post-marketing surveillance studies have been carried out in primary care in the United Kingdom by the Drug Safety Research Unit (DSRU) on all of the major SSRIs. These studies recorded 120 suicides in over 44,000 patients being treated in primary care in Britain. The DSRU methodology has since been applied to mirtazapine, where interestingly there have been 13 suicides reported from a population of 13,554 patients. The figures for SSRIs permits the comparisons outlined in Table 8.
Table 8: Drug Safety Research Unit Studies of SSRIs & Mirtazapine in Primary Care in the United Kingdom.
244 (C.I. 168 340)
173 (C.I. 110 255)
269 (C.I. 192 365)
183 (C.I. 114 274)
96 (C.I. 53 158)
There are two points to note about these studies. First, they do not prove there is not a problem with SSRIs, as SSRI companies claim. What they indicate is that SSRIs do not differ one from the other although there may be differences from Mirtazapine. The critical issue in assessing these results is to ask what would have the suicide rate been for these patients if left untreated.
A second point to draw from these studies is that the suicide rate in UK primary care on these antidepressants is of the order of 200/100,000 patients.
The Jick study was also undertaken in British primary care. This investigated the link between antidepressant prescriptions in 143 suicides from over 200,000 patients. It produced a statistically significant doubling of the relative risk of suicide on Prozac compared with the reference antidepressant, dothiepin, when calculated in terms of patient exposure years. Controlling for confounding factors such as age, sex and previous suicide attempts, left the relative risk at 2.1 times greater for Prozac compared to dothiepin and greater than for any other antidepressant studied, although statistical significance was lost in the process. Controlling for confounding factors in the case of mianserin and trazodone, however, led to a reduction in the relative risk of these agents compared to dothiepin in a way that did not happen for Prozac.
To provide comparability with other figures, I have recalculated the data from Jick in terms of absolute numbers and have separated the figures for Prozac from other figures (Table 9).
Suicides on Antidepressants in Primary Care in the United Kingdom:
Absolute Suicide Numbers
70 (C.I. 53 91)
26 (C.I. 8 61)
60 (C.I. 41 84)
80 (C.I. 38 144)
47 (C.I. 20 90)
69 (C.I 17 180)
78 (C.I. 43 129)
99 (C.I. 31 230)
166 (C.I. 86 285)
52 Suicides in 74,340 Pts
4 Suicides in 15,177 Pts
29 Suicides in 48,580 Pts
9 Suicides in 11,239 Pts
7 Suicides in 15,009 Pts
3 Suicides in 4,329 Pts
13 Suicides in 16,599 Pts
4 Suicides in 4,049 Pts
11 Suicides in 6,609 Pts
11 Suicides in 11,860 Pts
Total excluding Fluoxetine 132 Suicides per 195,931 Patients
67 Suicides per 100,000 Patients
The figures in the Jick study allow comparisons between antidepressants. They shed no light on the comparison between treatment with antidepressants and non-treatment or on the efficacy of antidepressants in reducing suicide risk in primary care. The Jick study also provides a good estimate of the rate at which suicides can be expected in UK primary care in the course of treating mood disorders.
The traditional figures with which the DSRU studies and the Jick study are compared by pharmaceutical companies are a 15% lifetime risk for suicide for affective disorders. This yields a suicide rate of 400-600/100,000 patients. Compared with this the figures for SSRIs from the DSRU studies or the figure of 189/100,000 patient years from Jick for Prozac could be seen as simply indicating SSRIs are not as good as other antidepressants at lowering suicide rates.
However, the 15% lifetime risk of suicide figure was derived from hospitalized samples of melancholic depressives in the pre-antidepressant era. There are very few empirical figures available for suicide rates in primary care depression, the sample from which the Jick and DSRU figures come.
One set of figures stems from Sweden, which gives a suicide rate of zero per 100,000 patients in non-hospitalized depression, and the traditional 400-600 suicides per 100,000 patients for hospitalized depression. This study is repeatedly cited by Lilly as indicating depressed patients are at 70-80 times the risk of suicide as the normal people.
One of the few other studies from primary care figure in Holland gives a suicide rate of 33 per 100,000 patient years.
Finally Simon and VonKorff from Puget Sound, based on a study of 65,000 patient years, and 36 suicides, give figures for patients with any secondary mental health service contact as 64/100,000 patient years. Primary care depression treated with antidepressants had a suicide rate of 43/100,000 patient years while primary care depressions not treated with antidepressants had a suicide rate of 0/100,000 patients. This latter group of patients being treated with psychotherapy may have been somewhat more mildly ill. But the existence of a suicide rate this low nevertheless proves that the suicide risk in many of the patients getting SSRIs is effectively indistinguishable from that of the normal population.
Utilizing a database of 2.5 million person years and 212 suicides from North Staffordshire, Boardman and Healy have modeled the rate for suicide in treated or untreated UK depressives and find it to be of the order of 68/100,000 patient years for all affective disorders. The figure of 68/100,000 gives an upper limit on the figure of suicides in mood disorders that are compatible with observed national rates of suicide in the United Kingdom. The Boardman and Healy study gives a figure of 27/100,000 patients per annum for primary care primary affective disorders.
From these various studies the following estimates can be offered:
Primary care depression suicide rates are < 68/100,000 patients
Primary care SSRI suicide rates are circa are > 180/100,000 patients
These figures overlap with the RCT data, which give figures of:
RCT suicide rates on placebo = 64/100,000 patients
RCT suicide rates on SSRIs = 168/100,000 patients
Two other epidemiological studies are of note. First, a study of 222 suicides by Donovan et al reported on 41 suicides that had had an antidepressant in the month before their suicide; this study demonstrated a statistically significant doubling of the relative risk of suicide on SSRIs compared to tricyclic antidepressants.
In a second study of 2,776 acts of deliberate self-harm, Donovan et al demonstrated a doubling of the risk for deliberate self-harm on SSRIs compared with other antidepressants, with in fact up to a 5 times greater risk on Prozac compared to imipramine for instance.
The response to this from companies and their consultants has been that this was not a randomised study and that prescribers may have preferentially given SSRIs to patients at risk of self-harm.
An intriguing set of studies was undertaken by Stuart Montgomery that sheds light on just this issue. In the first, patients with recurrent brief depressions, a group of high-frequency suicide attempters were randomised to either Prozac or placebo. This study was suggested in 1990 by a consultant to Lilly, presumably Montgomery, who was also likely to have views that would be of significance to the regulators.
The study was already underway at this point but remained unpublished until 1994, when publication coincided with the Wesbecker trial in the USA. The title and contents of the publication appear to reveal no problem with Prozac compared to placebo, aside from Prozacs lack of efficacy in this patient group and what might be inferred from the fact that the study recruited less than 2/3rds of its target sample, and approximately 50% of recruited subjects dropped out. Without details of the reasons for drop-out it is impossible to be certain what happened, but an internal Lilly memo makes it clear that placebo outperformed Prozac at a 0.006 level of significance on some central measure in the study.
Despite this result, it appears that Dr Montgomery persuaded SmithKline Beecham to fund a comparable study of Seroxat/Paxil in a similar patient group. This study was aborted after approximately 35 patients had been recruited, when the projected annualised rate suicide attempts in the Seroxat/Paxil group was running at 45 compared with 11 for placebo. These results were not published.
These are a similar patient group to those recruited to the Donovan study. In this case randomisation and placebo control makes it clear that far from the Donovan results being explainable away on the basis of biased prescription of SSRIs, the results seem likely to reflect the effects of SSRIs in this patient group.
Healthy Volunteer Studies
From the introduction of the antidepressants, there was recognition that these drugs regularly had serious adverse mental effects in some individuals. This was in contrast to, for example, the benzodiazepine tranquilizers, which seemed less likely to have comparably odd effects in either healthy individuals or patients.
The testing of drugs later marketed as antidepressants begins with healthy volunteer or phase 1 studies. An understanding had emerged by 1980 that odd reactions to antidepressants were common in healthy volunteer studies and that, in general, groups of healthy volunteers may be left worse off after antidepressants. This deterioration in functioning stood in apparent contrast to the improvement brought about by antidepressants in depressed patients. The apparent differences in responses between volunteers and depressives were explained by some on the basis of abnormalities in the brains of people who were depressed leading to a qualitatively different response to these drugs.
The argument that there might be a difference between healthy volunteers and depressed patients had some plausibility in the 1960s and 1970s when patients who were depressed were liable to have a condition called melancholia. These patients were commonly admitted to hospital for treatment where they were kept under close observation. These patients were also likely to be Electroconvulsive Therapy (ECT) responsive. These were patients who looked very different to normal volunteers.
In the 1970s, the SSRIs came on stream and even more than with previous antidepressants the use of these drugs in healthy volunteers, especially in multiple dose studies has led consistently to agitation and turmoil in these volunteers. These problems have shown a clear dose response curve.
In the case of Prozac, Lilly have indicated that only 12 out of 53 healthy volunteer studies have been reported.
As I understand it Lilly may have used an unusual healthy volunteer population compared for instance to Glaxo SmithKline who often used their own employees. It is extremely unlikely that Lilly had professionals trained to assess alterations in mental state monitor these healthy volunteer studies.
In the case of Seroxat/Paxil pre-launch data from 14 healthy volunteer studies out of a total of approximately 35 studies have appeared. In the case of published work, the data reported in these company studies commonly excludes material concerning behavioral toxicity, including the suicide of volunteers.
Many of SmithKlines healthy Volunteer studies with Seroxat/Paxil were done with employees of the company from the Harlow Pharmacology Group. Other studies were done in the West Middlesex Hospital at Isleworth, then a small hospital specializing mainly in GUT medicine, now closed down. The interest of the main investigator was gut physiology. Beecham at this point was a company whose best-selling drugs were active on the gut, and whose development pipeline was steered toward further drugs active in this area. De facto, their studies on Seroxat/Paxil reveal a company more interested in the effects of this drug on the gut than on any other organ system. This is not surprising in that it was uncertain at the time whether Seroxat/Paxil would be developed as an antidepressant in view of a relatively unfavorable clinical trial profile.
There do not appear to have been any tolerance studies dedicated to establishing the psychotropic effects of the new compound. The studies in the series are largely standard studies investigating possible interactions between Seroxat/Paxil and a range of other compounds with which it might be co-administered.
In addition to having been conducted in settings unlikely to detect psychotropic effects, many studies were single dose and as such relatively uninformative.
I have reviewed most of the studies undertaken by SmithKline prior to initial market authorization for Paxil/Seroxat. Methodologically, I have assumed that all placebo periods occurring after Seroxat/Paxil intake cannot be interpreted in the same way as other placebo periods. To do so would minimize the severity and frequency of the problems apparently caused by Seroxat/Paxil, as reports on the various studies by SmithKline Beecham personnel provide a strong justification for not considering the discontinuation placebo period as equivalent to other placebo periods SmithKline Beecham monitors note the common occurrence of certain features in the withdrawal period and as of the mid-1980s they had set up studies to map the events occurring during the withdrawal period. These included violent dreams and agitation.
In my assessment, GSKs multiple dose tolerance studies show high dropout rates up to 50% in one study and approx 13% otherwise. From studies where the design permits an assessment, withdrawal syndromes appear to run at roughly 50%, but are found in up to 85% in one study. Mood change appears to occur at a 25% rate, with agitation occurring at a 33% rate and sexual dysfunction where recorded at approx 40%. Extrapyramidal features such as tremor, and mouth/throat dyskinesias or dystonias occur in roughly 30%. In one of these studies one volunteer committed suicide following the study. It cannot be assumed that this suicide did not relate to intake of Paxil.
SmithKline Beechams single or repeated single dose studies also show a significant adverse event profile, with studies being abandoned after a single dose and volunteer drop-outs after a single dose or a greater rate of drop-outs on Seroxat/Paxil than for instance on amitriptyline. The profile of side effects appears similar to that in the multiple dose studies nausea, insomnia, lethargy or malaise and a range of extrapyramidal symptoms such as agitation.
In an effort to distance themselves from these findings GSK have claimed that these events do not happen at a statistically significantly greater rate on Seroxat than on placebo. These studies however were not powered to make such determinations. There was commonly only 3-4 subjects given placebo; against this background the occurrence of any side effects on placebo makes it possible for the company to claim that there was no greater frequency of problems on drug even though the side effects happening on the drugs are now recognized as signature side effects of SSRIs.
The findings from GSK healthy volunteer studies furthermore map closely onto the findings from phase 1 SSRI healthy volunteer studies using Lustral/Zoloft.
In the case of Zoloft/Lustral, as few as seven of approximately 35 pre-launch studies appear to have been published.
In the case of Lustral/Zoloft, I have reviewed a majority of the protocols undertaken prior to market authorization.
In the multiple dose studies, a clear preponderance of these studies have shown significant levels of agitation, turmoil or emotional dysregulation, sometimes after a single dose only. In studies where different doses were used, there has been a clear dose-dependent relationship. It would seem that in all cases the problems cleared up following the discontinuation of treatment. In some protocols, there is evidence of a dose-dependent relationship between Zoloft and agitation and extrapyramidal problems. In these protocols anywhere from 50 to 80% of subjects are affected with clear extrapyramidal problems. In these cases, problems appeared shortly after the institution of treatment and cleared with discontinuation. Pfizer monitors at the time furthermore clearly attribute problems of apprehension and agitation to serotonin reuptake inhibition. Similar attributions have been made since in print by Pfizer employees.
One of these studies conducted by Ian Hindmarch in 1982, which remains confidential produced the following outcome: "[O]f 12.. healthy volunteers entered into this study, 5 in the first week of the study were randomized to Sertraline, and 7 to placebo. And of the 5 randomized to Sertraline, all dropped out in the course of the first week for what appears to have been fairly severe anxiety or agitation ". This striking finding cannot be explained easily other than by sertraline induced agitation amounting to suicidality. Such an interpretation is consistent with a study of sertraline undertaken in this department see below, and with linkages made by Pfizer personnel to drug induced agitation.
In single dose studies, utilizing different doses or preparations of Zoloft there was a clear challenge-dechallenge-rechallenge response visible. Among the single dose studies, in one published study the authors concluded there was a significant deterioration in wellbeing between the 4th and 8th hours after Zoloft, with complaints of difficulties lasting up to 48 hours after the last dose. The authors note that "both the 200 and 400 mg doses could be considered poorly tolerated by volunteers". A similar pattern was seen in other protocols.
In general across clinical trials with SSRIs in both depressed and non-depressed populations, the results show a consistent pattern of drug-induced conditions variously described as agitation-anxiety-nervousness-hyperkinesis-tremor. These show a regular onset after treatment starts and a dose-response relationship, with the problem clearing on discontinuation.
Very few of the healthy volunteer studies utilizing SSRIs have been published. (One of the relatively few published studies using Lustral/Zoloft is by Saletu et al). In the studies that have been conducted, both published and unpublished, the problems associated with SSRIs have been coded for under a number of possible headings, as mentioned above, including agitation, nervousness, anxiety, tremor, restlessness and hyperkinesis. These overlapping terms, a number of which were applied to the same individuals in some instances, indicate the presence of a drug-induced state that was relatively novel at least in its intensity for the clinicians conducting these studies. Despite the clear emergence of this syndrome in company studies in the early 1980s, and the relative ease with which it could be investigated in healthy volunteers, this condition, which has clear implications for the wellbeing of any depressed subjects later put on SSRIs, remains uninvestigated to this day. The attention of the wider academic community was effectively diverted from the problem as either the studies or these data from the studies remained for the most part unpublished.
There is close consistency between these findings and findings from SmithKline Beechams single or repeated dose studies, which show a similar adverse event profile, with studies being abandoned after a single dose and volunteer drop-outs after a single dose. The profile of side effects appears similar in the single dose studies as in the multiple dose studies, with reports of high rates of nausea, insomnia, lethargy or malaise and a range of extrapyramidal symptoms such as agitation see below.
North Wales Study
In most of these company healthy volunteer studies, the protocols were designed to preclude detailed questioning of the volunteers. This ban on exploratory questioning makes a healthy volunteer study conducted in North Wales of some importance. This was a randomized double blind controlled study comparing reboxetine, a drug with no actions on the serotonin system, with Lustral/Zoloft, which found that Lustral/Zoloft induced significant suicidality in two of our healthy volunteers. The healthy volunteer group was comprised mainly of medical and nursing staff. A detailed description of the suicidality induced has been published in a peer-reviewed journal and presented at Royal College of Psychiatrists, BAP and ECNP Annual Meetings in the course of 2000. The remaining data from the study have also been published in a peer-reviewed journal.
The North Wales study demonstrated a similar profile of neuropsychiatric and other side effects, including obscure side effects such as dystonia or "pharyngitis", in a similar ratio to company sponsored healthy volunteer studies. In general both Pfizer and SmithKlines studies and our study found a 40+% rate on average of subjects poorly tolerating SSRIs. We, however, were freer to explore further the consequences of some of these side effects. Our study makes it clear that some subjects in company sponsored studies who became agitated will almost certainly have become suicidal and perhaps even homicidal and indeed that the one in ten rate of suicide induction in our study was possibly low compared to some of these other studies. The mental states of these volunteers do not appear to have been explored in company studies.
Despite conditions unfavorable to a proper exploration of the issues, studies of Lustral/Zoloft such as that conducted by Hindmarch for instance have produced clear statements of apprehension from the volunteers and assessments by them directly corroborating the findings from our healthy volunteer study.
Studies have also been carried out by the SSRI producing companies of their compounds in patients who are not depressed, on for example obesity or smoking cessation programs. Many of these studies have not been published. Companies are under no onus to publish either studies that have been conducted or inconvenient details of results within those studies, which are otherwise published.
Significance of Healthy Volunteer Studies
The significance of healthy volunteer studies and studies in non-depressed populations in this case is as follows. In the 1960s, 1970s and 1980s cases of depression as mentioned above were more likely to be cases of what would now be called melancholia. Cases of nervousness were interpreted as anxiety states and it was thought appropriate to treat such patients with benzodiazepine, minor tranquilizers. One way to see this is that patients walking in through the clinic doors with nervous conditions were seen by the physicians as cases of Valium to whom a corresponding diagnosis of anxiety would be given.
In the 1990s following concerns about dependence on benzodiazepines, under the influence of the aggressive marketing of the SSRIs these cases of Valium became cases of Prozac in the first instance and subsequently cases of Prozac, Zoloft and Paxil. Through educational and marketing campaigns, the SSRI companies have produced a situation where individuals, who would never have been given an antidepressant in the 1960s, 1970s and 1980s, are now given one after cursory questioning by a physician.
These are often anxious patients, with no hint of suicidality, who may be at most risk of the reactions experienced by healthy volunteers. These patients are then left unwarned and unmonitored during the critical first few weeks of treatment. Companies transfer the onus to warn to the treating physicians, while the studies that might alert these physicians to serious problems with the drugs remain unpublished. Where in the 1960s depressed patients looked very to different healthy volunteers, these new patients much more closely resemble healthy volunteers, albeit stressed volunteers, than they do melancholic depressives.
When it comes to these cases of depression in the community, companies cannot avoid the fact that they have the studies demonstrating that dangerous reactions not only happen but are likely to happen, that antidepressants adversely impact on the affective well-being of a significant number of subjects and that on assessment these adverse effects have been squarely attributed by company employees to serotonin reuptake inhibition.
The SSRI group of drugs was always destined for community patients. These drugs were singularly ineffective in hospital depression. None of the companies in their clinical trials submitted to the FDA for approval were able to demonstrate efficacy in hospital depression. The treatment effects that were demonstrated were demonstrated in outpatient conditions, the kind of nervousness that in the 1970s and 1980s led to treatment with a minor tranquilizer. This is a key point behind the relevance of healthy volunteer studies to the question of suicidality on SSRIs.
A second point of significance regarding the healthy volunteer studies is that they undercut the disease not the drug argument, which companies have used to defend themselves. Patients taking Valium were never thought to be at risk of suicide. Valium and other benzodiazepines were never promoted on the basis that they would minimize risks of suicide.
The current SPc and PIL ambiguously refer to suicide risk increasing in the early phase of treatment. These statements are made against of background where the majority treatment of depressive disorders is in patients with milder conditions, likely to benefit little from treatment, but more importantly in a patient group treated in the community unsupervised and unwarned.
Again and again the story from patients recently demonstrated in programmes like Panorama is that when they encounter problems during the early phase of treatment and consult their GP they are likely to be told that their dysphoria stems from their illness and they are commonly encouraged not just to adhere to treatment, but often to increase the dose. This response can be justified by GPs on the basis of an appeal to the current SPc and senior figures linked to the regulatory system and others have portrayed the current UK wording as meaning just this.
Patients who get worse in the course of treatment often do not make the connection to treatment, and accordingly they do not gainsay their GP. This situation is even worse in children, whose complaints have to be mediated through parents who trust their GP and are not actually experiencing the problem.
Aside from the need to recognise treatment induced suicidality in order to save lives, if emergent suicide linked to drug treatment is not correctly attributed to the treatment, in my experience patients suffer a long-lasting injury to their self-esteem and self-confidence as a consequence.
If patients have engaged in actual suicidal acts as a result of treatment and the connection to treatment is not made, given that prior suicide attempts appear to increase the risk of future successful suicides, it appears possible that the risk of a future successful suicide has been increased accordingly.
1/ Antidepressant drugs can improve your mood, make you feel calmer and reduce your risk of suicide. But in a proportion of cases, particularly in the first weeks of treatment, the drug you are now taking may cause you to feel worse. You may become more tense or agitated, and you may have disturbing or unusual thoughts, such as thoughts of self-harm, you have never had before. Should this happen consult your doctor immediately, as dose lowering, or changing to another antidepressant may be more appropriate for you.
2/ Physicians should be aware that in rare instances SSRI compounds such as Seroxat may produce acute homicidal and suicidal states. Close monitoring of patients is indicated in the course of the first six weeks of prescription of these drugs, especially when there is a history of unusual anxiety, hypomania or akathisia.
This warning was drafted by Professor JT Maltsberger at the request of GSK's lawyers and tacitly endorsed by the jury who returned a verdict against GSK in Tobin v SmithKline.
Aside from akathisia, there are two other mechanisms by which SSRIs may lead to or contribute to suicide emotional blunting/lability and psychotic decompensation.
The evidence that SSRIs cause emotional blunting lies in the fact that these drugs are used to treat a wide variety of anxiety states and that Seroxat for instance advertises itself the anxiolytic antidepressant. An anxiolytic effect is by definition an instance of emotional blunting. The term blunting is applied when the degree of this effect gets to the extent that an individual perceives it to be excessive. This is very clearly described recently in the Guardian.
The very many clinical trials that all SSRI companies have done on their drugs in anxiety states attest to the emotional blunting capacities of the SSRIs. This action of SSRIs is therefore in fact abundantly supported by randomized placebo-controlled trial evidence. This clinical trial evidence is supplemented by a growing body of case studies, which make it clear that the emotional blunting SSRIs produce, the fear reduction, can proceed too far and become an abnormal absence of fear that has consequences for behavior. In addition to the above in their phase 1 healthy volunteer studies, company monitors have regularly recorded the occurrence of mood change on Seroxat/Paxil and other SSRIs, and coded this to emotional lability.
Discontinuing treatment will often rapidly lead to a restoration to normal, and subjects stopping SSRIs comment on this. But in some patients while emotional blunting may clear up, lability may increase.
The significance of these affective changes is that such an effect can be expected to make an individual less sensitive to the consequences of their actions than they would be in the normal course of events making it possible to act without fear of the consequences, or not to be inhibited by any moral consideration of the consequences of an action. This will amount to a de facto disinhibition.
There is likely to be and in clinical practice there appears to be significant interactions between this disinhibition and the disinhibition linked to alcohol intake. Each of the major companies has up to 100 reports of homicides or attempted homicides and hundreds of reports of serious assaults on their books. Where there has been concomitant alcohol intake, this is used to explain away the event. In the case of up to a third of SSRI associated homicides or assault, alcohol use may have been also reported. But this explanation does not take into account the possibility that while alcohol may be contributory, clinical experience indicates that the combination of SSRIs and alcohol may be particularly disinhibiting for some people and that this linkage between alcohol and SSRI intake with assault or homicide points to an association that should lead due warnings rather than an association that should be used to explain away problems with the SSRI.
Since the first administration of imipramine to patients, it was also noted that patients at risk of psychotic decompensation became worse on this drug. This has been a regular feature of the testing of SSRIs, with for example in the case of Prozac, numerous early reports from hospital studies of patients with schizoaffective type disorders becoming markedly worse on this drug at what was probably a greater rate than for other drugs or patients.
Having reviewed trials from the clinical trial databases of Prozac, Zoloft/Lustral and Seroxat/Paxil, I can state that at present all SSRIs that I have reviewed have caused psychotic decompensation in some patients. This happens at a higher rate with SSRIs than occurs on placebo. This data has not been published. This problem clears up on discontinuation of the SSRI.
A recently published study from Yale suggests that up to 8% of patients admitted to psychiatric facilities may suffer from SSRI induced mania or psychosis.
A link between psychotic decompensation and suicide is part of standard psychiatric knowledge and finds a place in all textbooks. This cannot be denied by any of the companies. The only argument can be on the issue of whether a particular SSRI produces this effect or not. Evidence presented in Tobin v SmithKline makes it clear that Seroxat/Paxil for instance can do just this, with even Mr Preuss, the attorney for SmithKline in his closing address arguing that it would be a mistake to think that hallucinations for instance did not happen on Paxil.
Table 10: Incidence of Suicides and Suicide Attempts in Antidepressant Trials From FDA Medical Reviews
Suicides & Attempts as a % of Patient No
All Investigational drugs
SSRI Trial Placebo
|David Healy MD FRCPsych|
|Director, North Wales Department of Psychological Medicine (Honorary Consultant Psychiatrist)|
|Uned Hergest, Ysbyty Gwynedd, Bangor, Gwynedd LL27
Hergest Unit, Gwynedd Hospital, Bangor, Gwynedd LL27 2PW
Tel: (01248) 384452 Fax: (01248) 311397
CLICK HERE for index of David Healy's correspondence