David Healy’s interesting article in Advances in Psychiatric Treatment entitled “The Antidepressant Tale: figures signifying nothing?” deserves some attention. There are two main points, one relating to the safety of antidepressants and another relating to efficacy.
Let’s start with safety. Suicide is a rare event and, as such, is difficult to predict with much statistical certainty. Thus, antidepressant trials lack the statistical power to show differences in suicidal acts between antidepressant and placebo. While clinical trial data have rather consistently shown an increased relative risk for suicide and suicidality on antidepressants versus placebo, the absolute risk remains low, thus resulting in differences that are not statistically significant.
So do we say, “oh well, it’s not less than 5% likely than the differences between antidepressant and placebo are due to chance alone, so prescribe all the drugs you want?” Apparently many think so. Do we need to be 95% sure that a drug is related to significantly more suicides before we do something about it?
Healy starts off discussing an analysis conducted by Lilly which showed the relative risk for suicidal acts on fluoxetine was nearly twice that of such acts on placebo but that the difference was not statistically significant (due to the general rarity of such acts overall):
“The analysis gave the relative risk of suicidal acts for patients taking fluoxetine
compared with placebo as 1.9 (95% CI 0.2–16.0). This led Beasley et al to state ‘Analysis of the incidence of suicidal acts (suicidal attempts and completions) revealed no statistically significant differences in the act rates between fluoxetine-treated and placebo treated patients’. And the conclusion they drew from this lack of significance was that ‘Data from these trials do not show that fluoxetine is associated with an increased risk of suicidal acts or emergence of substantial suicidal thoughts among depressed patients’ (Beasley et al, 1991)…
Using this analysis, if we return to the Beasley et al (1991) paper it is clear, that although the statement ‘the analysis of the incidence of suicidal acts revealed no statistically significant differences in the act rates between fluoxetine treated and placebo treated
patients’ is supportable, it should not lead to the conclusion that ‘data from these trials do not show that fluoxetine is associated with an increased risk of suicidal acts or emergence of substantial suicidal thoughts among depressed patients’.”
What Healy fails to mention here is some of his own work, which I found disappointing. For example, in the Beasley meta-analysis, it was later found out that “no mention was made that benzodiazepines had been coprescribed in the clinical trial program in order to minimize the agitation that Lilly had recognized fluoxetine could cause[32] … no reference was made to the 5% of patients who dropped out for anxiety and agitation. This drop-out rate, which is statistically significantly greater than for placebo, holds true for other SSRIs as well.” (Healy 2003 in Psychotherapy & Psychosomatics). So with coadministration of benzodiazepines to control agitation and removing the people who dropped for agitation and anxiety (perhaps the most likely to be suicidal), the results are not statistically significant but certainly point much more toward suicidality on fluoxetine than placebo.
Later, Khan et al (2003 in American Journal of Psychiatry) examined the rates of suicides in SSRIs, active comparators, and placebo and found a relative risk of 1.4 (95% CI .56-3.62) in favor of greater suicide on SSRIs and, when including all antidpressants found a RR of 1.62 (95% CI .66-4.02). Let’s keep in mind that the above issues with the coadministration of benzo’s applied in many of these studies, as well as some studies (though it’s unclear how many) counting suicides on placebo washout (when all participants were receiving placebo – nobody was taking antidepressants) as placebo suicides, which biases the data in favor of drugs appearing safe. See links on inappropriately attributing suicide/suicidality to placebo here and here. So even with the deck stacked in favor of showing drug safety, there is still an enhanced suicide/suicidality risk associated with antidepressants compared to placebo. Curious, eh? Perhaps if the deck was not stacked, this difference would have actually achieved the magic p-value of .05 or less.
But wait, there’s more. Next Healy discusses another Khan et al meta-analysis (2002 in Journal of Affective Disorders). Here’s what Healy said in his most recent article:
“One of the most striking instances of the unwillingness to think that signals might offer evidence of a hazard posed by a therapy came from Khan et al in a 2002 review of deaths by suicide of adults participating in clinical trials of fluoxetine, paroxetine, sertraline,
clomipramine, fluvoxamine, clonazepam and venlafaxine for obsessive–compulsive disorder, posttraumatic stress disorder, social phobia, generalized anxiety disorder or panic disorder. From this dataset, they concluded: ‘We found that suicide risk among patients with anxiety disorders is higher than in the general population by a factor of 10 or more. Such a finding was unexpected... The sample of patients selected was considered at minimal risk of suicide’ (Khan et al, 2002). Anxiety disorders, they suggested, posed a risk of suicide. In fact, this dataset on 12 914 patients taking active treatment and 3875 patients taking placebo reported 11 suicides in the active treatment groups and no suicides in the placebo groups. The data on suicidal acts were incomplete but combined data for suicidal acts and suicides show a relative risk of 1.65
for active treatment over placebo. It takes a prior judgement that antidepressants could not trigger suicide to interpret such data as evidence for the suicide risk posed by anxiety disorders rather than by the antidepressants used to treat them (p.322).”
But wait, there’s still more. According to the most comprehensive meta-analysis on the subject, the relative risk of suicidality on SSRIs is 2.93 versus placebo (Fergusson et al., 2005 in BMJ). And, to quote more from Healy:
“In December 2004, the Medicines and Healthcare products Regulatory Agency (MHRA) published a report on antidepressants and suicide (Committee on Safety of Medicines Expert Group on the Safety of Selective Serotonin Reuptake Inhibitors, 2004). This
gave data for suicides in adult placebo-controlled trials of sertraline, citalopram, escitalopram, fluvoxamine, venlafaxine and mirtazapine from which the relative risk of completed suicide on active treatments compared with placebo emerged as 2.42. The relative risk for suicidal acts on active treatment was 2.37 compared with placebo and the relative risk for the combination of both suicides and suicidal acts compared with placebo was 2.38. These values of 2.37 and 2.38 are statistically significant but have not led to clear warnings.”
So depending on the meta-analysis that is examined, there is sometimes statistical significance, but there is always a strong trend for a relative risk of about two or greater for suicidal acts or actual suicides on drug compared to placebo. This occurs despite some studies inappropriately assigning suicidality or suicide to placebo when it occurred during a placebo washout period, when the placebo was not being compared to the medication, and that benzodiazepines were coadministered in many of the original studies.
Healy goes on to point out that we may in fact be entirely too enamored with conventional significant testing in many cases. In the case of safety, I agree with him. Differences in rare events may be difficult to detect statistically, but when the heavy bulk of the evidence falls clearly on one side of the equation indicative of iatrogenic effects of a treatment, it is essential that we pay attention to this evidence.
I’ll post separately on the other issue mentioned by Healy regarding efficacy since this post is already long enough.
No comments:
Post a Comment