Association between suicide attempts and selective serotonin reuptake inhibitors: systematic review of randomised controlled trials
BMJ 2005; 330 doi: https://doi.org/10.1136/bmj.330.7488.396 (Published 17 February 2005) Cite this as: BMJ 2005;330:396
All rapid responses
Dear authors,
I am a psychiatrist practising in University of Malaya, Malaysia. I read
your systematic review entitles association between suicide attempts and
SSRIs. It is a very interesting review for psychiatrist to read and
digest.
I am very greatful if you could clarify to me how could you reach the
conclusion of "the odd ratio of non-fatal suicide attempts was 0.85 and
the odd ratio of fatal suicide attempts for SSRIs compared with tricyclic
antidepressants(TCAs) was 7.27". This indeed a very high value and
interprets a very high risk of fatal suicide attempts in patients treating
with SSRIs as compared to TCAs. It means significant finding in term of my
practice. Therefore, your explanation is very much helpful. Thank you.
Competing interests:
None declared
Competing interests: No competing interests
Editor,
Perhaps naively, I thought that Fergusson and colleagues might appreciate me highlighting that either their paper contained erroneous data or there had been a miscalculation. Although they have been kind enough to respond to two independent comments, and acknowledge an “error in the odds ratio estimate for fatal attempts in the SSRI versus TCA analysis” I am surprised that that authors have not taken the opportunity to correct this error in their e-response. Specifically, is the error in the raw data (table 1) or in the calculation? In either case what are the correct figures?
If I understand the authors correctly, there is in fact no increased risk of fatal outcomes with SSRIs in this dataset and they wish to correct the “What this study adds” box that “Evidence from this study supports the association between the use of SSRIs and increased risk of fatal and non-fatal suicide attempts” (my highlight)
Whilst accepting the point that weighting should be taken into account, I believe that the substantial differences in the effect sizes I pointed out in my original response require further clarification. I hope the authors will not take offence if I respectfully suggest that the statisticians employed by the BMJ could help by looking at the primary dataset; after all, isn’t that their primary role?
Competing interests:
None declared
Competing interests: No competing interests
Editor:
We thank Fergusson (Rapid Responses, March 4) for replying to our
letter and agree with him [1] that metaanalyses require weighted
calculation. However, the rough check we carried out (see our earlier
letter) using the aggregated data in his table yielded ORs and confidence
intervals very close to those published by him for the SSRI vs. placebo
series. If the pooled ORs in his metaanalysis were indeed significant
overall (which we cannot check without the original data) and yet the OR
was indistinguishable from 1 (Fergusson et al, Fig. 2) it would strongly
suggest extreme heterogeneity amongst the ORs, indicating that there are
other factors that are influencing the results and that perhaps pooling
was inappropriate.
Since the original paper did not carry the weighted data for the
individual studies, using the aggregated data in their table was the only
available method for us to evaluate the published results.
In the case of the SSRI vs. TCA analysis, in the paper an OR of 7.27
was based on 5 deaths in a SSRI sample of 6126 vs. 4 deaths in a TCA
sample of 5401. Clearly, such an OR looked totally disproportionate to the
clinical findings and thus potentially misleading to physicians and
especially to the general public. In fact we know that it has been seized
upon and headlined by newsmedia as yet another cause for public anxiety in
the current furor over the risk of provoking suicidal events through the
use of SSRIs. Publishing such a questionable result, without commenting
on the very wide confidence intervals and small numbers, could serve as a
prime example of the misuse of statistics, and it risks the possibility of
disservice to the cause of evidence-based medicine.
[1] Streiner D. Using metaanalyses in psychiatric research. Canadian
Journal of Psychiatry, 1991; 36: 357-62.
Competing interests:
None declared
Competing interests: No competing interests
We thank Dr. Marcia Valenstein from the University of Michigan for
quickly and kindly alerting us to a possible error in the odds ratio
estimate for fatal attempts in the SSRI versus TCA analysis. Indeed, this
was an undetected error in manuscript preparation and we let BMJ know
immediately. We regret and take full responsibility for the error.
Fortunately, it does not change our primary conclusions.
As for our analyses being “miscalculated dramatically” or “blatantly
misleading” as suggested, respectively, by Mitchell and Sakinofsky and
Streiner in their BMJ rapid responses, we strongly disagree. Ignoring
individual trials and calculating summary statistics from lumped data, as
they have done, is an incorrect approach to meta-analysis.
Nowhere in our Methods section do we state that we calculated odds
ratios from the aggregated data. Instead, we conducted meta-analyses of
randomized controlled trials using Peto’s method for calculating odds
ratios of rare event data. Failing to preserve the randomization of
subjects at the trial level introduces bias and confounding.(1)
Furthermore, simply calculating odds ratios from aggregated data fails to
account for the influence of chance in small trials. By weighting each
study by its precision, smaller trials have less influence than larger
trials.
(1) Egger M, Davey Smith G, Altman D. Systematic reviews in health
care. Meta-analysis in context. London: BMJ Publishing, 2001.
Competing interests:
None declared
Competing interests: No competing interests
I wonder how many of these subjects were Hypogonadal, either in the
first place, or as a result of taking SSRI's.
The American Association of Clinical Endocrinologists alerts us to
the prevalence of Hypogonadism misdiagnosis in its adult male Hypogonadism
guidelines. The patient is normally placed on anti-depressants instead.
SSRI drugs have been found to cause low Free Testosterone as well as
sexual dysfunction. Low Free Testosterone is indicative of Hypogonadism.
Hypogonadism has symptoms of lethargy, mood swings,depression and
"brain fog", amongst many others, including for some, sexual dysfunction.
In severe, or long term untreated cases of Hypogonadism, suicidal
tendencies can occur.
Hypogonadism, contrary to popular thought perhaps, does significantly
exist in teens and increases in prevalence as time goes on.
All that being the case, why are we surprised that SSRI's, and other
drugs causing Hypogonadism, of which there are many, lead to depression,
loss of hope and ultimately, suicidal thoughts, especially in the young.
An already confused situation is exacerbated with the introduction of
SSRI's as they slowly demolish endocrine health.
Whilst the limitations of some studies referenced are obvious in
terms of numbers and perhaps testing methods, there must be a serious
question raised as to why these patients are not being properly evaluated
for Hypogonadism, before putting them onto anti-depressants.
References:
1. American Association of Clinical Endocrinologists Medical
guidelines for clinical practice for the evaluation and treatment of
hypogonadism in adult male patients—2002 update Petak SM, Nankin HR, Spark
RF, Swerdloff RS, Rodriguez-Rigau LJ AACE Hypogonadism Guidelines, Endocr
Pract. 2002;8(No. 6)
2.Crenshaw TL, Goldberg JP: Sexual Pharmacology: Drugs That Affect
Sexual Functioning. New York, NY: WW Norton & Company, 1996.
3. Antidepressant-Induced Sexual Dysfunction Associated with Low
Serum Free Testosterone
Presented by Alan J. Cohen, M.D.,
Private Practice and Assistant Clinic Professor of Psychiatry, UCSF
Psychiatry On-Line 1999
revised 10/3/2000
4. Seidman S and Rabkin J (1998) Testosterone replacement therapy for
hypogonadal men with SSRI-refractory depression. J Affect. Disord Mar;
48(2-3):157-161
5. Bioavailable Testosterone and Depressed Mood in Older Men: The
Rancho Bernardo Study
Elizabeth Barrett-Connor, Denise G. von Mühlen and Donna Kritz-Silverstein
Competing interests:
No competing Interest, however, author of the Testosterone Deficiency Centre at www.androids.org.uk
Competing interests: No competing interests
Editor:
The debate on whether antidepressants, especially SSRIs, are
responsible for provoking suicidal behaviour in patients has attracted
worldwide media reaction and generated fearfulness in many patients who
might benefit from these drugs. One of us (IS) directs a consultative
clinic that focuses on patients thought to be at high risk for suicide
and, like other colleagues, has experienced increasing patient resistance
to taking SSRIs (and indeed other antidepressants) even though these
patients are suicidally depressed and psychological treatments (including
interpersonal therapy or CBT) have failed to help them. The recent
article in your journal by Fergusson et al(1) caught our attention, as
well as the rapid response by Mitchell(2), which suggested that most of
the odds ratios given in this paper were incorrect. This prompted us to
recalculate the odds ratios for all the trials in the table in their
article. Unlike Mitchell, our calculations for the SSRIs versus placebo
studies showed odds ratios that were similar to those published by
Fergusson et al. However, we agree with Mitchell that the reported
calculations for SSRIs versus TCA's and for SSRIs versus active drugs
other than TCA's are incorrect. Fergusson et al cite an odds ratio for
completed suicides for SSRIs versus TCA's as 7.27 (95% confidence limits =
1.26 to 42.03), based on 5 deaths in an SSRI group of 6126 and 4 deaths in
a somewhat smaller TCA group of 5401. Even just glancing at the data on
which they are based should make the calculated odds ratio suspect, and
one is surprised that this result got past the authors themselves and even
more so, the reviewers. We recalculated the data ourselves as showing an
odds ratio of 1.10 (95% confidence interval = 0.30 to 4.11), which is not
significant. For fatal and non-fatal events for SSRIs versus active drugs
other than TCA's, Fergusson et al report a statistically significant odds
ratio of 1.94 (1.06 to 3.57) and, like Mitchell, we get an odds ratio of
1.54 (0.85 to 2.80), not significant. We think that, given the current
climate of anxiety among patients, to publish such blatantly misleading
analyses is unjustifiable and recommend that the authors publish an
erratum.
We suggest also that odds ratios alone (and all other indices based
on ratios) do not give an indication of absolute risk. For example,
doubling a mortality rate from 1 in 100 to 2 in 100 yields an OR of 2, but
so does doubling it from 1 in 1,000,000 to 2 in 1,000,000, even though the
responses to these two situations should be very different. It is also
necessary to examine the Number Needed to Harm (NNH), i.e. the reciprocal
of the absolute difference in the suicide rates(3). In the Fergusson et al
study, for all trials that report fatal and non-fatal events in patients
taking SSRIs vs. placebo (OR = 2.24), we calculate the NNH as 708, meaning
that this number of people must take SSRIs for there to be one excess
event (fatal or non-fatal) in the SSRI group. This actually compares quite
favourably with other examples in medicine, e.g. the NNH of 179 for fatal
or non-fatal events comparing clopidogrel versus aspirin for patients with
stroke in the CAPRIE trial(4).
Only 49% of the trials screened in this study (345 of 702 considered
eligible) reported the suicide attempts (143 in total) that occurred, and
were included in the analysis. Surely suicide attempts must have occurred
in the trials that did not report them. What guarantee do we have that
the trials that were included are representative of the total? In fact,
the authors tell us that the larger trials (those potentially more of
interest) tended to be the ones that did not report these events. Had the
comparative incidences in them been lower, the ORs reported by Fergusson
et al would have been lower too, and maybe not significant. We wonder if
the findings from the selected efficacy trials that were included can be
generalised to clinical experience at large (effectiveness).
The statistical excess of suicidal attempts reported among patients
taking SSRIs versus those on placebo has, of course, been shown before,
notably by Khan et al(5) and your editorial by Cipriani et al(6) provides
a reasonable explanation for this finding. We like also the suggestion by
Jones(7) that under reporting may be more common among placebo-treated
patients because overdoses among them would be non-toxic. One wonders
also about the randomisation process in some of the reported trials,
especially the smaller ones, and whether they were genuinely at
armslength. Clinicians are notoriously unhappy about assigning their
depressed patients to the placebo arm of a trial and are likely to subvert
the process if they can, so that their sicker cases might land up in the
active drug arm(8). In any case, the whole debate keeps losing track of
the fact that these trials were not designed to assess suicidality as an
outcome but to satisfy regulatory agencies that the drugs were efficacious
in controlling a specific disease process, hence their unreliability for
reporting of suicidal events. Suicidality (past suicidal behaviour or
present suicidal ideation) is in fact generally an exclusion criterion for
these trials. To lay these issues to rest we badly need prospective
studies with suicidality as the outcome variable of interest, but these
have very rarely been done for ethical and practical (sample size)
reasons(9). Meanwhile, the numerous, large-scale, population-based
observational studies in existence that show a fall in suicide rates
related to increased treatment of depressive illness together with
increased use of novel antidepressants including SSRIs should not be
ignored(10).
References:
[1] Fergusson D, Doucette S, Glass KC, Shapiro S, Healy D, Hebert P,
et al. Association between suicide attempts and selective serotonin
reuptake inhibitors: systematic review of randomised controlled trials.
BMJ 2005; 330: 396-9.
[2] Mitchell A. Risk of suicide using SSRIs: data appears to be incorrect.
BMJ 2005; 330: 396.
[3] Laupacis A. Sackett DL. Roberts RS. An assessment of clinically useful
measures of the consequences of treatment. New England Journal of
Medicine. 318(26):1728-33, 1988
[4] CAPRIE Steering Committee. (1996). A randomised, blinded trial of
clopidogrel versus aspirin in patients at risk of ischaemic events.
Lancet, 348, 1329-1339.
[5] Khan A, Warner HA, Brown WA. Symptom reduction and suicide risk
in patients treated with placebo in antidepressant clinical trials. An
analysis of the Food and Drug Administration database. Archives General
Psychiatry 2000; 57: 311-7.
[6] Cipriani A, Barbui C, Geddes J. Suicide, depression, and
antidepressants: patients and clinicians need to balance benefits and
harms. BMJ 2005; 330: 373-4.
[7] Jones H. SSRI use and non-fatal self harm .BMJ 2005; 330: 396.
[8] Schulz KF. Subverting randomisation in controlled trials. JAMA
1995; 274: 1456-8.
[9] Gunnell D, Saperia J, Ashby D. Selective serotonin reuptake
inhibitors (SSRIs) and suicide in adults: meta-analysis of drug company
data from placebo-controlled randomised controlled trials submitted to the
MHRA’s safety review. BMJ 2005; 330: 385-8.
[10] Licinio J, Wong M-L. Depression, antidepressants and
suicidality: a critical appraisal. Nature Reviews/Drug Discovery 2005; 4:
165-71.
Competing interests:
None declared
Competing interests: No competing interests
Demonstration of a causal link between Selective serotonin reuptake
inhibitors (SSRI) use and self-harm would have significant clinical
implications for health services. Indeed the extra monitoring might make
antidepressant use wholly impractical within a risk averse service.
Two findings stand out. First is the remarkably low reported
incidence of self-harm in these studies with estimates of 0.05 – 0.001 %.
Even during a short trial such a low incidence among patients with
significant depression is not credible. The surely unique finding that
rates of fatal and non-fatal self-harm are identical among placebo treated
patients further emphasises how unsustainable these reported figures must
be. It is perhaps unsurprising that in studies designed to evaluate self-
harm but efficacy there is gross underreporting of such acts.
It is likely that underreporting may be more common among placebo
treated patients. Although the methods of self-harm are not stated one
might speculate that overdose of trial tablets accounts for a number of
these reports. The trial treatments are unequal since SSRI’s will cause
physical effects (GI disturbance) that might prompt emergency attendance.
Placebo should provoke no such reaction and thus less need to report such
an act of self-harm.
Such data has not emerged to a barrage of opprobrium from the
pharmaceutical industry. Although popular and endorsed in recent NICE
guidelines SSRI’s no longer have novelty value and are rapidly losing
their patents. Data that discredits such old drugs may serve well in
marketing the new generation of antidepressants that will be launched in
coming years.
Competing interests:
previous receipt of research grant from Eli Lilly.
Competing interests: No competing interests
Fergusson et al (2004) report that the risk of suicide attempts is significantly greater for patients enrolled in short term RCTs of SSRIs compared with placebo (odds ratio of 2.28 p<_0.02 and="and" other="other" interventions="interventions" not="not" including="including" tricyclics="tricyclics" odds="odds" ratio="ratio" of="of" _1.94.b="_1.94.b"/>[1] Remarkably, they are also one of the few groups to report that completed suicide (fatal attempts) were also higher with SSRIs than tricyclics (odds ration 7.27). This appears to have led Geddes et al to state that there is almost a two risk of fatal and non-fatal risk of suicide for those taking SSRIs.[2]
An increased risk of fatal overdoses when using SSRIs is hard to understand, particularly when the comparison is tricyclics, given their acknowledged toxicity in overdose.[3-5] For this reason I re-checked the odds ratios from the data given by Fergusson, and found most of them to be incorrect. For example, the authors state that the odds of suicide attempts in SSRIs vs others is 1.94; but with 27 SSRI cases out of 4130 treated patients vs 18 control cases out of 4233 treated patients generates an odds of 1.54 (0.85-2.8) non-significant. Changing the denominator to 8856 and 9059 (all trials) makes no difference to the result. Similarly, the authors state that the odds for non-fatal attempts was 2.25 whereas I calculate it at 1.89 (0.96 –3.73) again non-significant. Perhaps most incomprehensibly regarding the odds of completed suicide vs tricyclics, the number of SSRI cases is 5 and TCA cases 4; obviously a non-significant difference. This is backed up by the calculation which reveals an odds of 1.1 (0.29-4.1) and not 7.27 as stated in the paper. This kind of error really does not help the debate in what is already a noisy area.
In conclusion either most of the raw data printed in table 1 is wrong; or one of us has miscalculated dramatically. I enclose a copy of my “working” for one of the above sums in case anyone else offers to clarify this.
Alex J Mitchell
Consultant in Liaison Psychiatry, Leicester General Hospital
Example Re-Calculation
Chi-square test (2 by 2)
Observed values and totals:
27 18
4103 4215 /8318
4130 4233 /8363
Expected values:
22.222887 22.777113 /
4107.777113 4210.222887
Uncorrected Chi² = 2.0398 P = 0.1532
Yates-corrected Chi² = 1.635152 P = 0.201
Measures of association:
Pearson's contingency = 0.015616
Cramér's V (signed) = 0.015618
References
[1] Fergusson D, Doucette S, Glass KC, Shapiro S, Healy D, Hebert P, et al. Association between suicide attempts and selective serotonin reuptake inhibitors: systematic review of randomised controlled trials. BMJ 2005;330: 396-9.
[2]Andrea Cipriani, Corrado Barbui, and John R Geddes. Suicide, depression, and antidepressants BMJ 2005; 330: 373-374.
[3]Cheeta S, Schifano F, Oyefeso A, et al. Antidepressant-related deaths and antidepressant prescriptions in England and Wales, 1998-2000 BRIT J PSYCHIAT 2004 184: 41-47.
[4] Henry Ja, Alexander Ca, Sener Ek. Relative mortality from overdose of antidepressants. British medical journal 1995 310 (6974): 221-224.
[5] Shah R, Uren Z, Baker A, Majeed A. Deaths from antidepressants in England and Wales 1993-1997: analysis of a new national database. Psychological Medicine 2001: 31 (7): 1203-121.
Competing interests:
None declared
Competing interests: No competing interests
corrected odds ratio for fatal suicide attempts with SSRIs as compared wtih TCAs
The correct odds ratio for fatal attempts in the SSRI versus TCA
comparison is 1.08 (0.28 to 4.09). The corrected estimate is now posted on
the BMJ website.
Competing interests:
None declared
Competing interests: No competing interests