select search filters
roundups & rapid reactions
before the headlines
Fiona fox's blog

expert reaction to study on association of antidepressant use with adverse health outcomes

A study, published in JAMA Psychiatry, examined the existing evidence on antidepressants and links to negative health outcomes. 


Dr Sameer Jauhar, Honorary Consultant Psychiatrist and Research Fellow, Institute of Psychiatry Psychology & Neuroscience, King’s College London (IoPPN), said:

“In science we are dependent on summaries of evidence (meta-analyses), which pool studies together. Given the widespread use of antidepressants it is vital that we keep track of their effects, and side-effects as evidence emerges.

“If we wish to examine risk factors for a disease or treatment we can look at real world observational studies, which examine risk factors associated with illness or intervention. One problem with interpreting these studies is the difficulty in assessing causality. An association may be due to another factor, separate from the one examined, or the reason for the intervention, i.e. the illness itself (confounding by indication).

“Umbrella reviews, like this one, grade the quality of meta-analyses, and require authors to go through all the studies making up these analyses and reclassify them.

“They are the highest level of evidence synthesis for observational studies.

“This is the first umbrella review of SSRIs, examining adverse effects of these drugs, and should be welcomed.

“The authors examine a number of associations identified in meta-analyses, and when an association was identified they examined how good the evidence was, and whether the reason for the drug (e.g. depression) better explained the risk.

“They found that some identified associations, from high quality evidence, e.g. antidepressants and preterm birth, were not as clear when the effects of illness were accounted for.

“What does this mean?

“When giving advice to people about antidepressants it gives clinicians a better idea of risks and benefit of these drugs, and emphasises how important it is to treat the underlying illness itself. Some risks, e.g. in pregnancy, may not be as high as studies indicated, and treatment of the underlying illness should be emphasised.

“It also emphasises the importance of future studies controlling for the effects of illness.

“As a review of existing evidence it can only comment on studies already performed, and cannot comment on findings from RCTs, which have already led to warnings about risks in some populations, e.g. children and adolescents and suicide attempts.

“In summary, this is a high quality piece of work, summarises the evidence for adverse effects of antidepressants well, and concludes that a lot of the associations for these adverse effects are affected by the illness itself.”


Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:

“This systematic review is different from most of the systematic reviews one sees. It is an umbrella review, that is, a systematic review of systematic reviews. It looks across a considerable number of previous systematic reviews, and (among other things) assesses their quality in various ways, summarizes them as far as that can be done, and looks for gaps in what is known. It is a review of adverse health outcomes that might be associated with antidepressant use. Each of the systematic reviews that it builds on looked at just one or a few potential adverse outcomes, and in many cases just at one class of antidepressant drugs. Because this umbrella review looks across all these, it can look for consistencies and inconsistencies in the evidence, though with such a wide range of potential adverse effects and drug types, consistencies may be hard to find. The new review does not attempt to look at how effective antidepressants are in treating mental health problems such as depression. It also does not look at any evidence from randomised clinical trials of antidepressants. This last point might seem surprising – there have been a great number of antidepressant trials, and though most would have concentrated on the efficacy of the drugs, they do collect some information on adverse effects too. The researchers’ reasons for not considering randomized trials are that the participants are often not typical of the populations who are treated with the drugs outside the trials, that adverse outcomes are not consistently reported, and usually the length of the trials is not great, so that adverse outcomes that take some time to show up will not be observed. I think the omission of data from trials is reasonable, but it does lose a certain amount of information that would otherwise be available. Statistical methods for systematic umbrella reviews are not so well developed or standardised as for the more common type of systematic review, but I see no problems with the way these researchers did their review.

“The researchers examined and re-analysed each systematic review according to a clear protocol, and assessed the credibility of evidence from each of the meta-analyses in the reviews. (A meta-analysis is the standard way of putting together evidence from several separate studies in a systematic review, and each of the reviews they considered included one or more meta-analyses.) If a meta-analysis satisfied all of seven different criteria, the researchers classified it as providing convincing evidence of an increased risk of the adverse health outcome it was considering. If it satisfied just three of these criteria, they classified it as providing highly suggestive evidence, and there were other categories below that. There was a separate assessment of the quality of each meta-analysis in terms of the statistical and other methods it used, using a standard method for such assessments, but the main conclusions of the new research seem to be based on the scoring of the credibility of evidence.

“Overall, only three potential adverse health outcomes were classified as having convincing evidence. These were an increased risk of autism spectrum disorders in the children of mothers who took antidepressants before pregnancy, an increased risk of autism spectrum disorders in the children of mothers who took a particular class of antidepressants (SSRIs) during pregnancy, and an increased risk of suicide attempts or suicide in children and adolescents who took SSRIs. It’s important to note that, even though two of these relate only to SSRIs, that’s no indication that the effects cannot occur with other antidepressants, only that no previous systematic reviews had provided good evidence on other antidepressants for these outcomes. Another eleven potential adverse health outcomes were classified as having highly suggestive evidence, 21 as having suggestive evidence, and the rest (85 outcomes, or over 70% of all the outcomes they considered) were classified as having only weak evidence or no evidence from the reviews that were reviewed.

“The researchers then carried out what they call ‘sensitivity analyses’, where they reconsidered and re-classified all of the potential adverse effects on the basis of just a certain subgroup of the studies that were involved. This, unsurprisingly, changed the classification of the strength of evidence for some of them, often downgrading it but in a few cases upgrading it. This can be a useful procedure, particularly if a previous meta-analysis included studies that used different methods or included different types of patients.

“One point that particularly came out of these sensitivity analyses is that, if the analysis is restricted to studies where an adjustment was made for confounding by indication, no adverse health outcomes were classified as having convincing evidence. Confounding by indication is a potential problem in the interpretation of pretty well any observational study of a medical treatment. In an observational study, patients aren’t allocated to different treatments (or to no treatment) at random by the people running the study, as would be the case in a randomised trial. Instead, generally, the clinicians treating the patients decide which patient gets which treatment, and they will make that decision on the basis of their diagnosis of the patient (and possibly other things) – that is, in the jargon, on the indication for the treatment. So, on average at least, patients who get different treatments will differ in terms of the condition that led them to have a need for treatment. These differences in indication might be the real cause of any difference in the outcomes of the treatments, and not the actual treatments themselves. So, for instance, in an observational study comparing antidepressant treatment with no treatment, doctors may well be more likely to give antidepressants to patients with more severe symptoms. If the study is looking at adverse health outcomes, perhaps those would be more common in patients who initially had more severe depression, because of their depression, and not because of the antidepressants they took – but this could show up in the results as a difference between the patients who took antidepressants and those who didn’t. If information is available to researchers on the reasons for the choice of antidepressant treatment, statistical adjustments can be made to allow, at least in part, for this confounding by indication. But many studies don’t do this, often because the information on indications for treatment just isn’t available.

“In this new research, when only studies that did adjust for confounding by indication were included, no potential adverse effect was classified as having convincing evidence (though some still had highly suggestive evidence). What this doesn’t mean is that, if we consider the possibility of confounding by indication, there’s no good evidence of increased risk of adverse health outcomes in people who take antidepressants. Some effects did still have highly suggestive evidence, and another important point is that, in the studies that did not or could not adjust for confounding by indication, we don’t actually know how much their results would have changed if they had been able to make that adjustment.

“The new research report does point out clearly that some potential harmful effects, even if the evidence for them is far from perfect, can in any case be prevented medically by additional treatments, or by avoiding the use of antidepressants with certain types of patient. Also, we’ve got to remember that all drugs carry a certain risk of adverse effects. The important thing is to try to get the right balance between the good and the harmful effects. Importantly, what was found in this study about most potential adverse outcomes was that the evidence for their existence was not strong. Lack of evidence for something does not mean that it does not exist. That’s one possibility, but another is that it does exist but research so far hasn’t been able to find it. Crudely put, maybe researchers were looking at the wrong people, or in the wrong way. This isn’t a criticism of those researchers; some of these potential effects are really hard to study. This is why I support the authors of this new study when they call for further research, perhaps using more informative types of randomized trial where possible.”


Dr Gemma Lewis, Senior Research Associate in Psychiatric Epidemiology, University College London (UCL), said:

“This is a very good study which has looked at the quality of the evidence on whether antidepressants lead to negative health outcomes, and how convincing the data is.

“This is good because, often, reviews of the evidence don’t group studies according to how good they are or what biases they have controlled for, and then examine how this might change the findings.

“Initially there was convincing evidence that antidepressants might increase the risk of suicide in young people and that, when used by pregnant women, they might lead to a range of negative outcomes, including autism, in the children. However, these were naturalistic studies of people taking and not taking antidepressants in the general population (as opposed to well controlled clinical trials). We know that, in the community, antidepressants are prescribed to people with more severe depression, who might already be at higher risk of suicide and other outcomes. In some studies it might therefore look as though antidepressants increase the risk of suicide and other negative events when, actually, it is the more severe depression that causes this. When the study focused only on data that had controlled for this type of bias, there was no convincing evidence that antidepressants led to any of these negative outcomes.

“Although these results are reassuring, they show that we need more, higher quality, studies in this area.”


Prof Wendy Burn, President of the Royal College of Psychiatrists, said:

“This very large review of evidence, pulling together data from more than a million people over many separate observational studies, is reassuring in that it confirms that antidepressants are generally safe, and indicates that they protect from suicide in adults.

“After adjusting for the fact that young people prescribed antidepressants are a higher risk group with more severe mental health problems, the authors found no convincing evidence of an association between antidepressants and suicidality.

“It is important that if an antidepressant is to be prescribed to this age group it should only be following assessment and diagnosis by a child and adolescent psychiatrist. Specific arrangements must be made for careful monitoring of any adverse drug reactions.

“This area still needs more research. More trials are needed that include all the types of patients that doctors see in their clinical practice.”


Association of Antidepressant Use With Adverse Health Outcomes, A Systematic Umbrella Review’ by Elena Dragioti et al. was published in JAMA Psychiatry at 16:00 UK time on Wednesday 2nd October. 

DOI: doi:10.1001/jamapsychiatry.2019.2859


Declared interests

Dr Sameer Jauhar:

Sameer Jauhar works at the same organisation as some of the authors of this study, but he was not involved in this work.

Sameer is Co-investigator on a research study in psychosis, funded by Alkermes. Sameer has not received fees for being a Co PI or any expenses for this.

King’s College, London, has received fees from Lundbeck for lectures Sameer Jauhar has given on psychosis.

Funding: National Institute for Health Research Biomedical Research Centre at South London and Maudsley National Health Service Foundation Trust and King’s College London SJ is funded by a JMAS (John, Margaret, Alfred, and Stewart) Sim Fellowship from the Royal College of Physicians, Edinburgh.

Prof Kevin McConway: Prof McConway is a member of the SMC Advisory Committee, but his quote above is in his capacity as a professional statistician.

Dr Gemma Lewis: No conflicts of interest

None others received.

in this section

filter RoundUps by year

search by tag