Your grief and guilt overwhelm you. You are so tired you cannot think straight. Your simple joys are lost in an invisible agony. You have pain in your head and back and stomach, real pain. The swamp of your soul suffocates you with despair. All this is your fault, you are worthless, and you might as well die. This is how depression can feel, though people’s experiences of it, including the severity of symptoms, can vary widely. This terrible disease affects about one person in 10 at some point in life and, to treat it, many millions of people have taken antidepressants. Unfortunately, we now have good reasons to think that antidepressants are not effective.
To know if antidepressants work we must, of course, pay close attention to the best evidence about these drugs. There have been many empirical trials of antidepressants, and in the past 10 years or so there have been some good meta-analyses of these trials (a meta-analysis pools data from multiple trials into a single analysis). However, there is a problem: experts disagree about the merits and problems of these empirical studies, and about what we should conclude based on them. Philosophy can help. Philosophy of science is the discipline that studies the concepts and methods of science, and offers a lens through which we can understand what scientific evidence shows us about the world. After witnessing the darkness of depression and the struggle by some of my dearest friends and family to treat this disease with drugs, I began to use my training as a philosopher to understand the evidence about antidepressants. Diving into the details of how antidepressant data are generated, analysed and reported tells us that these drugs are barely effective, if at all.
Depression affects many of us. To the extent that you find the arguments in this essay convincing, the message here could be disappointing. If you are already taking antidepressants, you might decide to stop, but I urge caution. We have little reliable evidence about coming off antidepressants, though there is evidence that people can suffer from withdrawal. Moreover, we have little reliable evidence about alternative modes of intervention, such as talk therapy or lifestyle changes. So, patients should be extra cautious when considering changes to their medications, or foregoing them for other kinds of treatments. A quick essay on a difficult subject must sacrifice depth; for a fuller presentation of the arguments that follow, please see my book Medical Nihilism (2018). If you are depressed, your physician or psychiatrist has clinical experience and insight into your condition – despite the fact that most physicians overestimate the benefits and underestimate the harms of antidepressants, you should continue to consult with them, perhaps with this essay in hand.
The best evidence about the effectiveness of antidepressants comes from randomised trials and meta-analyses of these trials. The vast majority of these studies are funded and controlled by the manufacturers of antidepressants, which is an obvious conflict of interest. These trials often last only weeks – far less than the duration that most people are on antidepressants. The subjects in these trials are selected carefully, typically excluding patients who are elderly, who have other diseases, or who are on several other drugs – in other words, the very kinds of people who are often prescribed antidepressants – which means that extrapolating the evidence from these trials to real patients is unreliable. The trials that generate evidence seeming to support antidepressants get published, while trials that generate evidence suggesting that antidepressants are ineffective often remain unpublished (this widespread phenomenon is called ‘publication bias’). To give one prominent example, in 2012 the UK pharmaceutical company GlaxoSmithKline pleaded guilty to criminal charges for promoting the use of its antidepressant Paxil in children (there was no evidence that it was effective in children), and for misreporting trial data.
Every trial on antidepressants uses a scale to measure the severity of depression of subjects before and after the trial. These scales are deeply flawed, and they bias the research toward overestimating the effectiveness of antidepressants. A typical scale that is often used is called the Hamilton Rating Scale for Depression. This scale has 17 questions, each of which has several possible answers. Each answer receives a particular score, and then the scores for all the questions are added together to give an overall measure of depression severity, for a maximum score of 52 points. The hope when testing a new antidepressant in a trial is that the depression-severity score of subjects in the drug group will decrease more than the depression-severity score of subjects in the placebo group. The scale was invented in 1960 by the psychiatrist Max Hamilton in the UK, and has been in use ever since (from here on, when I mention depression-severity scores, I am talking about this scale).
The problem with this scale is that large changes in a subject’s score can occur as a result of trivial changes in a subject’s real depression. For example, there are three questions about the quality of a subject’s sleep, with a total of six possible points, and there is a question about how much a subject is fidgeting, with up to four points. So a drug that simply made people sleep better and fidget less could lower one’s depression score by 10 points. To put this in context, recent clinical guidelines in the UK have required drugs to lower depression scores on this scale by an average of only three points. When a measurement scale measures what we want it to measure, we say the scale has ‘construct validity’. The general problem with depression-severity scales is that they lack construct validity, and this contributes to overestimating the effectiveness of antidepressants.
If a trial subject gained weight, she might accurately guess she was in the drug group
The placebo effect is when patients improve merely as a result of the medical care they have received rather than as a result of the biochemical properties of their drug. The idea is that the mere expectation that you will get better after receiving medical care can itself contribute to you getting better. Some diseases are more responsive to placebo than others, and depression is one of the most placebo-responsive of all diseases. Since much clinical research aims at discovering the true biochemical effects of drugs, trials include a control group that receives a placebo (sometimes control groups receive competitor drugs), and the allocation of subjects to the drug group or the placebo group is concealed from subjects (this is sometimes called ‘blinding’). To estimate the active biochemical effects of the drug, the measured outcomes in the drug group are compared with the measured outcomes in the placebo group.
Blind-breaking is when subjects accurately guess which group of a trial they are in. This can occur because of the presence or absence of side-effects – for example, two common side-effects of antidepressants are weight gain and problems with sexual functioning, and so if a subject in a trial on a new antidepressant gained weight and developed difficulty achieving orgasms, she might accurately guess that she was in the drug group. This accurate guess could then lead to an expectation that her symptoms of depression will improve, and then her symptoms could in fact improve, by the placebo effect alone. There is not much empirical evidence on the frequency of blind-breaking in trials of antidepressants, though some experts believe that it is very high. (A simple improvement to trials would be to ask subjects to guess their group at the end of the trial, which would give researchers some indication of the extent of the placebo effect in the trial – this is sometimes but not often done, yet could easily be done in all trials.)
Because blind-breaking is occurring in trials on antidepressants, and because depression itself is so placebo-responsive, some prominent researchers (such as Irving Kirsch at Harvard Medical School and Peter Gøtzsche, formerly of the Nordic Cochrane Centre in Denmark) argue that whatever small positive effect is observed in such trials could be due entirely to the placebo effect.
Once researchers have actual measurements from a trial on antidepressants, they must analyse the data in a way that turns the numbers into meaningful evidence about the effectiveness of the drug. The best way to do this is to measure the decrease of depression severity in the drug group and in the placebo group, and then compare the difference between the two. The result is called an ‘effect size’. It gives you – as a real, average patient – a rough indication of how much you could expect your symptoms of depression to improve thanks to the drug. In a moment I will tell you the result when this is done as carefully as possible with all of the data we have on antidepressants. First, though, a cautionary reminder that statistics can be weapons of deception.
There are many ways that researchers can analyse data from trials that render the evidence meaningless and unreliable. Here is one example. In 2018, a meta-analysis about antidepressants was published in The Lancet (one of the world’s most important medical journals). This article, by the psychiatrist Andrea Cipriani at the University of Oxford and colleagues, included many sophisticated analyses. But one simple statistic was widely discussed. This was the ‘odds ratio’ of benefiting from antidepressants. In such studies ‘benefit’ is often defined as occurring when depression severity goes down by more than half. The odds ratio is the odds of subjects in the drug group benefiting divided by the odds of subjects in the placebo group benefiting. The result of their analysis was an odds ratio of about 1.5. On the face of it, this is an extremely modest result. But, in fact, it tells us very little.
To see this, consider an analogy. Imagine we are testing a drug for weight loss. For every 100 subjects in the drug group, three subjects lose one kilogramme and 97 subjects gain five kilos. For every 100 subjects in the placebo group, two lose four kilos and 98 subjects do not gain or lose any weight. How effective is the drug for weight loss? The odds ratio of weight loss is 1.5, and yet this number tells us nothing about how much weight people on average gain or lose – indeed, the number entirely conceals the real effects of the drug. Though this is an extreme analogy, it shows how cautious we must be when interpreting this celebrated meta-analysis. Unfortunately, however, in response to this work, many leading psychiatrists celebrated, and news headlines misleadingly claimed ‘The drugs do work.’ On the winding route from the hard work of these researchers to the news reports where you were most likely to hear about that study, a simple number became a lie.
When analysed properly, the best evidence indicates that antidepressants are not clinically beneficial. The meta-analyses worth considering, such as the one above, involve attempts to gather evidence from all trials on antidepressants, including those that remain unpublished. Of course it is impossible to know that a meta-analysis includes all unpublished evidence, because publication bias is characterised by deception, either inadvertent or wilful. Nevertheless, these meta-analyses are serious attempts to address publication bias by finding as much data as possible. What, then, do they show?
In meta-analyses that include as much of the evidence as possible, the severity of depression among subjects who receive antidepressants goes down by approximately two points compared with subjects who receive a placebo. Two points. Remember, a depression score can go down by double that amount simply if a subject stops fidgeting. This result, found by both champions and critics of antidepressants, has been replicated year after year for more than a decade (see, for example, the meta-analyses led by Irving Kirsch in 2008, by J C Fournier in 2010, and by Janus Christian Jakobsen in 2017). The phenomena of blind-breaking, the placebo effect and unresolved publication bias could easily account for this trivial two-point reduction in severity scores.
We saw above how clinical guidelines have held that drugs must lower severity-depression scores by three points to be deemed effective. On this standard, antidepressants do not pass. Worse, some psychiatrists have argued that this standard is too low – they say that, for an antidepressant to be clinically significant, it must lower depression severity by at least seven points, compared with a placebo. No drug does this.
We treat many normal aspects of life: morning drowsiness with coffee, shyness with alcohol
In short, we have plenty of reasons to think that antidepressants have no clinically meaningful benefits for those suffering from depression. Conversely, we know that these drugs cause many harmful side-effects, including weight gain, sexual problems, fatigue and insomnia. Some studies have demonstrated a link between antidepressants and the risk of violence, suicide, childhood and teenage aggression, and psychotic events in women.
An early theory about depression is that it is constituted by a low concentration of serotonin. Since the class of antidepressants known as ‘selective serotonin reuptake inhibitors’ (SSRIs) helps to increase serotonin levels, it was widely thought that there was a solid theoretical basis for treating depression with SSRIs. However, most researchers now think that this is a grossly simplistic and misleading theory of depression.
One of the main reasons for the serotonin-deficiency theory was the belief that SSRIs are effective at treating depression. The thinking was as follows. Premise one: SSRIs modulate pathological serotonin levels; premise two: SSRIs treat depression; conclusion: depression is constituted by pathological serotonin levels. Notice that, even if this reasoning was persuasive, it would not provide independent grounds for thinking that SSRIs are effective, since that is a premise of the reasoning. So, one cannot respond to this essay by saying ‘but we have theoretical reasons for thinking that antidepressants are effective’. Moreover, the thesis of this essay calls into doubt the second premise.
Conversely, there is another theoretical consideration that seems to speak against antidepressants. Some critics claim that many diagnosed cases of depression are not cases of real disease, but rather involve the ‘medicalisation’ of normal life – normal grief, stress, anxiety or simply suburban sadness being brought into the jurisdiction of medicine. If a case of sadness is inappropriately medicalised, goes this thought, then treatment with pharmaceuticals is also inappropriate. However, I do not find this criticism of antidepressants compelling. Beneath its surface lurk controversial premises about the concept of disease, the nature of normality and the proper jurisdiction of medicine. We treat many normal aspects of life with external aids, such as morning drowsiness with coffee, shyness with alcohol and erectile dysfunction with drugs. So, in short, these theoretical considerations – about the pathophysiology of depression or about the medicalisation of depression – are not persuasive one way or another regarding the effectiveness of antidepressants.
That said, we ought to be suspicious of theories that characterise depression as a simple disease, constituted by a deficiency of such-and-such a chemical – as most researchers recognise, depression is not like scurvy (constituted by a deficiency of vitamin C) or Type 1 diabetes (constituted by a deficiency of insulin). We can cure scurvy with vitamin C, and treatment of Type 1 diabetes with insulin is miraculous. Since depression is a complex disease, it is implausible to expect that it could be successfully treated just by nudging chemical levels, as we do with scurvy.
I have been focusing on evidence from trials on antidepressants. Despite all of the problems I have noted about these trials, they are nevertheless our best source of evidence about the effectiveness of antidepressants. However, there is another source of evidence we could consider: the experience of real patients. You – or your friends and loved ones – might have taken antidepressants, and this could have convinced you that such drugs can be effective for some people.
Attending to the testimony of patients is integral to good medicine. But such testimony is typically not a good guide to causal inference. First-person reports are unreliable when determining whether or not antidepressants are effective. There are at least three reasons for this. First, the severity of symptoms of depression fluctuates and improves over time, and people tend to seek treatment when their symptoms are more severe. So, after being treated, symptoms are likely to improve, not because the treatment is effective but merely because time passes, like the gradual healing of a wound.
Second, depression is very placebo-responsive. For a large proportion of subjects in the placebo group of trials, depression-severity scores decrease by as much as 10 or 15 points. The placebo effect is amazing – for example, bigger placebo pills elicit greater effects than small pills. Third, confirmation bias is the tendency of people to notice evidence that confirms their expectations and ignore evidence that disconfirms their expectations. This cognitive failure affects us all. After taking an antidepressant, people tend to notice signs of their health improving more than they notice opposing signs.
So, if you hear of someone benefiting from antidepressants, this was likely due to the natural course of the disease fluctuating or improving over time, confounded by the placebo effect, and exaggerated by confirmation bias. This is not to doubt the testimony of patients. Their first-person experience is, ultimately, the most real and most important phenomenon in medicine. We must listen to it. But, away from the clinical encounter, sitting at our desks with statistics, science and sober reflection, when we listen, what do we hear? Placebo, not Prozac.
