The Long-term Consequences of Antidepressant Use: An Interview with Michael Hengartner

MIA Editors

Researchers at the University of Zurich, led by Michael Hengartner, recently reported that antidepressant usage was associated with worse outcomes in patients followed over 30 years. In this interview, Hengartner provides more information about the study methodology and their findings, and their plans for future research on this topic.

Michael Hengartner

Michael P. Hengartner, PhD, is senior researcher and lecturer at the Zurich University of Applied Sciences in Switzerland. He holds a doctorate in clinical psychology and a qualification for professorship in medicine. His main research interests are psychopathology, psychosomatics, psychiatric epidemiology, social psychiatry, and clinical psychology. He is married and father of three children. Follow him on Twitter @HengartnerMP

Q. You recently published a study that found that antidepressant usage was associated with worse outcomes over a thirty-year period. Can you tell us a bit more about the nature of the study?

A. We used the data of a longitudinal epidemiologic survey, known as the “Prospective Zurich cohort study,” to conduct this analysis. This remarkable longitudinal study was designed and conducted by principal investigator Dr. Jules Angst, and later on by Dr. Wulf Rössler, both affiliated with the University Hospital of Psychiatry in Zurich, Switzerland. I had the privilege to access this unique dataset because I am a former research associate of both Dr. Angst and Dr. Rössler.

The Zurich Study was sponsored by the Swiss Science Foundation to advance our knowledge of psychiatric epidemiology. The Zurich study began in 1978 with the enrollment of a representative sample of 4547 young adults (the men were 19 years old and the women 20 years old) from the province of Zurich, Switzerland. Based on a brief psychiatric screening test, the participants were classified as either at high risk or at low risk of mental disorders.

Next, a sample of 591 people, comprising two-thirds of high-risk subjects and one-third of low-risk subjects, were randomly drawn from this initial screening sample to participate in the longitudinal survey. This procedure is referred to as a stratified sampling procedure and is common in psychiatric research due to the low prevalence of some mental disorders in the general population.

The first comprehensive assessment of this group was conducted in 1979, when the participants were 20/21 years old, with a semi-structured clinical interview, which captures psychopathology, treatment, social functioning, and physical health. These comprehensive psychiatric assessments were repeated in 1981, 1986, 1988, 1993, 1999, and, finally, in 2008. That is, the sample was followed over a total observation period of 30 years as participants progressed from age 19/20 years to 49/50 years.

Please note that this cohort study, at the outset, was not designed to measure the long-term effect of psychopharmacological treatments. The main objective of the Zurich study was to determine the prevalence and course of mood and anxiety disorders in the community, which, back in the late 70s and early 80s, were mostly unknown. So, the present analysis was necessarily a post-hoc analysis that I initiated due to my growing interest in the long-term effects of antidepressant pharmacotherapy.

Q. So what was the primary outcome in your study, and how was it measured?

A. The primary outcome in the present study was the severity of depressive symptomatology within the past 12 months as assessed at each assessment wave (i.e. in 1979, 1981, 1986, 1988, 1993, 1999, and 2008). As detailed in the paper, we defined the following four levels that reflect increased illness severity: 1) no depression symptoms, 2) transient and minor depression symptoms, 3) subthreshold depressive disorder (distressing symptoms that do not qualify for a psychiatric diagnosis), and 4) major depression (according to DSM-IV diagnostic criteria). We preferred this continuous outcome (no symptoms to clinically relevant symptoms) over a dichotomous diagnosis (i.e. disorder present vs. absent), because changes in psychopathology are often subtle and they are inadequately captured by broad diagnostic categories. Full remission and acute fully symptomatic diagnosable disorders are extreme poles along a dimension where most observable change (i.e. improvement and deterioration) occurs in between these extremes within the moderate range of illness severity.

Imagine for instance a person who had major depression and who was started on an antidepressant drug. A few years later his/her symptoms were re-assessed and it was found that the person still had debilitating depression symptoms, but these did not cross the diagnostic threshold (i.e. subthreshold depression). Another person also had major depression at baseline, but did not use antidepressants and at follow-up, this person reported no symptoms at all. Now imagine we had only assessed psychiatric diagnoses. In both cases major depression was present at baseline and absent at follow-up, so no drug-effect would have been detected. However, by looking at a continuous depression outcome graded according to illness severity, it becomes evident that the person who used drugs experienced only a slight improvement in depression symptoms, whereas the non-user experienced a full remission. In consequence, a potential adverse drug effect only has been detected because subtle changes along a dimensional gradient of illness severity were assessed (akin to the dimensional scores based on rating scales for depression applied in the clinical trials).

Q. Can you explain a bit more about what you found during the periodic assessments?

We used a time-lagged regression model to test the association between antidepressant use at any time point and subsequent depression severity. This means that a separate association was computed for all consecutive assessment waves. In detail: Antidepressant use in 1979 (baseline) was related to depression severity in 1981 (follow-up), antidepressant use in 1981 (baseline) to depression severity in 1986 (follow-up), and so on up to antidepressant use in 1999 (baseline) related to depression severity in 2008 (follow-up). In total there were thus 6 unique prospective effects that were pooled statistically to obtain one single effect size estimate for the whole 30-year observation period.

To minimize confounding by indication, which qualifies the degree to which an association between treatment (at baseline) and outcome (at follow-up) is biased by illness severity at baseline and other factors, we statistically controlled for various potential confounders. Imagine for instance that only people with severe major depression use antidepressants (which is certainly not true, but let’s just assume it), then it would follow that treatment relates to a poor outcome, because people with severe forms of depression typically have a poorer outcome independent of treatment received. Therefore, we included several markers of severe depression, such as the presence of severe suicidality at baseline, comorbid anxiety disorder, or high subjective distress at baseline.

Since the mean time interval between consecutive assessments was approximately five years, the interpretation of the reported effect is that at any given time between the age of 20 and 50, people with some kind of depressive symptoms who use antidepressant drugs have, on average, an 81% increased odds to have a more severe illness at five-year follow-up than people who did not use antidepressants (when statistically adjusted for sex, education level, marriage, any affective disorder at baseline, severe suicidality at baseline, and family history of depression).

More specifically, if a person had only minor depression symptoms at baseline, he/she had an 81% increased odds to experience subthreshold depression at an average follow-up of 5 years. If a person had subthreshold depression at baseline, then antidepressant use related to an 81% increased odds to have diagnosable major depression at follow-up (controlling for the potential confounders detailed above).

Q. So you found this association between antidepressant use and subsequent worse outcomes at every assessment?

A. Yes, it was present at every assessment, that is, between all consecutive time points. However, due to small number of antidepressant users at some time points, it did not always reach statistical significance. This is exactly the reason why it is important to use sophisticated statistical methods that consider repeated measurements and that provide a pooled estimate across all measurements. In some way this statistical procedure compares to meta-analysis, which pools effects of several individual studies to arrive at one single, mean effect size. Analysis of repeated measures has further advantages. For instance, it allows for examining time trends (i.e. age effects) and to control for time-varying confounders (e.g. varying illness severity of different depression episodes).

Q. Was there any further investigation you did, beyond anything set forth in the published paper, to test this association between antidepressant use and worse long-term outcomes?

A. To test the robustness and generalizability of a reported effect, researchers commonly conduct so-called sensitivity analyses. The aim of these supplementary analyses is to increase the confidence in and credibility of a reported association. We conducted several sensitivity analyses that did not make it into the published paper due to constraints on manuscript length, because the journal editor decided to accept the work as a short report only (instead of a full-length article).

So here I will add some interesting details that were not reported in the published article. As touched on above, the reported prospective effect refers to antidepressant use at some time point in association with subsequent severity of depression over an average follow-up period of 5 years (however, note that some time intervals were considerably smaller, such as that between the assessments in 1986 and 1988, and others much longer, such as that between 1999 and 2008). Not reported in the published article is whether the age at baseline assessment played a role in the strength of association (i.e. potential age effects). Sensitivity analysis revealed that this was indeed not the case. Age at baseline didn’t play a role in the strength of the association. The reported effect was stable across time and applies to the entire age span from 20 to 50 years.

We also didn’t report, in the published article, whether long-term antidepressant medication (i.e. antidepressant use present at two consecutive time points) related to a different outcome than discontinuation of pharmacotherapy (i.e. antidepressant use at the baseline time point but discontinued use at the subsequent follow-up time point). Although case numbers were small, sensitivity analysis revealed that drug discontinuation and persistent non-use related to a significant reduction in depression severity over time, whereas long-term therapy related to chronic/recurrent symptoms (i.e. no apparent change in symptoms over time).

Q. In other words, although the numbers were small, no use was better than exposure followed by discontinuation, and discontinuation was better than continual antidepressant use? Is that right?

A. Yes, this is exactly what this additional analysis revealed, although the difference between no use and discontinuation revealed only a trend towards statistical significance. However, and most importantly, both non-use and discontinuation related to a markedly better outcome than long-term use. So, put differently, the association between antidepressant use and worse outcome reported in the paper was mainly due to long-term drug use, although discontinuation was also related to a slightly increased risk of worse depression. But again, there were only few records of long-term use in this community cohort (specifically 21), so this effects needs to be replicated using a larger sample to increase credibility.

Q. Did you do any other sensitivity analyses?

A. A third sensitivity analysis focused on intra-individual change over time in both medication use and symptom severity. Analysis of intra-individual change adopts a slightly different statistical modelling approach. But even with this alternative analysis the effect reported in the published paper was fully and consistently reproduced. That is, the reported prospective association between antidepressant use and subsequent depression severity was robust and credible.

Q. Do you have further research planned on this issue, which is obviously so important?

A. As discussed in the paper, the major limitation of this research (and similar other papers published before) is non-random treatment assignment. Although we controlled for various potential confounders, we cannot rule out that unmeasured factors, such as an individual’s coping skills and personality traits, did bias the reported association. Unfortunately, randomized controlled trials are no alternative, because they cannot follow-up large samples over such long observation periods. Therefore, other approaches are necessary to eliminate selection bias.

One solution is to use propensity score matching analysis, a statistical technique which links drug users to comparable non-users based on a multitude of clinically important covariates. This statistical method ensures that drug users and non-users differ only with respect to treatment selection, but not with respect to other confounding variables (such as social functioning, number of preceding hospitalizations, receipt of disability benefits, and so on). At this moment we have such a paper under review and like the one just published, it shows that drug-treated people have the poorer long-term outcome than unmedicated patients.

In contrast to the present analysis which was based on a community sample, this work is focused on a clinical sample of psychiatric inpatients treated at two different sites in the province of Zurich, Switzerland, and the primary outcome was rehospitalization rates within twelve months after discharge from the index hospitalization. However, as you well know, due to institutional corruption within academic psychiatry it is quite difficult to successfully pass the review process with such papers. Most psychiatric experts reviewing for the leading scientific journals refuse peremptorily any report calling into question the merits of psychiatric drugs. So it could take some time until this work gets published.

Q. Are there any other research projects you are working on?

A. Yes, of course. Another important limitation of the present work is that we do not exactly know for how long someone was using antidepressants, because we only had data from seven assessment waves spread across 30 years (and not an assessment each year). Another aim of future research should thus be to consider the duration of pharmacotherapy and the adverse consequences of stopping long-term maintenance therapy. That is, we not only have to look at the long-term outcome of acute pharmacotherapy, but also at the long-term outcome of long-term antidepressant use (i.e. maintenance pharmacotherapy over several years). As detailed above, it appears that long-term use is particularly problematic.

Such analyses should also focus on physical health, as long-term drug use likely plays an important role in the disruption of adaptive bodily functions (such as sexuality, digestion, immunity and metabolism) that may increase vulnerability to serious physical diseases. We are currently designing such a prospective longitudinal study on mental and physical health consequences of antidepressant use based on comprehensive long-tern medical data derived from insurance claims. So there is definitely more to come from our research lab on this topic.


  1. This comment, while undoubtedly true, is a devastating critique on modern psychiatry practice, which ought be to be acutely sensitive to harms of any and all of our interventions: “…due to institutional corruption within academic psychiatry it is quite difficult to successfully pass the review process with such papers. Most psychiatric experts reviewing for the leading scientific journals refuse peremptorily any report calling into question the merits of psychiatric drugs.”
    FIRST DO NO HARM – and such exemplary work as Michael Hengartner and crew are doing should be front and centre of the leading journals. Thanks for the excellent and ongoing work!