Summary of findings
Our large pragmatic trial of the effectiveness of maintenance antidepressants in primary care found that the rate of relapse was twice as high in those who received placebo as in those who continued to take antidepressant medication during 12 months of follow-up (hazard ratio 2.04, 95% CI 1.55 to 2.68). Most relapses occurred 12–26 weeks after randomisation and 4–18 weeks after medication was tapered.
Our analyses of secondary outcomes found strong evidence that depressive and anxiety symptoms at 12 weeks were higher in those receiving the placebo than in those receiving antidepressant medication, and the size of this difference was likely to be clinically meaningful.76 Weaker evidence of a smaller difference in depressive and anxiety symptoms remained at 6 months, but attenuated thereafter. People in the discontinuation group were more likely than those in the maintenance group to self-report worsening of mood and had lower mental health-related quality of life at 12 weeks, although these differences were not evident at any other time point. At 12 weeks, 44% of those who discontinued their antidepressants reported feeling worse, compared with 21% of those who remained on their medication. Such global changes in mental health are used to calculate minimal clinically important differences and are regarded as patient-centred indicators of clinically meaningful change.77 We also found strong evidence that patients who discontinued antidepressants stopped their trial medication sooner than those who remained on maintenance treatment (HR 2.26, 95% CI 1.67 to 3.07). By the end of the trial, 39% (95% CI 32% to 45%) of participants in the discontinuation group had returned to an active antidepressant prescribed by their clinician, which may explain why treatment effects attenuated after the 12-week follow-up.
People who discontinued their antidepressant experienced more withdrawal symptoms than those who remained on maintenance treatment at every time point except 52 weeks, by which time the difference had attenuated substantially (and many people in the discontinuation group had returned to taking an antidepressant). The difference in withdrawal symptoms between the groups was largest at 12 weeks and reduced as follow-up progressed. There was no evidence of a difference in physical symptoms that could be attributed to antidepressant side effects (in contrast to withdrawal effects) until 39 and 52 weeks after randomisation, when those receiving an antidepressant reported slightly more physical symptoms than those receiving placebo. It is possible that any difference in side effects was masked by the increase in depressive and/or withdrawal symptoms in the discontinuation group.
Summary of economic evaluation findings
The cost–utility analysis suggests that there is a low probability that discontinuing antidepressant medications with replacement by placebo is cost-effective compared with continued maintenance of antidepressant treatment in this population. Participants randomised to the discontinuation group experienced significantly fewer QALYs over 12 months than those in the maintenance group, with this difference primarily driven by an increased rate of relapse. Participants who were randomised to the discontinuation group also had higher GP consultation costs and Improving Access to Psychological Therapies costs. The difference in GP costs appeared to be driven by a shorter time to relapse in the discontinuation group, potentially because participants arranged to see their GP following relapse to review their medication, with 53% (95% CI 44% to 62%) of those who relapsed in the discontinuation group having returned to a known antidepressant before the end of the trial.
A limitation of the analysis is that the time horizon is the same as the follow-up period for the trial (i.e. only 12 months). It is possible that there are longer-term impacts on costs and QALYs, such as medication side effects, feelings of stability derived from maintenance medication or feelings of liberation derived from having become ‘medication free’, that are not captured by this analysis. However, decision analytic models of depression rarely go beyond 12 months, and we are not aware of any other studies with longer time horizons on which a decision model with a longer time horizon could be based; therefore, any modelling would have required many untested assumptions. In addition, given the relatively large difference in QALYs between tapering and maintenance compared with the relatively small difference in costs, it seems unlikely that discontinuation could potentially be cost-effective over the long term. Missing values also represent a limitation to the analysis; this analysis used multiple imputation, which carries the assumption that values are missing at random, but we cannot know whether or not this is correct. A further limitation of the analysis is that the effectiveness measures used were generic health-related quality-of-life measures and, although there is no evidence to suggest that other quality-of-life measures might perform better in this context, it is still possible that information regarding other factors that are important to participants was not captured. Finally, we compared maintenance antidepressants with the replacement of antidepressants with a placebo over a tapering period, although the use of placebo after discontinuation is not an indicated treatment. We chose to include a placebo to ensure that we were studying the pharmacological effects of the antidepressant drug, which is essential to inform policy. However, it is likely that the difference between maintenance and discontinuation without a placebo would be larger.
At a £20,000 to £30,000 threshold for a QALY gain (i.e. the preferred NICE threshold78) there would need to be a cost saving of £200–600 per person for tapering over the lifetime horizon following the first 12 months to balance the QALYs lost during the first 12-month period by discontinuing the medication; it is not clear where in the longer-term patient pathway this could potentially occur. This highlights the need for further research into the longer-term implications of discontinuing antidepressant use compared with continuing with long-term maintenance for the prevention of relapse, in terms of both longer-term costs and longer-term benefits to patients.
Overall, there was no substantial cost to the health-care system of discontinuing antidepressants; there were costs only to patients in terms of their well-being. Although the evidence from our analysis would not support recommendations at a national level to discontinue antidepressants based on cost-effectiveness, there is the opportunity to provide participants with the information that they require to make a more informed choice about their continued antidepressant prescription. There was a difference in utility scores at 3 months, although this difference had disappeared by 12 months, meaning that any detriment due to relapse or tapering the original medication, although significant, was, on average, short-lived. Some relapses, however, can be potentially severe and disabling, and our analysis offers no prediction of the impact on any particular person. Patients who wish to discontinue their antidepressant, and are willing to accept the risk of a shorter time to relapse, have the possible future benefit of eventually being medication free. The key costs would be mainly to themselves in terms of an increased likelihood of a relapse that could be potentially severe, for a short-term reduction in health-related quality of life, plus health service costs of an additional GP appointment to manage a potential relapse and other treatment options, such as cognitive–behavioural therapy. In societal terms, we also did not find a significant impact on productivity, partly because of the very wide CI around the result.
Strength and limitations
To our knowledge, the ANTLER trial is the largest individual randomised trial of antidepressant maintenance treatment that is not funded by the pharmaceutical industry. Clinical trials are often criticised for using narrow inclusion criteria, which can reduce external validity.79 Our design ensured that we recruited people currently receiving long-term maintenance antidepressants in primary care. Participants had been receiving antidepressant medication for ≥ 9 months (the majority for > 3 years), which, to the best of our knowledge, was a longer treatment period than any previous trial. Our results are, therefore, more readily generalisable than the results of previous trials to the population currently receiving long-term maintenance antidepressants in primary care. We used the four most commonly prescribed antidepressants in primary care. Given that the vast majority of antidepressant prescriptions are for long-term treatment, including maintenance, we think that our results will apply to most people receiving maintenance treatment. We investigated three SSRIs, which have a similar pharmacological profile and act via similar mechanisms. Our results should, therefore, be generalisable to other SSRIs when used in this population. Although there are similarities in the mechanism of action of all commonly used antidepressants, it is more difficult to generalise our findings to other classes of antidepressant. We were unable to investigate whether or not treatment effects differed for mirtazapine because of the small number of participants receiving this antidepressant. We investigated the usual doses of antidepressants used in the UK. Our results should generalise to higher doses, given that there is no evidence of added efficacy,22 but it is difficult to generalise our findings to maintenance treatment at lower doses.
Attrition is a limitation of most randomised controlled trials (RCTs), and more participants in the discontinuation group than in the maintenance group dropped out. However, attrition was modest and results were unaltered after we adjusted for variables associated with missing data. Participants in the discontinuation group correctly guessed which group they were in more often than those in the maintenance group. This is consistent with prior studies suggesting that patients can distinguish placebo from active treatment. In principle, this could bias the outcome, but it could also be a consequence of relapse or withdrawal symptoms resulting from allocation to discontinuation (and on the causal pathway from exposure to outcome).
Although our design was embedded in clinical practice, it introduces some potential limitations. We had to ask retrospectively about past history of illness and treatment. In particular, we do not have detailed information of the original clinical decision for prescribing the antidepressant or any diagnostic information at that time. Participants who had more recently become well may have been more vulnerable to relapse, although this is unlikely to have biased the treatment effect owing to randomisation. Although our sample is likely to be more representative than prior trials, only a small proportion of those potentially eligible participated, which is common in RCTs. We also recruited people who had experienced at least two prior depressive episodes, so our findings do not necessarily apply to those receiving treatment for their first depressive episode. Participants were also mostly white, married and employed, and were recruited predominantly from moderately sized general practices in urban areas. Those who participated were older than those invited who did not participate. This limits generalisability but is still an improvement on existing literature and provides more realistic estimates of what might happen if someone on long-term maintenance treatment stops their antidepressant, at least for the first 12 months after discontinuation. A more important limitation is the ability to generalise to other health systems, although there are striking similarities in how antidepressants are used in wealthy countries.80
In our trial, we used a novel assessment (the rCIS-R) to measure the reappearance of depressive symptoms. The results of our test–retest reliability study nested within the ANTLER trial provide strong evidence that the rCIS-R is a reliable measure of assessing reappearance of depressive symptoms. There was excellent agreement for definitions of relapse, the individual symptoms that were assessed and the sum of the symptoms scores. The main advantage of the rCIS-R is that, to our knowledge, it is the only fully structured interview assessing time to relapse. The pragmatic advantages that deserve considerations are time required for completion, simplicity of scoring and absence of any special training requirements.
Implications for health care
Our findings have several implications for the management of depression in primary care. Our evidence that discontinuing long-term maintenance antidepressants increases the risk of a clinically significant relapse in the following 12 months suggests that, for many patients, long-term treatment is appropriate. However, the severity of relapse varies. Although 56% of people who discontinued their treatment experienced a relapse, only around half of them (53%, 95% CI 44% to 62%) chose to return to an antidepressant prescribed by their clinician. It is possible that some relapses, or possible withdrawal symptoms, were not severe enough for the individual to decide that they needed to return to their medication. If six (95% CI 3 to 19) people stopped their medication, we estimated that one would experience a relapse who may not have experienced a relapse if they had remained on maintenance treatment. Our results illustrate that remaining on maintenance antidepressants does not guarantee well-being and is offset by any adverse events and the reluctance of many people to stay on medication for many years. If people who want to discontinue their antidepressants are regularly monitored in primary care during the first 6 months, relapse prevention may be possible with alternative treatments, such as psychological therapies. For example, there is good evidence that mindfulness-based cognitive therapy (with support to taper or discontinue antidepressants) is as effective as maintenance antidepressants at preventing relapse.53
Current best practice is to engage with patients’ priorities and collaborate in coming to a decision about medication. For the individual patient, it is possible to know only about the average likelihood of relapse and that the severity of potential relapses will be unpredictable. Our findings will give patients and physicians an estimate of the likely benefits and harms of stopping long-term maintenance antidepressants to inform shared decision-making in primary care.
Recommendation for research
The ANTLER trial identified the following research needs:
Further research into the longer-term implications of tapering antidepressant use compared with continuing with long-term maintenance for the prevention of relapse, in terms of both longer-term costs and longer-term benefits to patients.
In addition to having longer follow-up duration, it would be beneficial if future studies were larger so that differences according to antidepressant type could be identified.
Further research into distinguishing between withdrawal symptoms and depressive symptoms in the first few weeks of tapering antidepressant medication. This could involve experimental designs, new measurement instruments or factor analyses.
The rCIS-R was designed for research purposes, although it may also have application in clinical practice as a simple way of assessing relapse in depression. This could be investigated in future studies.
