Antidepressant withdrawal symptoms are common

A recent systematic review identified 24 studies that evaluate withdrawal symptoms from antidepressants, including randomised controlled trials (RCTs), observational studies and survey data (Davies Reference Davies and Read2019). Given the paucity of data in this area of psychiatry, all studies are of interest but double-blind RCTs are particularly instructive because they include standardised evaluation of symptoms and allow greater causal attribution. One representative and well-conducted RCT published by Rosenbaum and colleagues in 1998 is worth examining in detail because of its high quality and reliability: it uses several measures of withdrawal effects and symptom scales, allowing direct comparison in the same group of patients in a double-blind procedure. Patients (n = 242) with remitted depression who were treated with fluoxetine, sertraline or paroxetine for, on average, 11 months, had their treatment interrupted by either placebo or their usual antidepressant in a double-blind manner for 5–8 days (Rosenbaum Reference Rosenbaum, Fava and Hoog1998). Nocebo effects – when a negative outcome occurs owing to the belief that an intervention will cause harm, the opposite of the placebo effect – were rendered less likely in this study as there were two possible time periods where placebo substitution could occur in a double-blind manner.

In this study, 14% of participants on fluoxetine, 60% of those on sertraline and 66% of those on paroxetine reported four or more symptoms on the DESS (Table 1), the criterion for experiencing a ʻdiscontinuation syndrome' (Rosenbaum Reference Rosenbaum, Fava and Hoog1998), on treatment interruption. As fluoxetine has a long elimination half-life (7–15 days), 5–8 days was probably too short a period to demonstrate withdrawal effects from this medication, leading to an underestimation of their frequency. Participants who did not have their antidepressants disrupted did register small increases in withdrawal symptoms (calculated weighted average: 2.3 symptoms on the DESS), consistent with small nocebo effects. Notably, the numbers of symptoms detected in participants interrupted from sertraline or paroxetine were much greater: 5.7 and 7.8 respectively. Withdrawal symptoms in the sertraline and paroxetine groups resolved after their antidepressants were reinstated (unbeknownst to the participants), consistent with a genuine pharmacological effect, rather than a nocebo effect. Furthermore, the fluoxetine group might approximate a ‘drug continuation’ group because of the drug's long half-life, so 14% might represent the nocebo withdrawal incidence – smaller than that seen for sertraline and paroxetine.

TABLE 1 The 20 most common withdrawal symptoms reported by people stopping antidepressants^a

a These data (rounded to two significant figures) are derived from Rosenbaum et al (Reference Rosenbaum, Fava and Hoog1998).

b Physical symptoms.

In a systematic review combining six RCTs with five naturalistic studies and three large surveys, 27–85% of patients were found to have experienced withdrawal symptoms on stopping an antidepressant (mostly SSRIs and SNRIs), with a weighted average of 56% (Davies Reference Davies and Read2019). Restricting the analysis only to double-blind RCTs from this review, the incidence of withdrawal effects is 53.9% (6 RCTs, 731 participants). There was also some variation in incidence rates between different antidepressants in these RCTs, although the number of studies was small for many drugs: paroxetine (58.8%, 3 RCTs), fluoxetine (50%, 3 RCTs), sertraline (59%, 3 RCTs), citalopram (70%, 1 RCT), escitalopram (27%, 1 RCT) and venlafaxine (71%, 1 RCT). Another systematic review reported withdrawal symptoms for 35–78% of people who stopped venlafaxine (Fava Reference Fava, Benasi and Lucente2018). Another study not captured in the review did not detect a withdrawal syndrome for agomelatine following short-term use (Montgomery Reference Montgomery, Kennedy and Burrows2004).

Nocebo effects (that is, withdrawal effects experienced when placebo is discontinued or when an antidepressant is continued) are worth examining in order to compare with the single-arm frequencies provided for antidepressants. In the six RCTs captured by the Davies & Read review (Davies Reference Davies and Read2019) and the five RCTs reviewed by Baldwin (Baldwin Reference Baldwin, Montgomery and Nil2007), six studies included such a ‘nocebo group’. In three studies a placebo was stopped, with incidence rates of withdrawal of 12.2% (n = 123), 1.9% (n = 116) (Baldwin et al (2004b) and Lader et al (2004) respectively, cited in Baldwin et al (Reference Baldwin, Montgomery and Nil2007)) and 13.5% (n = 52) (Oehrberg Reference Oehrberg, Christiansen and Behnke1995). In two studies an antidepressant was continued while patients were masked: in Montgomery et al (Reference Montgomery, Durr-Pal, Loft and Nil2003) (cited in Baldwin et al (Reference Baldwin, Montgomery and Nil2007)) 9.2% of 125 patients met criteria for withdrawal. In the other study (Zajecka Reference Zajecka, Fawcett and Amsterdam1998), 75% of 299 patients continued on fluoxetine reported one or more withdrawal symptom, although this is difficult to compare with other studies which used thresholds of four or more symptoms to detect a withdrawal syndrome. Lastly, stopping fluoxetine, which might be considered a measure of continuation because of its half-life, yielded an incidence of withdrawal symptoms of 14% (Rosenbaum Reference Rosenbaum, Fava and Hoog1998).

Excluding the outlier study for fluoxetine continuation (with an atypical definition of withdrawal syndrome – one symptom) the weighted average of the nocebo effect in these studies is 11.8%. Subtraction of this rate from the overall detected rate in RCTs (53.9% − 11.8%) might therefore give a more reasonable estimate of 42.1%. However, it should be kept in mind that most studies did not measure the severity of withdrawal effects and it is not clear that ‘dizziness’, for example, reported in nocebo conditions is equivalent to that in antidepressant withdrawal (where patients have been referred to emergency departments for investigation of stroke) (Haddad Reference Haddad, Devarajan and Dursun2001). Nevertheless, it is probably the case that a small minority of withdrawal is explained by nocebo effects.

Critical authors recalculated the incidence of withdrawal effects as 44%, when including further studies conducted by a drug company, published as supplements in journals (Jauhar Reference Jauhar and Hayes2019b). The lowest value obtained by that paper, of 31% (for RCTs alone), still represents 2.4 million people at risk of withdrawal in England. A further critique only analysed the five drug company studies, in which patients were treated, on average, for only 12 weeks and in which the authors subtracted out a high estimate for nocebo effects (Jauhar Reference Jauhar, Hayes and Goodwin2019a). The arbitrary study selection, short duration of treatment and idiosyncratic transformation of values in this study make its results difficult to interpret in relation to the general population of people using antidepressants (more than half of whom have been on antidepressants for more than 2 years) (Johnson Reference Johnson, Macdonald and Atkinson2012).

Withdrawal symptoms can be severe and/or long-lasting

A systematic attempt to assess the average severity and duration of withdrawal effects was limited by the lack of studies assessing these dimensions (Davies Reference Davies and Read2019). As regards symptom severity, only data from four large surveys were available. These indicated that about half of patients (46%) who reported withdrawal effects from antidepressants reported that these were ‘severe’, selecting the highest option on a Likert scale used to indicate severity in these studies (Davies Reference Davies and Read2019). These data were not derived from a representative sample and this limits their general applicability: those motivated to answer may have a worse than average experience of withdrawal, although 85% of patients in one included survey felt that antidepressants were helpful to them (Read Reference Read and Williams2018), indicating that negative views towards this class of medication were not widespread. Notably, patients in these surveys did have a duration of use of antidepressants (mode >3 years) similar to the wider population of antidepressant users in England (Johnson Reference Johnson, Macdonald and Atkinson2012).

Ten of the included studies evaluated duration of withdrawal symptoms. These studies are vulnerable to bias themselves – for example, studies derived from patient self-report on online forums – and did not provide enough consistency for a weighted average to be calculated. However, in seven out of the ten studies the average duration of withdrawal symptoms was greater than the 2 weeks previously supposed to be typical of withdrawal effects (Davies Reference Davies and Read2019). Indeed, the mean duration of withdrawal symptoms was 18 months in one study (after, on average, 5 years of use) (Stockmann Reference Stockmann, Odegbaro and Timimi2018) and in another (with 75% of patients using antidepressants for more than 2 years) more than one-third of patients experienced withdrawal symptoms for more than a year (Davies Reference Davies, Pauli-Jones and Montagu2018), suggesting that at least for some people, and especially those with longer-term use, withdrawal symptoms may be long-lasting.

It is commonly thought that protracted withdrawal symptoms, lasting months or years, are pharmacologically implausible; correspondingly, sometimes patients are told that long-lasting symptoms could not be drug-related because it is ‘out of their system’. This misconstrues the pathophysiology of withdrawal symptoms. Repeated use of psychotropic medication causes adaptations in the brain and body; these can be thought of as resetting of the homeostatic set-point for neurotransmitters (Hyman Reference Hyman and Nestler1996; O'Brien Reference O'Brien2011), including downregulation of serotonin receptors observed in positron mission tomography (PET) studies (Meyer Reference Meyer, Kapur and Eisfeld2001; Haahr Reference Haahr, Fisher and Jensen2014). When the drug dose is reduced, the mismatch between what the system ‘expects’ and the input from the drug is experienced as withdrawal symptoms. The elimination half-life of the drug is important: the quicker it is eliminated the sooner withdrawal symptoms will start. However, the most important factor is the length of time taken for adaptations to the presence of the drug to resolve (Reidenberg Reference Reidenberg2011). There has been little research into this area regarding antidepressants but, for example, it has been shown that adaptations to antipsychotics in animals can persist for more than a human-equivalent year (Horowitz Reference Horowitz, Jauhar and Natesan2021). It is plausible that after long-term use of antidepressants these adaptations may persist for many months or longer, consistent with the experience of protracted withdrawal symptoms (Hengartner Reference Hengartner, Schulthess and Sorensen2020a).

Overall, although further research in this neglected area would be welcome it can be concluded that between a third and a half of patients will experience withdrawal symptoms, a portion of these reactions will be severe (with half possibly being an overestimate) and for some individuals withdrawal symptoms will last months or years. The RCPsych position paper, following this evidence, highlighted that withdrawal effects can be ʻsevere and long-lasting' in some people (Royal College of Psychiatrists 2019: p. 3). Further studies will need to survey representative populations of antidepressant users or conduct prospective discontinuation trials that carefully measure incidence, severity and duration of withdrawal effects to help refine our understanding of what characteristics influence risk of withdrawal (e.g. duration and dose of use, type of antidepressant) to help better inform patients of the risks.

Withdrawal symptoms of a psychological nature are common

The psychological symptoms that have been attributed to antidepressant withdrawal are shown in Fig. 1 (Rosenbaum Reference Rosenbaum, Fava and Hoog1998; Fava Reference Fava, Gatti and Belaise2015; Cosci Reference Cosci and Chouinard2020). Psychological withdrawal symptoms are common – for example, in the Rosenbaum study, four out of the five most commonly experienced withdrawal symptoms were psychological: worsened mood, irritability, agitation and confusion. The possibility remains that these symptoms might result from genuine relapse of the patients’ original condition. However, some aspects of the study make this interpretation less likely: the psychological symptoms had onset (and resolved) at the same time as physical symptoms clearly not characteristic of depression (such as dizziness, headache, nausea, sweating, muscle aches and an unsteady gait) and resolved swiftly (within a week following antidepressant recommencement (see in the following)) (Rosenbaum Reference Rosenbaum, Fava and Hoog1998). Patients report that psychological symptoms of withdrawal can be severe and persist for long periods, and can be distinguished from their original condition (ʻI've never felt this before') (Framer Reference Framer2021).

FIG 1 Psychological symptoms of antidepressant withdrawal.

Most of the listed symptoms are derived from the Discontinuation-Emergent Signs and Symptoms checklist (Rosenbaum Reference Rosenbaum, Fava and Hoog1998). Symptoms derived from other sources are referenced individually: a, Fava et al (Reference Fava, Gatti and Belaise2015); b, Cosci & Chouinard (Reference Cosci and Chouinard2020); c, Valuck et al (Reference Valuck, Orton and Libby2009); d, Rusconi et al (Reference Rusconi, Carlone, Muscillo and Piccione2009).

Distinguishing withdrawal from relapse in clinical practice

It has been proposed that there are three ways that withdrawal can be distinguished from relapse: timing, duration and response to medication reinstatement (Haddad Reference Haddad and Anderson2007). Haddad et al outline that withdrawal symptoms usually occur days after reducing or stopping an antidepressant, whereas relapse might be expected to occur weeks, months or years later. There is some variability in this – fluoxetine has a half-life of 7–15 days and so withdrawal symptoms might be delayed for weeks (Zajecka Reference Zajecka, Fawcett and Amsterdam1998). It has also been reported that symptoms of antidepressant withdrawal can be delayed even after short half-life antidepressants are stopped, although the mechanism is not well understood (Hengartner Reference Hengartner, Schulthess and Sorensen2020a). A short duration (1 day to 3 weeks) was previously thought to distinguish withdrawal symptoms from relapse; however, we now recognise that withdrawal symptoms can last for months or even years in some individuals (Davies Reference Davies, Pauli-Jones and Montagu2018; Stockmann Reference Stockmann, Odegbaro and Timimi2018; Hengartner Reference Hengartner, Schulthess and Sorensen2020a) and so this criterion is no longer as useful. Lastly, another means of distinction is the time taken for symptoms to resolve on reinstatement of the antidepressant: for withdrawal symptoms this will be more rapid than for relapse, as demonstrated by resolution within 1 week in one study (Rosenbaum Reference Rosenbaum, Fava and Hoog1998). This also matches clinical experience – when antidepressants are recommenced soon after cessation. However, reinstatement may be less successful when it is delayed for weeks or months after the onset of withdrawal symptoms (Hengartner Reference Hengartner2020b).

There are also distinctive symptoms that mark out withdrawal symptoms from relapse. This is particularly important: physical and psychological symptoms that co-occur are more likely to be withdrawal symptoms. For example, if a patient were to experience a surge of anxiety and lowered mood on stopping an antidepressant, and this was also accompanied by nausea, vomiting or dizziness and electric shock sensations (‘zaps’, a commonly experienced neurological symptom of withdrawal affecting the head, and sometimes the limbs), we would much more readily conclude that these are withdrawal symptoms rather than relapse of a depressive condition. Although a bout of gastroenteritis co-occurring might explain some of these symptoms, it is more likely that one condition will cause several symptoms rather than that several conditions are co-occurring, using Occam's razor (Wildner Reference Wildner1999). Some symptoms are so distinctive – such as the electric ‘zaps’ in the head, often experienced on lateral eye movements, which can persist for months or years (Papp Reference Papp and Onton2018) – that they could be considered pathognomonic of antidepressant withdrawal. Physical symptoms typical of antidepressant withdrawal are shown in Fig. 2.

FIG 2 Physical symptoms of antidepressant withdrawal.

Most of the listed symptoms are derived from the Discontinuation-Emergent Signs and Symptoms checklist (Rosenbaum et al, Reference Rosenbaum, Fava and Hoog1998). a, symptoms derived from Cosci & Chouinard (Reference Cosci and Chouinard2020).

The timescale and pattern of withdrawal symptoms can also be distinctive. After a small reduction (e.g. 10% of the most recent dose) of an antidepressant, as currently suggested in RCPsych guidance (Burn Reference Burn, Horowitz and Roycroft2020), withdrawal symptoms normally follow a wave-like pattern similar to withdrawal effects from other drugs: onset a few days after the reduction, worsening, reaching a peak of intensity normally within a couple of weeks, before lessening in intensity and finally resolving usually within a fortnight or month (although with great variability among patients) (Fig. 3). Relapse or recurrence does not follow this clear wave pattern. In practice, careful monitoring of patients’ withdrawal symptoms makes this pattern evident and allows delineation of symptoms from relapse. The major distinguishing features between withdrawal and relapse are summarised in Table 2.

FIG 3 An approximate representation of withdrawal symptoms following a small reduction in antidepressant dose.

Note the wave-like properties of the relationship, whereby symptoms begin a few days after dose reduction and increase in intensity to reach a peak several days later, before lessening in intensity and resolving. A relapse is likely to be much more delayed in onset (although some withdrawal symptoms can also be delayed) and to not have this clear crescendo–decrescendo pattern over time. Note that bigger reductions in dose can lead to withdrawal symptoms that take much longer to resolve (including months or years) in some patients.

TABLE 2 Distinguishing features between antidepressant withdrawal symptoms and relapse of an underlying condition

Of course, the physician should continue to be vigilant for genuine relapse of the patient's underlying condition, which may come on weeks or months after treatment is stopped and may have characteristics that are quite typical for the patient.

Withdrawal symptoms are likely mistaken for relapse in clinical trials

We now explore whether misdiagnosis of withdrawal symptoms as relapse can occur in clinical trials aimed at measuring the relapse prevention properties of antidepressants. The NICE guidelines on relapse prevention in depression (National Institute for Health and Care Excellence 2009), and the recent draft update, are primarily based on a meta-analysis of discontinuation trials by Geddes and colleagues (Geddes Reference Geddes, Carney and Davies2003), and other similar trials. In these trials, individuals with depression who have remitted on antidepressants are randomised to either continue taking antidepressants (the maintenance group) or to stop taking antidepressants and be switched to placebo (the discontinuation group). These two groups are then monitored with depression rating scales in order to detect relapse. The difference in relapse rates between the patients maintained on antidepressants and those discontinued is interpreted as the ability of the antidepressants to prevent relapse. The influential meta-analysis of these studies found that relapse rates were 41% in the discontinuation group and 18% in the maintenance group, which has been interpreted as an absolute risk reduction of 23% for antidepressants, or a relative risk reduction of about half (56%), findings similar to other meta-analyses (Glue Reference Glue, Donovan and Kolluri2010; Borges Reference Borges, Chen and Laughren2014). There are, however, several lines of evidence to suggest that some of the relapses diagnosed in these trials are misdiagnoses of withdrawal symptoms.

Rate of tapering employed in these studies

In these 31 studies, the most common method of tapering individuals from their antidepressants (which they had used for months or years) to placebo was abrupt cessation and the weighted mean of taper duration was 5 days. It has been demonstrated that tapering over 2 weeks does not reduce the risk of withdrawal symptoms compared with abruptly stopping (Baldwin Reference Baldwin, Cooper, Huusom and Hindmarch2006; Tint Reference Tint, Haddad and Anderson2008). The latest guidance suggests that for people who have been taking antidepressants for more than a few weeks, tapering should be at a rate that is tolerable to each individual, which can be months or years, likely depending on several factors, including the duration of previous treatment (Horowitz Reference Horowitz and Taylor2019; Burn Reference Burn, Horowitz and Roycroft2020; NICE Reference National Institute2021), so the tapering rate employed in the studies summarised in this meta-analysis are, by modern standards, rapid. The risk of withdrawal symptoms in these discontinuation studies is therefore considerable.

Withdrawal symptoms overlap with all commonly used depression rating scales

Change on depression rating scales is not specific for the detection or measurement of depression, given their measurement of a wide variety of symptoms. Withdrawal effects, both psychological and physical, overlap with the domains measured in depression symptom scores, including the two most commonly used symptom scales, the Hamilton Rating Scale for Depression (the HRSD or HAM-D) and the Montgomery–Åsberg Depression Rating Scale (MADRS) (Table 3). There are few categories in these scales that do not have corresponding antidepressant withdrawal symptoms.

TABLE 3 Symptom domains in the MADRS and HRSD and overlapping withdrawal symptoms derived from the DESS or other authoritative sources

MADRS, Montgomery–Åsberg Depression Rating Scale (Montgomery Reference Montgomery and Åsberg1979); HRSD, Hamilton Rating Scale for Depression (Hamilton Reference Hamilton1960); DESS, Discontinuation-Emergent Signs and Symptoms checklist (Rosenbaum Reference Rosenbaum, Fava and Hoog1998); n.a., not applicable.

a Cosci & Chouinard (Reference Cosci and Chouinard2020).

b Framer (Reference Framer2021).

c Valuck et al (Reference Valuck, Orton and Libby2009).

d Hengartner et al (Reference Hengartner2020b).

e Reisman (Reference Reisman2020).

Furthermore, withdrawal symptoms have been clearly demonstrated to register as significant increases on depression rating scales. For example, in the Rosenbaum et al (Reference Rosenbaum, Fava and Hoog1998) study (Fig. 4) there are larger changes in depression scores for paroxetine (often implicated as the antidepressant associated with the most severe withdrawal syndrome) than sertraline, and minimal changes on all scales for fluoxetine. Both DESS symptoms and depression scores on the MADRS and HRSD reduced rapidly when antidepressants were reinstated (Fig. 4), consistent with withdrawal scores driving the increase in depression scores.

FIG 4 Withdrawal and depression symptom scores from the study by Rosenbaum et al (Reference Rosenbaum, Fava and Hoog1998).

Mean withdrawal symptom and depression scores for patients treated with (a) fluoxetine, (b) sertraline or (c) paroxetine before placebo substitution (‘before’), during placebo substitution (‘during’) and 1 week after reinstatement of the original antidepressant (‘after’). Withdrawal symptom were identified on the Discontinuation-Emergent Signs and Symptoms checklist (DESS) and depression was rated on the Montgomery–Åsberg Depression Rating Scale (MADRS) and Hamilton Rating Scale for Depression (HRSD). Note the close temporal relationship between onset and resolution of withdrawal symptoms and changes in depression scores.

Withdrawal symptoms are severe enough to meet the threshold for depression diagnosis on depression rating scales

It has been shown that withdrawal symptoms increase scores on depression rating scales such as the HRSD (Table 4) to a degree that would meet the criteria for diagnosis of relapse in studies for a portion of patients (Rosenbaum Reference Rosenbaum, Fava and Hoog1998). The average HRSD score at baseline (while on antidepressants) in this study was about 7 points. The usual cut-off for depression relapse in discontinuation studies are HRSD scores >17, so an increase of 8 or more points is likely to meet the definition of relapse for a proportion of individuals; more so for those whose HRSD score increases by 10 or more points. From Table 4 it can be seen that about a third to a half as many patients as met criteria for ‘discontinuation syndrome’ would have met criteria for relapse in a discontinuation trial. We may consider fluoxetine to be a form of control group in this study (as there was not a significant increase in withdrawal symptoms for this drug): subtracting out the proportion of participants who experienced an increase in HRSD score of ≥8 points on the 8-item HRSD or ≥10 points on the 28-item HRSD on discontinuing fluoxetine (due to either natural variation or the nocebo effect) would slightly reduce the proportion of patients on sertraline or paroxetine who met these thresholds for pharmacological reasons. Recorded relapse rates in a discontinuation trial that observes participants for months may have been even greater than this because withdrawal symptoms can increase over time – perhaps especially relevant for the fluoxetine group.

TABLE 4 Patients who met criteria for ‘discontinuation syndrome’^a and experienced an increase in score on the Hamilton Rating Scale for Depression for different antidepressants

HRSD-28 and HRSD-8, 28-item and 8-item Hamilton Rating Scale for Depression.

a Discontinuation syndrome: ≥4 symptoms on the Discontinuation-Emergent Signs and Symptoms checklist (DESS).

Source: Rosenbaum et al (Reference Rosenbaum, Fava and Hoog1998).

Discontinuation trials do not specifically measure withdrawal symptoms to distinguish them from relapse

Although there are difficulties in distinguishing withdrawal effects from genuine relapse in antidepressant discontinuation studies, this distinction could be achieved if, alongside depression rating scales used to detect relapse (e.g. the HRSD), withdrawal symptom scales (e.g. the DESS) were also employed. In this case participants who experience significant withdrawal effects could be excluded from assessment of relapse. However, antidepressant discontinuation trials did not usually consider withdrawal symptoms – none of the trials in the Geddes et al (Reference Geddes, Carney and Davies2003) review measured withdrawal symptoms specifically and only ten mentioned the possibility of withdrawal confounding the detection of relapse (mostly dismissing the possibility).

Relapse in the discontinuation arm occurs in the time period when withdrawal effects are most likely

Relapses cluster in the period just after participants are switched to placebo in the discontinuation arm of antidepressant discontinuation studies (Hengartner Reference Hengartner2020b). One analysis found that the entire difference in relapse rate between maintenance and discontinuation arms occurred in the first 6 weeks after randomisation, the period when withdrawal effects are most likely (Récalt Reference Récalt and Cohen2019). Interestingly, this effect was not evident for agomelatine, a medication thought not to be associated with a withdrawal effect. A more recent analysis has found that 70% of the difference between maintenance and discontinuation arms in antidepressant discontinuation trials (average duration 39 weeks) submitted to the US Food and Drug Administration occurred in the first 12 weeks, more than twice the proportion that would be expected if the prophylactic effect of antidepressants were ‘evenly spread’ throughout the follow-up period (Hengartner Reference Hengartner and Plöderl2021).

There are two points of view on whether the prophylactic effects of antidepressants should be ‘evenly spread’ throughout the follow-up period – that is, that antidepressants should be equally protective in the first few weeks as they are later in follow-up. Depression is normally conceptualised as an episodic condition in which antidepressants are thought to prevent relapse. If the antidepressants are removed, and withdrawal effects are not pertinent, then the natural history of the episodic disorder should be exposed: there should be no reason why patients should relapse at the same time if they each have their own pattern of episodic relapse. It follows then that if there is a disproportionate preponderance of relapses soon after stopping antidepressants then it is likely that the consequences of the process of stopping itself (i.e. withdrawal effects) is the plausible cause. This has, for example, been demonstrated for lithium where individuals with bipolar disorder stopping lithium were seven times quicker to relapse than those who had never been treated with lithium (Suppes Reference Suppes, Baldessarini and Faedda1991).

The alternative explanation for early relapses is that patients are prone to early relapse because of the removal of the stabilising effect of the antidepressants. One version of this rationale argues that partially treated patients relapse soon after medication is discontinued, leading to a high-risk period for relapse soon after stopping (Jauhar Reference Jauhar, Hayes and Goodwin2019a). However, participants in these discontinuation studies are selected for having remitted and meta-analyses do not find a difference in relapse rates for participants with longer stabilisation periods before discontinuation (Geddes Reference Geddes, Carney and Davies2003; Glue Reference Glue, Donovan and Kolluri2010; Borges Reference Borges, Chen and Laughren2014), which would be thought to provide more complete treatment. On the other hand, the reduction in relative risk of relapse after antidepressant cessation was similar in magnitude between 12–36 months and 0–12 months in the Geddes review and this is not consistent with withdrawal effect confounding (although it must be noted that the 12–36 month period only included 6 studies, with 35 relapses, so this captures relatively few patients). Notably, agomelatine, which has not been found to be associated with a withdrawal syndrome, fails to demonstrate significant relapse prevention properties when unpublished data are taken into account (Koesters Reference Koesters, Guaiana and Cipriani2013).

Overall, there are two possible reasons for worsening symptoms when stopping medication: either uncovering of the underlying disorder or withdrawal effects (although withdrawal effects might precipitate genuine relapse this should be seen as a withdrawal-related effect). It seems to us that if relapses are clustered soon after stopping then this seems to fit more with a withdrawal effect. This dilemma could be resolved by stopping antidepressants gradually enough (i.e. over months or longer) to prevent withdrawal effects and observing the pattern of relapse.

The degree to which antidepressants prevent relapse is uncertain

Withdrawal symptoms are common: 53.9% in RCTs from the recent systematic review (Davies Reference Davies and Read2019) and 31% in very conservative estimates (Jauhar Reference Jauhar and Hayes2019b). Rates of relapse of depression after cessation of antidepressants are less certain. The Geddes meta-analysis found relapse rates of 41% in the discontinuation arm, 23% greater than in the maintenance arm. However, as outlined above, some proportion of these relapses are likely to be due to withdrawal effects: the original authors themselves concluded that ʻwe cannot exclude the possibility' (Geddes Reference Geddes, Carney and Davies2003). It is hard to estimate what proportion of these relapses might be due to withdrawal confounding but the data presented in Table 4 suggest that perhaps one-third to one-half of the 31–53.9% of patients who experience withdrawal effects (as estimated by different analyses of RCTs) may reach the threshold required for detection of relapse given an 8- or 10-point increase in HRSD-8 or HRSD-28 score (Rosenbaum Reference Rosenbaum, Fava and Hoog1998). This suggests that approximately 10–27 percentage points of relapses, or a substantial proportion of the difference in relapse rates between the maintenance and discontinuation arms in these studies, may be attributable to withdrawal – although there is considerable uncertainty in these estimates.

Therefore, the relapse prevention studies summarised in Geddes et al (Reference Geddes, Carney and Davies2003) may be demonstrating, at least partially, the effect of abrupt (or near-abrupt) cessation, through either withdrawal effects or destabilising participants’ conditions rather than the benefits of continuing treatment (Baldessarini Reference Baldessarini and Tondo2019). Further study is required to confirm the magnitude of the relapse prevention properties of antidepressants by evaluating the degree to which withdrawal confounds relapse detection in these studies. The major lines of evidence suggesting that withdrawal effects confound detection of relapse in antidepressant withdrawal studies are summarised in Box 2.