Antidepressant Withdrawal: Two One-Thirds, Two Different Claims

Two numbers in a new antidepressant-withdrawal paper look almost interchangeable: 33% and 32%. One describes symptom increases across observation intervals. The other estimates how many participants experienced pharmacological withdrawal over six months. Between them sit assumptions about causation, background symptoms, and whether a calculation made across intervals can be transferred to people.

That distinction matters in a consultation. A psychiatrist reviewing deterioration during a taper needs to understand what happened to this patient. A headline about "one-third" cannot decide whether the next step should be another reduction, a pause, reinstatement, or treatment for recurrent illness.

Published September 24 in Therapeutic Advances in Psychopharmacology, the study gives a reason to look more closely at the lower-dose phase. It also shows why stopping antidepressants remains an unusually difficult measurement problem: the medication, the underlying condition, and the act of changing treatment can all produce overlapping symptoms.

Rennwald and colleagues followed 32 adults in Swiss primary care for 26 weeks. Participants were in stable remission, had taken antidepressants for at least six months, and planned to discontinue. Median prior treatment was 29 months. Eighteen of the 32 took escitalopram.

The investigators related symptom changes to dose changes, expressing each medicine's dose relative to a study-defined minimum effective dose. Their own example makes the reference point concrete: escitalopram 10 mg counted as 100%, 7.5 mg as 75%, and 5 mg as 50%. The threshold refers to the dose reached, not the percentage removed from the starting prescription.

In the fully adjusted model, clinically relevant symptom increases occurred in an estimated 33% of intervals involving reductions below 75% of that reference dose, 8% of intervals involving reductions above it, and 13% of intervals without reductions. Compared with no reduction, the below-threshold odds ratio was 3.19 (95% CI 1.37 to 7.40). Above the threshold, it was 0.60, with a much wider interval of 0.17 to 2.18.

The three-way comparison is the contribution. Symptom increases were associated with where reductions ended on the dose range. The above-threshold estimate did not differ significantly from no reduction, but its width leaves room for several possibilities, and it cannot establish that higher-dose cuts are harmless. Its point estimate also sits below the no-reduction figure. Chance, easing adverse effects, and early symptom declines are possible explanations. The comparison does not establish that dose reduction protects against symptom increases.

The paper adds a striking detail: reductions were proportionally larger at higher doses. Above the 75% boundary, the average reduction was 67% of the previous dose. Below 75%, 50%, and 25%, the corresponding averages were 32%, 20%, and 12%. These lower-dose categories overlap and were not randomized groups.

The pattern challenges the assumption that a smaller proportional cut must be easier merely because it is smaller. It does not isolate dose position as the cause. Participants and clinicians chose the reductions, and earlier symptoms could have shaped later choices.

Only 12 participants received a specific tapering plan. Three stopped abruptly, and ten had reached zero by week eight. By week 26, fifteen had discontinued. This was a heterogeneous, real-world discontinuation process, not a comparison of standardized slow and fast schedules.

The study therefore speaks to the lower-dose phase, including reductions to zero. It cannot say how often the final milligrams become difficult during a carefully individualized, prolonged taper, or whether any particular schedule prevents that difficulty.

Recruitment fell far short of the planned 400 participants. Repeated measurements yield more information than a single assessment per person, but they do not turn 32 participants into hundreds of independent patients. The small sample, the predominance of escitalopram, and the number of adjustment variables make replication especially important.

The design still has real value. It observes long-term users trying to stop in ordinary care, a process that short, tightly controlled trials characterize poorly. Its strength is longitudinal observation of that process. Its weakness is separating the explanations for what it observes.

The investigators used an adapted 43-item Discontinuation-Emergent Signs and Symptoms Scale (DESS). Each symptom was rated from zero to three over a two-week recall window. A total increase of at least four points counted as clinically relevant: four to seven was classified as moderate, and eight or more as severe.

These are study definitions. A four-point increase can come from worsening of existing symptoms as well as new ones. It is not the same as four new symptoms, a clinician-adjudicated withdrawal syndrome, or any particular degree of functional impairment.

A total can also hide changes moving in opposite directions. As the authors note, adverse effects may ease as the dose falls while withdrawal symptoms emerge, leaving the sum unchanged despite a new problem. Conversely, a rising total can reflect an unrelated stressor or illness. The items and their timing matter even when the overall score looks reassuring.

Assessments took place at baseline and at weeks two, four, six, eight, sixteen, and twenty-six. The early gaps were two weeks; the final two were eight and ten weeks. A two-week recall at the end of a ten-week interval cannot capture every symptom or dose change within it, so the interval estimates do not represent a constant weekly risk.

The affective subscale deserves attention. It covered anxiety, irritability, sudden mood worsening, panic, tearfulness, agitation, and related experiences. Below-threshold reductions were associated with a greater increase than no reduction, with an adjusted coefficient of 0.92 points and a standardized mean difference of 0.27. The neurosensory subscale showed a larger standardized difference of 0.35. That accompanying sensory signal makes a withdrawal interpretation of the emotional symptoms worth taking seriously, though it does not settle their cause.

The finding brings emotional deterioration into the withdrawal question. It does not show that every affective change was withdrawal, nor does it exclude a depressive episode. A symptom scale records experiences; deciding their cause takes evidence beyond the label printed above its items.

A companion report by the same investigators carries a stronger title: Acute Affective Symptoms during Antidepressant Tapering Indicate Withdrawal Reactions Rather than Relapse. Its abstract appears to describe the same cohort, and it acknowledges that causality cannot be inferred. It should not be treated as independent replication, and the September paper does not resolve that diagnostic distinction.

The 33% figure comes from the adjusted interval analysis. Separately, 17 of the 32 participants (53%) had at least one clinically relevant increase during an interval involving a below-threshold reduction. Eleven participants were classified as moderate and six as severe.

The authors then estimated the pharmacological share of those events. They treated the 13% no-reduction frequency as background against the 33% below-threshold frequency. Because 13 divided by 33 is about 0.39, they attributed roughly 60% of the risk to pharmacological withdrawal.

Applying that 60% factor to the cumulative 53% gives approximately 32%. Applying it to the 19% classified as severe gives approximately 11%. The authors describe the 32% as moderate-to-severe withdrawal, so it already includes the severe subgroup; the two figures should not be added together.

The arithmetic is simple, but the attribution asks more of the data. No-reduction intervals must represent the background symptom changes that would have occurred during reduction intervals. Yet people may hold a dose because symptoms have appeared, and delayed effects of an earlier reduction may surface while the dose is unchanged. The authors acknowledge the latter.

Transferring an interval-based attributable fraction to a person-based cumulative proportion adds a further assumption. A participant with several affected intervals is counted once in the cumulative outcome. Unless we know how pharmacological and background events cluster within individuals, the interval ratio cannot tell us what proportion of people experienced pharmacological withdrawal.

That is why the two similar numbers need different descriptions. The 33% is an adjusted frequency of symptom increases across specified intervals. The 32% is a derived cumulative estimate that depends on the authors' attribution model. Neither is a general incidence rate for everyone who stops an antidepressant.

The paper offers a biological rationale for the 75% threshold and reports that it fitted the data better than alternatives of 100% and 125%. But that comparison was made within the same 32-person dataset, so part of the fit may reflect this particular sample. The direct check is to specify the same boundary in advance and test it in new data.

Update, September 29, 2026: In email correspondence, corresponding author Michael Hengartner confirmed that the 75% threshold was not specified before analysis and was "determined empirically," an approach he noted is outlined in the paper, with additional analyses reported in its supplement. The clarification reinforces the need to test the threshold, fixed in advance, in independent data.

Until then, the finding supports further investigation of dose position. It does not establish 75% as a clinical line separating reductions that cause withdrawal from those that do not. The threshold also enters the study's definition of a withdrawal reaction, which required a dose below 75% of the reference dose, so the 32% and 11% estimates depend on the same boundary. For the next decision, a patient's symptoms and trajectory are more informative than which side of the cutoff their dose falls on.

Time is a further unresolved issue. Average symptom scores were highest at baseline, partly because of three unusually high scores. The largest average decreases occurred between baseline and week two, and again between weeks sixteen and twenty-six. Dose position also shifts as tapering progresses. A stronger analysis would test whether the lower-dose association survives more direct accounting for time and for influential observations. Baseline adjustment helps, but regression to the mean remains a plausible concern that the published results neither confirm nor rule out.

The mechanism is credible enough to pursue. A systematic review by Sørensen, Ruhé, and Munkholm of 17 imaging studies involving 294 participants found that serotonin-transporter occupancy rises steeply at lower antidepressant doses and approaches a plateau at higher ones. Low milligram doses can therefore retain substantial biological activity. The imaging evidence does not, however, identify an optimal taper or show that occupancy change predicts an individual's withdrawal.

The cohort builds on earlier tapering arguments by Horowitz and Taylor; Horowitz is a coauthor. Replication by independent groups is the natural next test. Comparing actual tapering strategies would answer a separate question: whether changing the schedule improves outcomes.

The wider literature offers no single uncontested rate. Henssler and colleagues' 2024 meta-analysis reported discontinuation symptoms in about 31% of people after stopping an antidepressant and 17% after stopping placebo. In a later published response to criticism, those authors called subtraction across different sets of studies a debatable approximation and emphasized the smaller, eight-percentage-point difference found within randomized comparisons.

Moncrieff and colleagues challenged the reliance on spontaneous adverse-event reports, short exposure and follow-up, and uncertain severity measures. Their 2025 reanalysis, restricted to five studies with systematic and relevant assessment, estimated that 55% experienced at least one symptom, before subtracting background effects. The original authors countered that this selection discarded substantial evidence and could overrepresent drugs with greater discontinuation effects.

Kalfas and colleagues' 2025 review took yet another approach. Its one-week DESS analysis found an average difference equivalent to about one additional symptom after antidepressant discontinuation, compared with placebo discontinuation or continued treatment. Dizziness, nausea, vertigo, and nervousness were increased, while depression symptoms were not in the relevant short-term comparisons. Most DESS trials followed participants for no more than two weeks, which limits what the review can say about later symptoms.

An average additional symptom, the percentage reporting any symptom, a severity-score increase, and a pharmacologically attributed syndrome answer different questions. Structured questioning can detect experiences that spontaneous reporting misses, while nonspecific items can also pick up background distress. Short follow-up limits what can be said about later events. These measurement choices deserve more attention than the competing headline percentages.

For the clinic, the disagreement argues for a practical discipline: be clear about which evidence informs the conversation, then return to the patient's actual trajectory. Neither a reassuring average nor a high cumulative estimate identifies the cause of today's deterioration.

The affective and neurosensory findings suggest a focused clinical question: how did the emotional change relate to the sensory change? Establish whether dizziness, imbalance, or unfamiliar sensations preceded the anxiety, accompanied it, or were absent. Then compare that sequence with the patient's previous illness. The aggregate subscale results cannot establish the order for any individual; the interview has to recover it.

Place those symptoms beside the actual exposure: medicine, formulation, doses taken, dates of reductions, missed doses, and supply interruptions. The intended taper and the doses actually taken may describe different timelines. Record function alongside symptoms, particularly when an apparently small change has made driving, working, or basic self-care difficult.

NICE lists early onset after reduction, unfamiliar symptoms, and symptoms qualitatively different from the original condition as clues favoring withdrawal, and it also recommends considering new pathology. Neither a sensory complaint nor an affective score should end that assessment.

Timing helps without providing a stopwatch diagnosis. The Royal College of Psychiatrists describes withdrawal as commonly beginning soon after a reduction and recurrence as often developing over a longer period. Fluoxetine and other delayed-onset presentations complicate the distinction. Relatively prompt improvement after reinstatement can inform the assessment, alongside the rest of the clinical picture.

The cohort's measurement gaps carry a further lesson: a retrospective score can miss the sequence that matters. Preserve dates and item-level changes between reviews, including symptoms that resolved before the next questionnaire. Because an unchanged total can conceal both easing adverse effects and emerging difficulties, ask what changed rather than relying on the direction of the sum.

The ANTLER randomized trial is essential evidence for the maintenance discussion. Among 478 UK primary-care participants who felt well enough to consider stopping long-term treatment, relapse by 52 weeks occurred in 39% of those assigned to maintenance and 56% of those assigned to discontinuation. Withdrawal symptoms were also more common after discontinuation.

The regimen matters. Participants taking citalopram, sertraline, or mirtazapine received half their usual dose for one month, then alternated that half-dose with placebo for another month before switching to placebo. Those taking fluoxetine 20 mg alternated medication with placebo for one month before switching to placebo.

ANTLER did not compare maintenance with every possible individualized taper. Its report discusses the difficulty of distinguishing withdrawal from depression, and treatment allocation became apparent to many participants. These limitations affect interpretation, but they do not erase the observed difference between randomized groups.

Two decisions therefore need their own reasoning. Whether maintenance offers protection depends on the patient's illness history, previous response, adverse effects, preferences, and risk. What explains symptoms during the current reduction also depends on exposure and symptom timing. A difficult reduction cannot carry the entire case for indefinite treatment, just as evidence of withdrawal cannot show that maintenance has no value.

The immediate plan can change while the eventual endpoint stays open. Pausing to reassess tolerability need not mean abandoning discontinuation, and continuing treatment after review can also be a reasoned choice. The objective is a defensible decision the patient understands, with a clear way to revisit it.

The study's two-week recall window does not establish an optimal two-week follow-up schedule. Nor do its later eight- and ten-week gaps justify making a symptomatic patient wait that long. NICE recommends basing monitoring frequency during reduction on the person's clinical and support needs, while assessing both withdrawal and returning depression.

For someone developing symptoms at lower doses, check whether the current review arrangement can actually inform the next planned reduction. The questions are concrete. When is that step due? What information will be available before it? How can the patient get a review if symptoms change? Shorter intervals may suit the clinical situation, but the 75% research boundary cannot set the schedule.

NICE also recommends staged reduction, attention to pharmacokinetics and treatment duration, smaller reductions as doses fall, and a pace agreed with the patient. For more severe withdrawal, its guidance includes considering reinstatement at the previous dose, followed by slower reductions once symptoms resolve.

The study's distinction between achieved dose and size of reduction is useful here, so document both. Confirm that the intended dose can be supplied and that the patient and pharmacist understand the same administration instructions. A nominally small reduction offers little reassurance if the actual exposure is uncertain.

The next note should explain what changed, the leading alternative explanations, and why the agreed action fits the present symptoms and risk. It should also record what would prompt reassessment before another reduction. That lets a later clinician evaluate the reasoning instead of inheriting an unexplained label of either withdrawal or relapse.

Severe deterioration requires assessment on its own terms. Suicidal thoughts, inability to maintain basic care, and other urgent changes should not wait on agreement about whether the cause is withdrawal or recurrent illness. NICE recommends direct assessment of suicidal ideation and urgent referral when immediate risk is considerable, and the Royal College likewise directs people who develop suicidal thoughts during reduction to seek prompt help.

For the clinician, the operational question is what level of support is needed now. "Probably withdrawal" does not answer it, and neither does "probably relapse." The explanation informs treatment, but the patient's current condition determines how urgently to assess and protect them.

Make the contingency plan concrete. Which kinds of deterioration warrant earlier contact, who receives that contact, and what happens if routine access is unavailable? A scheduled follow-up should organize care while leaving a clear route to reassessment before that date.

The September study strengthens the case for paying attention as doses fall, but it supplies neither a diagnostic threshold nor a universal taper. Its most lasting contribution may be a question it makes harder to avoid: when symptoms emerge near the end of treatment, have we examined the medication change with the same care we would give to starting treatment? The useful number is the one whose meaning survives the journey from the paper to that decision.

This article concerns planned antidepressant reduction in adults and is not an individualized dosing protocol. Its interview and documentation suggestions are editorial synthesis.

Educational Disclaimer: The Psychiatric Record provides general educational information for psychiatric and mental-health professionals. Content does not constitute medical, legal, regulatory, compliance, billing, or other professional advice; does not establish a standard of care; and is not a substitute for independent professional judgment. Appropriate assessment and treatment depend on the individual, setting, and current authoritative guidance. Patients should coordinate medication changes with their treating clinicians.