Some Predictions About the Future of “Withdrawal Studies”
What we talk about when we talk about antidepressant withdrawal
In communities organized around antidepressant withdrawal and the broader iatrogenic harm and the deprescribing ecosystem, the term “withdrawal” has acquired an extraordinary breadth and scope. It’s an instance of concept creep on steroids. The term now gets used for a week of brain zaps after stopping paroxetine all the way to years of fatigue and cognitive fog attributed to a drug discontinued long ago. Acute antidepressant withdrawal syndrome is well established at this point; it is more common and more severe than the psychiatric mainstream was willing to admit for an embarrassingly long time. But beyond acute antidepressant withdrawal, there is a lot of confusion and uncertainty, and because after decades of neglect antidepressant withdrawal is finally attracting public attention, there is a strong incentive to stuff more and more forms of medication-related difficulties into this bloated concept.
A taxonomy of post-discontinuation and withdrawal phenomena
The way “withdrawal” is used in iatrogenic harm communities, there appear to be two broad concepts being utilized.
The first is clinical: any clinical difficulty experienced following the discontinuation or dose reduction of a psychiatric medication that isn’t clearly and unequivocally a relapse of the original psychiatric problem. The time duration for “following” ranges from immediately post-discontinuation to several months later.
The second is pharmacological: any clinical difficulty that arises due to the development of neuroadaptation to the presence of a psychiatric medication followed by discontinuation or dose reduction of that medication.
The two concepts aren’t identical, but there is a tendency to assume they are two sides of the same coin. The problem is that clinical difficulties can arise post-discontinuation for a variety of reasons that have little or nothing to do with neuroadaptation, or, in other cases, with just neuroadaptation (e.g. what is experienced is psychopathology altered by neuroadaptation). The causal inference is the tricky part. And we are not yet in a position to directly observe and measure “neuroadaptation.” There are no established and validated biomarkers for this purpose.
Given the very broad nature of these definitions and the difficulties of causal inference, when someone says “withdrawal,” they could be talking about any of the following possibilities. Some of these are extensions of established pharmacology, some are speculative and hypothetical, and some are not withdrawal at all.
Acute withdrawal/discontinuation syndrome is the clearest case. The analogue would be acute alcohol withdrawal. It shows up within days of a dose reduction or discontinuation, brings a recognizable cluster of symptoms (brain zaps, vertigo, nausea, irritability, dysphoria, insomnia, agitation), and is generally self-limiting over days to weeks. Brain zaps in particular are close to pathognomonic when they are present; they don’t show up in depression or anxiety, and you can’t readily explain them away as relapse. Acute withdrawal also tends to respond quickly to resuming the antidepressant at the prior dose.
Symptom rebound is a cousin but conceptually distinct. The amplified, transient return of the original target symptoms past their pre-treatment baseline. Rebound insomnia after stopping a benzodiazepine is the textbook case. By definition it is time-limited; it reflects homeostatic overshoot. For SSRIs, this is usually rebound anxiety and dysphoria. Symptom rebound can possibly occur with or without acute withdrawal syndrome.
Relapse of original psychiatric disorder (e.g. clinical depression or anxiety) triggered by acute discontinuation or rebound. The discontinuation/rebound destabilizes the system to such a degree that a new major depressive episode or generalized anxiety syndrome develops, and requires to be treated as such because it behaves like regular depression or anxiety (it is not transient and doesn’t respond rapidly to restarting the medication)
Protracted withdrawal syndrome covers symptoms persisting for months or years, well past the acute phase. It gets progressively more difficult to say that pharmacological speaking, this is the same sort of thing as acute withdrawal; I will have more to say about it below. The analogue would be post-acute alcohol withdrawal syndrome for alcohol, but that also remains rather poorly characterized with validity issues and remains without formal recognition in diagnostic manuals.
Delayed-onset withdrawal. This is the scenario when symptoms emerge weeks or months after the last dose with no preceding acute phase. Pharmacologically speaking, I find this to be the most puzzling. It requires a mechanism that produces no symptoms during and shortly after the period when the drug is actually leaving the body and occupancy is actually changing, and intense symptoms appear weeks later that are neither relapse nor, proponents allege, better explained by a different psychopathological process.
Several other things routinely get folded into the category of “withdrawal.”
Hypothesized receptor “supersensitivity” syndromes. This is by analogy to tardive dyskinesia, where chronic receptor modulation leaves lasting changes that outlast the drug. Dyskinesias related to dopaminergic supersensitivity are well-established; the extrapolation to serotonergic systems is entirely hypothetical at this point.
Post-SSRI sexual dysfunction (PSSD). Cases of persistent genital anesthesia, anorgasmia, loss of libido continuing for months or years have also been reported to emerge after stopping the SSRI (the more usual case is emergence during treatment). Because of some reported cases of emergence post-discontinuation, it sometimes gets discussed under the withdrawal umbrella, but if these post-discontinuation reports are genuine, they are really a sort of persistent medication-induced adverse effect, not withdrawal in any traditional sense.
Persistent neurocognitive or neurological changes, claims of lasting cognitive impairment or emotional blunting long after discontinuation. The analogue would be lasting cognitive impairment from severe and prolonged alcohol use. Again, to the extent this happens (I’m not sure), this isn’t technically withdrawal, it’s medication-related injury.
Discontinuation-triggered psychiatric syndromes different from the original psychiatric diagnosis. This would include things like new-onset mania, or psychosis, or panic attacks precipitated by stopping the medication without any prior history of these problems. Discontinuation-related mania is among the better-evidenced post-discontinuation phenomena, documented in a few careful case series, but we know very little beyond that. This is a situation where the perturbation of discontinuation leads to a new psychiatric syndrome and the usual stress-diathesis considerations would apply.
Relapse or return of the original condition (without significant rebound and without withdrawal precipitating it). For episodic conditions like episodic major depression, the odds of this happening in the first few weeks of discontinuation are low, but for chronic conditions like dysthymia and generalized anxiety, symptoms can return with original intensity soon after discontinuation (similar to how controlled chronic pain would reemerge soon after stopping analgesics).
And then there is the category of things that may simply be misattributed to medication discontinuation or may have an indirect relationship to medication discontinuation. This would include functional neurological and psychosomatic symptom presentations; nocebo responses; ordinary somatic and psychological variation reinterpreted through a withdrawal lens by people primed to do exactly that, and the coincidental development of conditions like post-viral syndromes or chronic fatigue syndrome, etc.
Finally, “withdrawal” sometimes isn’t simply a clinical description anymore but also an identity. “Being in withdrawal” becomes a standing condition rather than a time-limited state, a mark of membership in the iatrogenic harm community, which is categorically unlike how the word functions in pharmacology or addiction medicine, where you go through withdrawal and come out the other side.
“Withdrawal” has been stretched from a reasonably precise pharmacological concept (neuroadaptive rebound after removal of a substance) into a catch-all, wastebasket of speculative inferences for a wide range of problems that follow any encounter with a psychiatric drug.
The withdrawal intellectuals
At this stage, let’s briefly examine how some prominent figures associated with the withdrawal discourse have approached these issues. I’m thinking of Guy and Virginie-Anne Chouinard, Giovanni Fava, Mark Horowitz, David Healy, and Adele Framer.
The Chouinards offer the most formally differentiated and defensible scheme. Their 2015 classification creates three diagnostically distinct categories, with proposed (not yet validated) criteria: new withdrawal symptoms (novel to the patient, self-limiting), rebound symptoms (the original symptoms at greater intensity, also self-limiting), and persistent post-withdrawal disorders (lasting syndromes that may reflect, hypothetically, receptor supersensitivity).
Fava embeds withdrawal in a larger framework he calls “oppositional tolerance.” On this view, withdrawal is just one expression of a complicated underlying process, i.e. the body’s compensatory adaptation to chronic antidepressant/medication exposure, pushing back against the drug’s intended effect. This creates a whole family of oppositional phenomena, e.g. loss of efficacy over time, paradoxical worsening with treatment, bipolar switching during treatment, withdrawal on stopping, persistent post-withdrawal disorders, refractoriness on reinstatement.
Horowitz (with Taylor) gives a mechanistic specific grounded in receptor occupancy. Because the relationship between dose and serotonin transporter occupancy is hyperbolic, conventional linear dose reductions produce disproportionately large perturbations at the low end. Hence proportional tapering, cut by a percentage of the current dose rather than a fixed amount.
What the occupancy curve story has not done, though, is tell us clearly what “withdrawal” means in a way that would make it reliably separable from the competing explanations. What Horowitz offers (most systematically in the 2022 paper with David Taylor in BJPsych Advancesv) is a set of heuristic distinguishing features, such as onset soon after a reduction (but not always), psychological symptoms co-occurring with distinctive physical ones like brain zaps (but not always), a prompt response to reinstatement (but not always), a “wave” pattern of onset-peak-resolution (but, you guessed it, not always). Some of these are useful pattern-recognition tools, but withdrawal is being defined negatively (by its contrast with relapse) rather than positively, with its own diagnostic criteria. The working definition reduces to something like: symptoms after dose reduction that aren’t relapse. And the criteria for “not relapse” are themselves soft and probabilistic, and inclusive of things like symptom rebound and discontinuation-triggered-relapse.
Once you grant that withdrawal can onset immediately or weeks later, can last days or years, can present with distinctive neurological signs or with symptoms indistinguishable from the previous psychiatric disorder, and can be confirmed by reinstatement response but also might not respond to reinstatement, at that point you have described a category that is far too flexible to do meaningful discriminatory work. That is not necessarily wrong; biological phenomena really can be heterogeneous, but it is diagnostically slippery and scientifically underdeveloped. Validated criteria for the differentiation simply do not exist at present.
Healy, from what I can tell, has been walking away from the “withdrawal” in favor of characterizations like “iatrogenic dysregulation.” The term “withdrawal” imports connotations from the substance abuse world; SSRIs act extensively on peripheral serotonin systems, so a neuroadaptive story centered on brain receptor recalibration may be incomplete in a fundamental way; and conditions like PSSD are not “withdrawal” but drug-induced injuries. The existence of these dysregulations and injuries, neurologically speaking, is speculative, but the move from “withdrawal” to “dysregulation” is a telling one.
Framer, who founded Surviving Antidepressants, seems to work with the organizing idea of nervous system destabilization or dysregulation, that withdrawal symptoms “indicate the nervous system is becoming destabilized,” and this destabilization is what accounts for the variety and variability of symptoms across body systems. Her description takes in acute symptoms, protracted symptoms, delayed-onset symptoms, and symptoms during tapering, all reflective of a destabilization of some sort.
What do we really know?
Once we get past acute discontinuation syndrome, the empirical evidence becomes progressively thin and causal attributions more and more speculative, and this is true no matter which side of the debate one is on.
Acute withdrawal syndrome is fairly well-established as a clinically significant phenomenon. The argument here is largely about the range of incidence and severity, not existence.
Symptom rebound is pharmacologically grounded and often clinically observed, but because few separate it from acute withdrawal, we have no clear idea of its incidence separate from aggregate withdrawal statistics. The same goes for relapse of the original psychiatric disorder (e.g. clinical depression or anxiety) triggered by acute discontinuation or rebound.
Discontinuation-related mania and other psychiatric syndromes have been reported in case series (Goldstein et al. 1999 for discontinuation-related mania, for example) and at least, based on reported cases, seem to be rare.
PSSD remains without a reliable estimate, and we know even less about PSSD onset post-discontinuation.
Protracted withdrawal syndrome… the evidence base consists of the Hengartner et al. 2020 analysis of 69 Surviving Antidepressants narratives, community surveys, and clinical observation from a handful of specialized deprescribing clinicians. There is no prospective, controlled study documenting protracted withdrawal. Nobody has followed a cohort of discontinuers for 6–12+ months with baseline assessments and a comparison group and shown that protracted withdrawal occurs at a rate above what we’d expect from chance and coincidence and other diagnostic possibilities.
We can say with reasonable confidence that some patients have persistent distressing symptoms after discontinuation that are not explained by return of the original psychiatric condition. The label “protracted withdrawal syndrome,” however, smuggles in the presupposition that it is “withdrawal” in some meaningful sense of the word “withdrawal” and a supposition of mechanistic unity that has not yet been demonstrated.
Delayed-onset withdrawal has the basic pharmacological problem I flagged earlier. What produces no symptoms during the window when the drug is clearing and occupancy is shifting, and then produces symptoms later, once the perturbation is over? It’s not impossible (something, something, cascading downstream effects), but the evidence is almost entirely retrospective self-report with uncertain causal attributions.
Does medicine offer any clean precedent for genuinely delayed withdrawal… physiological dependence where the clinical syndrome shows up well after the drug has cleared? Chronic exogenous corticosteroids suppress the HPA axis and, through prolonged ACTH suppression, cause physical atrophy of the adrenal glands. Stop the steroids and the adrenals can’t immediately resume adequate cortisol output. But the crisis (adrenal insufficiency) need not manifest with clinical symptoms right away. It may appear days to weeks later, often when some physiological stressor arrives that the atrophied glands can’t handle.
Perhaps something similar happens in the case of delayed SSRI withdrawal? The serotonergic system experiences some sort of functional loss. It is able to handle functional demands in the period after discontinuation but fails weeks or months later from the accumulated stress. What we are talking about in this sort of situation is a reduced or impaired physiological capacity secondary to prolonged drug exposure, akin to adrenal atrophy; it is conceptually distinct from classic withdrawal. As should be apparent, this is entirely speculative, and we have no idea what an analogue of adrenal atrophy would be like for the serotonin system and whether it even exists.
The basic problem underneath all this uncertainty is depressing as well as simple. The research infrastructure to study these discontinuation-related iatrogenic phenomena was never built and never funded. The studies that would settle these questions, large prospective cohorts with validated instruments and proper comparison groups, do not exist. Until they do, everyone is forced to rely on clinical observations, self-reports, extrapolations of established pharmacological principles, and unrestrained speculations.
The implausibility problem
When someone says they have been “in withdrawal” for three years, they might mean they have been tapering for three years, or they might mean the last dose was three years ago and they still have symptoms they believe to be caused by stopping the drug. The first isn’t technically withdrawal, and the second strikes most people as implausible.
“Withdrawal,” in pharmacology, is a function of neuroadaptation. When a substance is removed, there is neuroadaptive rebound; adjusted set points based on the continued presence of the substance produce dysfunction until the system readapts with new set points. This runs on timescales of days to weeks for occupancy changes, weeks to maybe months for receptor density normalization.
So what could be driving persistent symptoms years later? If the symptoms are a direct physiological consequence of medication exposure and discontinuation, the situation is that of a (hypothetical) persistent drug toxicity of some sort akin to alcohol-related cognitive deficits or some sort of (hypothetical) super-sensitivity akin to tardive dyskinesia or some sort of (hypothetical) post-viral syndrome-like picture, except it’s post-medication. These possibilities are not instances of “withdrawal” and it’s a misnomer to call them so.
And then there’s the possibility that what is happening at three years is not causally tied to the discontinued drug in the way the person believes. It is a post-viral syndrome or chronic fatigue syndrome that coincidentally developed in the weeks following antidepressant discontinuation. (Given the base rates of these conditions and antidepressant use, when we are looking at large populations, such coincidences are bound to happen.) It can also very well be some form of functional neurological or psychosomatic disorder. Or something else entirely not causally tied to the medication.
The dogmas and unknowns of hyperbolic tapering
PET studies showing hyperbolic SERT occupancy curves are well-established. The trouble is the inferential gap between the occupancy curves and the clinical symptomatology and tapering recommendations. Withdrawal symptoms are not direct consequences of changes in transporter occupancy. They are consequences of downstream intracellular and network-level changes such as altered postsynaptic receptor sensitivity, second-messenger modifications, gene expression shifts, neurogenesis effects, etc. The relationship between an occupancy change and these downstream adaptations need not be hyperbolic at all. If the downstream systems carry their own nonlinear dynamics, then the optimal taper shape is a function of some complicated composite of the occupancy curve and the downstream dose-response relationships, not of the occupancy curve by itself.
Let’s consider the asymmetry between starting a medication and stopping it. When we start a patient on 20 mg of fluoxetine, we take a brain at 0% SERT occupancy and jump it to 70-80% in a very short period of time. The brain adapts over a few weeks… nobody proposes hyperbolically titrating upward in 10% increments over months. Yet the framework implies that on the way down, a reduction from, say, 40% occupancy to 20% is a dangerously large perturbation. But 40% occupancy is pharmacologically subtherapeutic on the way up. So if the serotonergic system can absorb partial transporter blockade without meaningful functional consequence when the drug is being started (much as Parkinson’s produces no motor symptoms until 60–80% of dopaminergic neurons are gone), why would restoring the system toward its native state produce severe symptoms on the way down?
The therapeutic-threshold tells us something about the system’s functional sensitivity. If we need 70-80% occupancy to get a clinically meaningful antidepressant effect, then the system has substantial buffering capacity. It can absorb a lot of transporter blockade before the downstream signal changes in a therapeutically relevant way. Going from 0% to 50% apparently doesn’t do much of consequence.
Neuroadaptation should be proportional to the degree of functional perturbation from baseline, and if subtherapeutic occupancy produces minimal functional perturbation going up, the adaptations laid down at those occupancy levels should be correspondingly minimal.
The standard answer is that a drug-adapted brain is not a drug-naive brain. True, but for the hyperbolic logic to work, it would have to mean that the relationship between occupancy and neuroadaptation is itself nonlinear in a way that does not track the therapeutic dose-response curve. Meaning the system adapts substantially even at subtherapeutic occupancy, and the buffering capacity seen under usual circumstances is no longer present (why though?), and withdrawal is driven by a different unknown and unspecified downstream process than the one producing benefit.
And here is a further wrinkle: if the system has been re-adapting at every step, which is the whole premise of slow hyperbolic tapering, then by the time one reaches 40% occupancy after months of stepwise reduction, the downstream systems should have substantially readjusted, and the acute perturbation at each remaining step should be small. But people in the withdrawal communities routinely say that going down from something like 3 mg of escitalopram to 2 mg of escitalopram produces agonizing withdrawal, which makes little sense in light of the buffering capacity seen under usual circumstances and given the fact that the brain has supposedly re-adapted to the 3 mg dose.
There is also an open question of whether the specific “hyperbolic” shape of tapering is mechanistically relevant, or whether a slow taper of some semi-linear shape of equivalent duration accompanied by equivalent clinical support would do just as well.
Does hyperbolic tapering “work” through the passage of time? If it takes 2 years to taper, a lot would have and could have changed in those 2 years.
Does hyperbolic tapering work for psychological reasons that have nothing to do with pharmacokinetics? A structured protocol provides reassurance, a locus of control, a plan for when symptoms are prominent, and graduated confidence building at each small step. All of which are therapeutic mechanisms. If the active ingredient is structure and reassurance, not hyperbolic shape, then the same support for a slow linear taper of equivalent duration should perform identically.
Is there any evidence that hyperbolic tapering prevents protracted or delayed withdrawal? Not really. And if protracted symptoms reflect some sort of neurological injury (again, hypothetical), then the rate of tapering might make no difference at all, just as the persistence of alcohol-related cognitive deficits doesn’t depend on how fast you taper off alcohol.
Can very long tapers keep people “stuck”? The notion that one’s nervous system is fragile and must be handled with exquisite care can keep the medication the central organizing fact of one’s life in a multi-year taper. Every little fluctuation is filtered through the lens of withdrawal.
Some predictions, frankly speculative
What follows is educated guesswork and a series of personal bets about where the field of antidepressant withdrawal (and withdrawal from other psychiatric medications) goes over the next decade. These predictions are conditional on withdrawal receiving more attention from the medical research community and on the collection of high-quality data through funded research. I don’t know the future. I likely will be shown wrong on at least some predictions. Nobody really knows right now since there is no meaningful data on most of these questions. I’m laying out these predictions anyway, to stimulate further discussion along these lines and to outline my own thinking about these complicated matters.
“Withdrawal” as a broad iatrogenic concept is going to be problematized. As more people become interested in scientifically studying reports of withdrawal, the heterogeneity hiding under the term “withdrawal” will become impossible to ignore, and researchers will start calling for differentiated and more precise characterizations, à la Chouinards. The politics of iatrogenic harm communities will slow it down, because conceptual unity under the banner of withdrawal has been strategically useful for advocacy, and disaggregation opens the door to minimization of some groups with iatrogenic concerns. But the science will force the issue eventually.
Hyperbolic tapering will show modest or no superiority over slow quasi-linear tapering under blinded and randomized conditions. If the shape of the dose reduction matters less than the duration of taper, clinical framing, and the patient’s psychological relationship to the taper, then a slow quasi-linear taper of equivalent duration should perform comparably under controlled conditions. A Cohen’s d under 0.3 is my best guess. By quasi-linear I’m envisioning something like linear dose reduction down to the last standard dose (e.g. 5 mg of escitalopram) and then going down by half or quarter tablet intervals (e.g. by 2.5 mg or by 1.25 mg).
“Hyperbolic” will come to mean “nonlinear” rather than a specific geometric shape following the receptor occupancy curve. In practice I feel this is already underway. The precise matching of SERT occupancy curves will turn out to be unsupported, I suspect, and it will give way to a general principle of gradual, nonlinear (but not strictly hyperbolic) tapering, and research may turn up nonlinear shapes that predict symptom severity or guide tapering better than the exact shape of the receptor occupancy curve.
Hyperbolic tapering will show no advantage for relapse prevention over slow quasi-linear tapering, and both will show higher relapse rates than maintenance treatment. I do believe hyperbolic tapering is superior to a rapid linear taper for relapse prevention, but I doubt the superiority will hold once time duration and expectancy effects are taken into account.
A meaningful minority of people with years-long “withdrawal” will turn out to have something like ME/CFS, POTS, or post-viral syndromes, and no clear causal relationship to antidepressant discontinuation will be established. My estimate is that this will be applicable to something like 10–20% of patients with self-described protracted withdrawal.
Another meaningful minority of individuals in self-described protracted withdrawal will be good candidates for a functional neurological or somatization-related diagnosis. Again, maybe 10–20%. There is conceptual work to be done before this can be stated with confidence. The extension of FND to iatrogenic states needs better characterization.
Tapering outcomes will turn out to depend heavily on expectancy and placebo/nocebo. People who expect severe, protracted withdrawal will have longer, more symptom-laden tapers, even after we control for everything else. Expectancy and nocebo will not explain everything, but they will emerge as powerful predictors of variance. Something on the order of 30-40% of variance will be explained by expectancy and nocebo.
Outcomes will also track personality characteristics, in the way David Mintz’s psychodynamic psychopharmacology describes for pharmacological outcomes generally. Attachment style, distress tolerance, health anxiety, neuroticism, etc, I expect these to emerge as moderators.
The study of withdrawal will move toward a “biopsychosocial” model. In addition to the physiological aspects, the experience of withdrawal will turn out to be modulated by a variety of psychological factors, and the trajectory will be shaped by social context such as community membership, information environment, the therapeutic relationship with the deprescriber, etc, etc. The exclusive focus on receptor occupancy curves will come to be regarded as naïve and reductionistic.
Adjunctive medications will get studied for facilitating withdrawal. Lamotrigine comes to mind as a plausible candidate. Other mood-stabilizing and anxiolytic medications as well. It is also possible that the first agents shown to help will be the ones addressing symptoms like anxiety, insomnia, somatic hyperarousal (e.g. gabapentin, propranolol, clonidine) rather than ones selected by receptor-level reasoning. These medications will be resisted by old-school folks belonging to iatrogenic harm communities, but new patients with no ideological commitments, looking simply for relief, will welcome them from their physicians.
In other domains of substance dependence, the standard approach includes pharmacological management of the withdrawal state. Alcohol withdrawal is treated with benzodiazepines or phenobarbital. Opioid withdrawal is managed with buprenorphine. There is an ideological current in the current antidepressant withdrawal community that is suspicious of pharmacological solutions on principle. Proposing another drug to manage the withdrawal feels like doubling down on the original sin.
Iatrogenic harm identity will get recognized the way illness identity already is. Somebody is eventually going to write about this community in the vein of Rachel Aviv’s Strangers to Ourselves.1 Within a decade I expect one or two high-profile cases of people who lived “in withdrawal” for years publicly reconceptualizing the experience (for a glimpse of this sort of thing, see Lisa Wallace’s story).
Aviv, in Q&A with me: “I think there’s a way in which the oversimplified story of “medication fixed my chemical imbalance” can get mirrored, like a 180-degree flip, by the inverse story: “my suffering is 100% a result of psychiatry.” Of course there are many cases in which psychiatry intervenes in traumatic, paternalistic and damaging ways, but I think the iatrogenic harm story can involve a simplification that, oddly, mimics the “chemical imbalance” one in the sense that there’s a failure to think about social, economic, and structural reasons for a person’s distress. It can become another too-easy story that blots out ambiguities.”