TaperCommunity
Psychiatry has a confidence problem: we prescribe drugs whose mechanisms we do not fully understand, to treat conditions we cannot measure, using targets we cannot confirm, in brains of infinite complexity.
Take paroxetine. The biological psychiatrist’s reasoning goes something like this: prescribe the drug, the drug binds to the serotonin transporter, the transporter gets blocked, serotonin reuptake slows, serotonin levels rise, mood improves. Clean. Logical. Defensible.
Except that chain has at least four separate links, and every single one of them requires a leap of faith.
The first assumption is that paroxetine binds only to the serotonin transporter. Even if in vitro binding studies suggest high selectivity, that says almost nothing about what happens inside a living human brain with its feedback loops, competing receptor populations, and individual genetic variation. The second assumption is that blocking serotonin reuptake is the only relevant thing the drug does once it gets there. The third is that increased serotonin availability is causally responsible for any mood change that follows. The fourth is that whatever mood change occurs is the outcome you intended and not something else the drug is producing through a pathway you were not watching.
Each of these leaps is large. Stacked together, they are enormous.
This is not a fringe critique. It is the state of the literature.
The serotonin theory of depression has been under serious pressure for years. Joanna Moncrieff and colleagues published a 2022 umbrella review in Molecular Psychiatry examining the evidence for low serotonin as the basis of depression. Across multiple lines of research, including serotonin metabolite levels, serotonin transporter binding, and tryptophan depletion studies, they found no consistent evidence that depression is associated with reduced serotonin activity. The theory that SSRIs work by correcting a serotonin deficit is, at minimum, far less established than prescribers and patients have been told.
Pharmacodynamic selectivity (the idea that a drug acts on one target and one target only) is itself a convenient fiction. Paroxetine, for instance, has clinically significant anticholinergic and noradrenergic activity alongside its serotonin effects. These are not minor footnotes. They are part of why paroxetine produces discontinuation effects so severe that the Maudsley Prescribing Guidelines treat it as among the hardest SSRIs to stop. If the drug were simply correcting a serotonin deficit and nothing more, withdrawal would not look the way it does.
The problem extends far beyond antidepressants. Lithium has been used for bipolar disorder for over sixty years, and the mechanisms behind its mood-stabilizing effects remain genuinely contested. Clozapine is the most effective antipsychotic we have, and the reason it outperforms other agents with similar dopamine blockade profiles is not clearly understood. This is not a criticism of these drugs, some of which have strong outcome data. It is an acknowledgment that efficacy and mechanistic understanding are not the same thing.
The patient sitting across from you is not a receptor. They are a person with a unique genetic profile, a gut microbiome that affects drug metabolism, years of prior medication exposure, and a brain that has already adapted to whatever they are currently taking. The gap between the clean pharmacology diagram and that specific human being is where clinical harm lives.
This matters most in two situations. The first is when you are starting a medication and the patient asks why it works. Most clinicians give the serotonin story with more certainty than the evidence permits. That is a small dishonesty that compounds over time, especially when the drug does not work, or when trying to stop it turns into a months-long ordeal nobody warned them about.
The second situation is deprescribing. If a clinician genuinely believed they understood exactly what a medication was doing, they might be more careful about how they reversed it. Instead, the same false confidence that supports prescribing bleeds into stopping: taper over two weeks, check in a month, done. The cognitive dissonance that was useful at the prescribing stage becomes dangerous at the stopping stage. Patients experience protracted withdrawal, it gets misread as relapse, the dose goes back up, and the drug becomes permanent by default rather than by decision.
None of this means the drugs do not work. Some of them clearly do, for some patients, some of the time. What it means is that the clinical conversation needs to be calibrated to what we actually know.
When starting a psychiatric medication, the honest framing is: this drug appears to reduce these symptoms in a meaningful proportion of people who take it. The precise reasons are not fully understood. We will try it, monitor it, and revisit whether it is still serving you at regular intervals. That is not a weak position. That is an accurate one.
For tapering, the implication is practical and urgent. If we do not fully understand what a drug is doing on the way in, we should be even more humble about predicting what will happen when we remove it. Horowitz and Taylor’s work on hyperbolic tapering, published in Lancet Psychiatry in 2019, provides a neurobiologically grounded framework: receptor occupancy changes are not linear with dose, so dose reductions should not be linear either. Reducing from 20mg to 10mg is a very different pharmacological event than reducing from 2mg to 1mg. The Maudsley Deprescribing Guidelines operationalize this into practical schedules that most clinical training programs still do not teach.
The minimum standard should be: go slow, go low, individualize, and do not mistake withdrawal for the return of illness without ruling out the taper as the cause.
Prescribing a psychiatric medication requires, as one clinician put it, a level of cognitive dissonance where you know the leaps of faith are large but you proceed anyway. That is sometimes the right call. What is never the right call is pretending the leaps are not there. Patients who understand the uncertainty of the mechanism are better positioned to make real decisions about starting treatment, staying on it, and coming off it safely. Informed consent built on a story we cannot fully support is not informed consent. It is a comfortable fiction that serves the prescriber more than the patient.