Here is the version of this I see most often. A patient stops an antidepressant, feels terrible within a few days, and comes back convinced the medication was the only thing keeping them together. Or the prescriber sees the rapid deterioration and concludes the same thing, restarts the medication, the patient feels better, and everyone treats it as confirmation. The medication gets locked in for years on evidence that never proved what anyone thought it proved.
The problem is not that the patient felt terrible. The problem is that we did not know what the terrible meant, because we never asked which of three very different things was actually happening.
Many symptom flares after a medication change are reflexively labeled side effects even when their timing, quality, or relationship to dosing suggests something else is happening. The patterns are predictable once you know what to look for.
The three things clinicians collapse into one
Side effects are the pharmacological actions of the drug happening while it is on board. Many common early side effects are dose-related, mechanism-predictable, and most likely to appear at initiation or after a dose increase. Not all adverse effects fit that pattern: sexual dysfunction, weight gain, emotional blunting, and rare idiosyncratic reactions may appear later, vary less predictably with dose, and may not resolve with time.
Sertraline causes nausea in the first two weeks for a lot of people. Mild, expected nausea early in treatment is not automatically a reason to stop. The gut is full of serotonin receptors, and the same receptor-level activity happening in the brain is happening there too. Whether to push through it depends on severity, trajectory, and whether the patient can tolerate it. Severe or persistent nausea is a different question. And nausea is not evidence that the medication is working therapeutically: side effects do not predict efficacy. Vyvanse suppresses appetite because it is an amphetamine. That is an expected pharmacologic effect, but it still requires monitoring and can become treatment-limiting if it causes significant weight loss or nutritional compromise.
The clinical clue for a side effect: the symptoms correlate with the drug's mechanism and they appeared, usually, with initiating or when the medication was increased.
Withdrawal is the body reacting to the absence of a drug it has adapted to. It does not mean the medication was working for the condition it was prescribed for. It means the nervous system became physiologically adapted to its presence, which is a normal biological process that happens whether or not the medication was doing anything therapeutically useful. Physiologic adaptation is not the same as addiction, and it is worth being clear about that with patients. When a drug has been present long enough, the system adjusts around it being there. Remove it and the system has to recalibrate. The symptoms of withdrawal are what that process feels like from the inside.
Withdrawal tends to come on relatively quickly for shorter-half-life medications, often within one to three days of stopping or significantly reducing. For longer-half-life agents the onset may be delayed by days to weeks as the concentration falls more gradually. Withdrawal symptoms often include novel, sensory-heavy, or physical symptoms that do not resemble the original condition. But withdrawal can also produce anxiety, insomnia, depressed mood, irritability, and emotional lability that overlap substantially with the underlying disorder, which is part of what makes the distinction hard. Brain zaps. Dizziness. Flu-like achiness. Agitation. Sensory disturbances. Venlafaxine is particularly associated with discontinuation symptoms primarily because of its relatively short half-life, which can produce a rapid fall in pharmacologic exposure, especially after missed doses or abrupt reductions. Patients on Effexor report brain zaps if they are even a few hours late on a dose. That is not a psychiatric symptom. That is pharmacology.
The clinical clue for withdrawal: rapid onset after dose reduction or stop, symptoms that do not match the original diagnosis, and resolution with time. That last part matters. Withdrawal often improves on its own as the nervous system recalibrates, though duration varies substantially. Many patients improve within days to a few weeks. Others, particularly after long exposure or abrupt cessation of higher-risk agents, experience symptoms that persist longer and may warrant reinstatement, taper revision, or closer monitoring. For most patients the discomfort does have an end point that is not tied to staying on the medication permanently.
Rebound is the original target symptom returning, often briefly intensified, when the medication's effect dissipates. It is not the same as relapse and it is not the same as withdrawal. Insomnia that temporarily exceeds the pretreatment baseline after stopping a regularly used hypnotic. Anxiety that intensifies beyond baseline as a short-acting benzodiazepine concentration falls between doses. The classic one in ADHD practice: the evening return of cognitive activation as Vyvanse wears off around hour eight or ten, the sense that mental volume has increased and sleep onset becomes impossible. That may be wear-off, meaning the medication benefit has simply dissipated and the baseline symptoms are returning, or it may be true rebound, a transient overshoot past baseline that exceeds what the patient experienced before starting the medication. The distinction matters because wear-off usually calls for a duration adjustment while true rebound may call for a different agent or delivery mechanism. Either way it is not a mood disorder and not evidence of medication failure.
The clinical clue for rebound: symptoms match the original condition, onset correlates with the medication wearing off, and they resolve with reinstatement or dosing adjustment.
One nuance worth knowing: rebound can feel briefly worse than the patient's pre-medication baseline, and that can happen two different ways. The physiological version is real and common with shorter-acting drugs, particularly benzodiazepines. Neuroadaptation during ongoing benzodiazepine exposure can leave the nervous system temporarily more excitable as the drug concentration falls. When anxiety predictably intensifies beyond baseline as clonazepam concentrations fall, it may reflect the original anxiety plus an overshoot from a nervous system that adapted to the drug's presence. That is part of why benzo dependence is so self-reinforcing.
The perceptual version is different but equally real. A patient who has had two months of genuine relief now has a reference point they did not have before. The same baseline anxiety they were living with before treatment now feels intolerable by comparison because they know what better felt like. That is not exaggeration. The contrast is real, and it matters clinically because it changes how you counsel them about what they are experiencing.
The one most clinicians miss
The withdrawal-versus-relapse distinction is the most consequential and the most commonly fumbled.
When a patient stops an antidepressant and feels terrible within a few days, that rapid onset raises suspicion for withdrawal, particularly when the symptoms are novel, physical, or sensory rather than a return of the original presentation. These are probabilistic clues, not diagnostic rules, and both processes can occur together.
Withdrawal tends to come on relatively quickly because it reflects the nervous system's adjustment to the drug's sudden absence.
True recurrence runs on a different timeline. If a medication was genuinely treating the underlying disorder, removing it sets the neuroadaptive process in reverse. That process does not unravel overnight. A gradual return of the original syndrome after a period of stability, especially past the initial withdrawal window, raises more suspicion for recurrence than symptoms appearing within the first few days.
The most useful differentiation: are the symptoms novel and unlike the original disorder, or do they look and feel exactly like what the patient had before treatment started. Withdrawal tends to produce the former. Recurrence tends to produce the latter. Neither timing pattern alone is diagnostic.
Reinstating the medication does not settle this. A medication restarted during active withdrawal will resolve the withdrawal, and it will look like proof the medication was necessary. It is not proof. Withdrawal may improve rapidly after reinstatement even if the medication was not providing meaningful therapeutic benefit for the underlying disorder. Time is one of the most useful pieces of evidence available, and it is rarely given. So is symptom quality, the drug's half-life, missed-dose sensitivity, whether the symptoms are novel versus characteristic of the original illness, and how the patient's course looked before treatment started.
Why this matters for your prescribing
Every one of these misclassifications leads to a wrong clinical move.
Withdrawal called a side effect: this is the trickiest of the three because it requires you to recognize that a patient still technically on a medication can be experiencing withdrawal from it. It happens most often with short half-life drugs during a taper that moved too fast, or when doses are being missed, or when a patient is a few hours late and the level has already dropped enough to trigger symptoms. They call in and say the medication is making them feel terrible, dizziness, brain zaps, agitation, nausea. The prescriber hears the medication is causing problems and stops it or switches it, when what is actually happening is the level dropped fast enough to cause discontinuation symptoms. The response may be to return to the last tolerated dose, slow the taper, improve dosing consistency, or otherwise adjust the strategy. The wrong conclusion is that the medication itself has newly become intolerable.
Withdrawal called relapse: the prescriber restarts the medication, concludes it was necessary, and the patient is on it indefinitely based on evidence that does not hold up.
Rebound called a side effect: the prescriber reduces the dose or stops the medication, when what is needed is a different delivery mechanism, a longer-acting formulation, or a small evening booster.
Side effects called intolerance: the prescriber abandons a medication that might have worked if the patient had been told what to expect and given an appropriate period to acclimate, when clinically safe and acceptable.
The overlap zone
Patients rarely present these neatly. They say I feel worse or it stopped working or I feel off. That phrasing is not a classification. It is the beginning of a clinical question.
When a patient comes in with a symptom flare after a medication change, run through timing first. When exactly did the symptoms appear relative to the change? Were they worse in the morning, late afternoon, days after stopping, immediately after stopping? Then symptom quality. Do these symptoms resemble the original condition or are they new, physical, sensory? Then mechanism. What does this drug do, and could what they are describing be a direct pharmacological effect of it being present, absent, or wearing off?
Half-life is one of the most useful pieces of information for thinking through these distinctions. Paroxetine has a relatively short half-life, no long-lived active metabolite, and greater anticholinergic activity than most other SSRIs, all of which complicate discontinuation. Its withdrawal can begin quickly, can be severe, and is sometimes mistaken for the anxiety disorder it was prescribed for. Fluoxetine has a long half-life and an active metabolite that persists for weeks, so its concentration falls more gradually than shorter-acting agents. That lowers, but does not eliminate, the likelihood of acute discontinuation symptoms. The same property is why clinicians sometimes use a short course of low-dose fluoxetine as an off-label bridge when withdrawal from a shorter-half-life antidepressant has been difficult, though the approach must be individualized.
Documentation guidance
When you have classified it, say so.
For a side effect: "Nausea consistent with expected serotonergic activation at initiation. Patient counseled that this typically resolves within two weeks. No dose change indicated at this time."
For withdrawal: "Symptoms consistent with SSRI discontinuation syndrome: dizziness, sensory disturbances, irritability. Current symptom pattern is more consistent with discontinuation syndrome than recurrence of the prior depressive episode. Taper slowed and individualized based on medication half-life, duration of exposure, prior withdrawal symptoms, and response to each reduction."
For rebound: "Evening cognitive activation consistent with stimulant wear-off effect given Vyvanse's eight to ten hour duration at current dose. Sleep onset affected. Plan to evaluate extended coverage or small afternoon IR booster."
For the withdrawal-versus-relapse question: "Symptoms appeared within 72 hours of discontinuation, pattern currently more consistent with discontinuation syndrome than disorder recurrence. Will monitor trajectory over the coming weeks. If symptoms increasingly resemble the original disorder, progressively worsen, or remain functionally impairing after the physical and sensory symptoms would be expected to improve, concern for recurrence will increase and medication reinstatement or alternative treatment will be reconsidered."
That last one is the note most people never write, because they never ask the question. But it is the note that protects you if the patient comes back in three weeks feeling fine, because you already named what you were watching for and why.
P.S. The AI Scribe is built around this same problem. Before it writes the assessment, it stops and puts the clinical calls in front of you to confirm or change, so the reasoning is yours and the note shows it.