Do Antidepressants Cause Mania, or Reveal Bipolar Disorder? What the Evidence Can and Can’t Settle
TL;DR
Whether an antidepressant causes a manic switch or simply reveals a bipolar vulnerability that was already there is, for most individual patients, a genuinely unresolved question. Clinicians aren’t being evasive about it; the evidence just doesn’t settle it yet.
A large 2024 Danish study found no significant increase in mania risk after antidepressant treatment in bipolar depression. A 2026 study of bipolar I patients found a roughly 30% higher risk of manic or mixed hospitalisation after starting an antidepressant. Both can be true at once: they measured different populations and different outcomes.
Mixed episodes (agitation and impulsivity alongside depressive thinking) appear to carry more of the excess risk than classic euphoric mania, and are easy to miss.
Family history, earlier illness onset, and a depression-mania-interval course pattern are the more credible predictors of switching; none of them can reliably predict an individual patient’s outcome.
One switch does not automatically mean “this patient has bipolar disorder for life.” Diagnostic frameworks disagree on how much weight to give a single treatment-emergent episode, and precise documentation matters more than a quick label.
Introduction
Two well-conducted studies, a couple of years apart, appear to say opposite things. A 2024 Danish national registry study of patients recently discharged for bipolar depression found no significant association between antidepressant treatment and a return of mania. A 2026 study following nearly 8,000 bipolar I patients found roughly a 30% higher risk of manic or mixed hospitalisation after starting an antidepressant.
Both are probably right, because they asked slightly different questions of different populations. This piece works through why the question of whether the antidepressant caused it or just revealed it resists a single answer, what the best current evidence supports, and why the more clinically useful conversation is often not about causation at all, but about recognising and managing whichever it turns out to be.
Why is this question so hard to answer?
The central difficulty is that both explanations predict the same clinical picture: a patient who becomes manic or hypomanic while taking an antidepressant. If the drug caused it, stopping the drug should prevent recurrence. If the drug revealed an underlying vulnerability, the episode would likely have appeared eventually regardless, and stopping the drug treats the symptom without addressing the cause. Randomised trials that could cleanly separate these explanations are difficult to run: you cannot ethically randomise depressed patients to years of untreated depression to see who eventually becomes manic anyway.
What we have instead is a mix of acute randomised trials (which mostly examine short-term switch risk, not long-term causation), large observational cohorts (which can follow patients for years, but cannot fully rule out that clinicians already choose antidepressants differently for patients who look more or less “at risk”), and biological plausibility arguments on both sides. None of these designs, alone, resolves the question.
What does the evidence say about causation?
Several strands of evidence are more consistent with the antidepressant revealing, rather than creating, an underlying bipolar vulnerability. Acute randomised trials have not shown a clear excess of switching over placebo for individual antidepressants; a large one-year Danish study found no significant association between antidepressant treatment and mania recurrence in bipolar depression; and the strongest predictors of eventual bipolar diagnosis (family history, earlier age of onset) exist before any medication is started.
In the Danish target-trial emulation of 979 recently discharged bipolar-depression patients, the hazard ratio for mania recurrence with antidepressant treatment over one year was 1.08 (95% CI 0.72–1.61), not statistically significant, whether or not the patient was also on a mood stabiliser. The authors described the added risk as negligible for this specific population and outcome. That word “negligible” is often quoted without its context: it describes recorded mania recurrence in patients already diagnosed with bipolar depression, not risk in every population or every outcome measure.
What evidence points toward a genuine drug effect?
Other findings are harder to explain purely as unmasking, particularly evidence of a dose-and-timing relationship and results that differ by treatment strategy rather than by patient characteristics alone. A Swedish registry study found a substantially higher switch risk with antidepressant monotherapy (hazard ratio 2.83, 95% CI 1.12–7.19) that was not present when a mood stabiliser was given concurrently. That’s difficult to explain solely by pre-existing vulnerability, since the patients’ underlying diathesis presumably didn’t change based on what else was prescribed alongside the antidepressant.
More recently, a 2026 population-based cohort of 7,946 bipolar I patients, propensity-matched between antidepressant users and non-users, found antidepressant use was associated with a 27% higher risk of manic or mixed hospitalisation overall (hazard ratio 1.27), rising to 51% with monotherapy, and more than doubling in patients aged 18–35 on monotherapy. Combining an antidepressant with a mood stabiliser showed no significant increase in this composite outcome. Findings like these, where risk tracks treatment strategy rather than simply patient profile, are the strongest evidence currently available for something closer to a genuine, modifiable drug effect, at least under certain treatment conditions.
How can the 2024 and 2026 studies both be right?
They studied different populations, different exposure definitions, and, critically, different outcomes, so they are not actually contradicting each other despite the surface-level tension. The 2024 study followed patients already discharged for a bipolar depressive episode and measured recorded mania recurrence over one year. The 2026 study followed a broader bipolar I population and measured hospitalisation for mania or mixed episodes specifically: a more severe, and in this case more mixed-episode-weighted, endpoint.
That distinction matters because the 2026 cohort found the excess risk was driven predominantly by mixed episodes rather than pure mania, a pattern the 2024 study’s outcome definition may not have been positioned to capture as clearly. The honest summary isn’t that the newer study overturns the older one. It’s that the outcome worth watching most closely may have shifted from “did mania recur” to “did a manic or mixed episode become severe enough to need hospitalisation,” and mixed episodes appear to be where more of the risk concentrates.
Why do mixed episodes matter more than the mania-versus-not framing suggests?
A mixed episode (agitation, impulsivity, and increased activity occurring alongside retained depressive thinking, hopelessness, or suicidality) is arguably the more dangerous and more easily missed outcome, because it doesn’t look like the euphoric mania most people picture. In the 2026 cohort discussed above, every antidepressant strategy studied showed a substantially higher relative risk for mixed-episode hospitalisation than for pure-mania hospitalisation.
The practical implication is a diagnostic trap worth naming explicitly: a patient whose depression becomes more agitated, activated, sleepless, or suicidal on an antidepressant is not obviously “improving” or obviously “manic.” They may be in a mixed state that needs a different response than simply continuing or increasing the current treatment. Reflexively raising the antidepressant dose in this picture, rather than reassessing the mood state itself, is one of the more consequential recognition errors this evidence should discourage.
Can we predict which patients are at higher risk before starting an antidepressant?
Some factors are genuinely useful for calibrating vigilance, but none can reliably predict an individual patient’s outcome, so they should inform monitoring intensity rather than a strict prescribing rule. In one prospective study of 1,629 antidepressant-treated bipolar patients, a depression-mania-interval course pattern, older age, a higher number of episodes per year, psychiatric family history, and a previous suicide attempt remained independently associated with switching. Earlier age of illness onset and family history of bipolar disorder are the more consistently supported pre-exposure predictors across the wider literature.
Other frequently cited factors (bipolar I subtype specifically, rapid cycling, mixed features at the index episode) show up inconsistently, often in single studies or lower-quality designs rather than across the board. The clinically honest position is that these factors shift population-level probability and should raise the threshold for closer monitoring; they do not amount to a test that identifies which specific patient in front of you will switch.
Should patients be screened for bipolar disorder before starting an antidepressant?
A brief structured screen is worthwhile before starting an antidepressant, but its result needs to be interpreted through the setting’s underlying prevalence of bipolar disorder, because a positive screen in a low-prevalence population is far more often wrong than right. The Mood Disorder Questionnaire, at roughly 80% sensitivity and 70% specificity, yields a positive predictive value of only about 12% at an assumed 5% community bipolar prevalence. That means roughly seven or eight of every ten positive screens in that setting would be false positives.
That is not an argument against screening; it’s an argument for what a positive screen should trigger. A positive result is an invitation to a fuller diagnostic interview and collateral history, not a diagnosis in itself, and specialist settings with a higher underlying prevalence will see meaningfully better predictive value from the same instrument.
Does one switch mean a patient has bipolar disorder?
Not automatically, and the major diagnostic frameworks don’t fully agree on how to treat a single treatment-emergent episode. DSM-5-TR allows a fully syndromal manic episode, or a hypomanic episode following major depression, to count toward a bipolar diagnosis if it persists at full severity beyond the medication’s physiological effect, but a brief, nonspecific activation that resolves as the drug clears does not meet that threshold. Other bodies take a more cautious stance toward converting a single antidepressant-associated elevation, in a patient who otherwise looks unipolar, directly into a bipolar diagnosis.
What should travel with any such episode in the chart is precision, not a premature label: the exact symptoms and their duration, the timing relative to the medication change, whether the picture persisted after the drug was stopped and cleared, family history, and which diagnostic framework was applied. A clinically honest way to describe this to a patient: the elevated state clearly appeared during treatment, but whether the medication caused it or revealed something that was already present often cannot be determined immediately. Both are possible, and the treatment decision in front of you doesn’t have to wait for that answer to be settled.
Conclusion
The cause-versus-unmasking question is not a gap in clinical knowledge that will close with one more study. It reflects a genuinely difficult separation to make in individual patients, given the study designs available. What the evidence does support is a shift in where the clinical attention should go: less on settling causation in the moment, and more on recognising the episode precisely (including mixed presentations, which may carry more of the risk than classic mania), documenting the timeline and framework rigorously, and using known risk markers to calibrate vigilance rather than to predict any one patient’s outcome. That is a less satisfying answer than a clean “yes, it caused it” or “no, it was always going to happen,” but it’s the one the evidence currently supports.
FAQ
If a patient becomes manic on an antidepressant, does that always mean they have bipolar disorder?
Not automatically. DSM-5-TR allows a fully syndromal episode that persists beyond the medication’s physiological effect to count toward a bipolar diagnosis, but frameworks differ on how to treat a brief, transient elevation. The full clinical picture, timeline, and collateral history matter more than the single episode.
Is it more dangerous to continue or stop the antidepressant once a switch happens?
There’s no single trial establishing an optimal approach, and the decision depends on severity, the specific drug’s half-life, and withdrawal risk. Severe mania, psychosis, or dangerous behaviour generally favours faster discontinuation; milder presentations may allow a more gradual taper. This is a decision for the treating clinician, made case by case.
Why do some studies say antidepressants barely raise mania risk while others show a real increase?
Because they often measure different populations and different outcomes. A study following recently discharged bipolar-depression patients for recorded mania recurrence can show a different result than a study following a broader bipolar I population for hospitalisation from mania or mixed episodes. Both can be accurate descriptions of what they actually measured.
Can a genetic or blood test predict whether someone will have a manic switch on an antidepressant?
Not currently. No pharmacogenomic or biological test available today predicts treatment-emergent switching reliably enough to guide an individual prescribing decision. Family history and clinical history remain the more informative, if imperfect, indicators.
Continue reading this series
This post is for educational purposes only and does not constitute individualised medical advice. If you have concerns about symptoms or medication, please consult a qualified clinician.
