Which antidepressants carry the highest risk of manic switch?

And why the ranking you were taught falls apart when you look at the numbers behind it.

What does the best comparative evidence show?

The analysis that best answers the comparative question is a 2025 network meta-analysis in eClinicalMedicine — the only one to compare individual antidepressants rather than classes. Thirteen randomised trials, 1,362 patients, twelve agents plus placebo.

Risk ratios against placebo ran from 4.53 down to 0.31. Not one reached significance — against placebo or against any other agent. The authors rated their own confidence in the findings as low.

Two other network analyses sit alongside it. Together they are the whole comparative evidence base.

So why does everyone quote a ranking?

Network meta-analyses generate a P-score for each treatment, where 1 is safest. Here is the full safety ranking from the 2025 analysis.

Is there any real difference between agents?

Only two randomised trials have compared antidepressants head to head on switch outcomes, and one of them produced the single reproducible significant result in this literature.

Randomised, 10-week acute phase with up to a year of continuation, all three given as adjuncts to a mood stabiliser, 174 patients. The difference in threshold switching was significant (χ² = 12.45, p = 0.002).

What makes it interpretable is the efficacy column. All three worked equally well. So the switch difference is not the artefact of one drug simply being more effective and pushing patients further.

One caution if you cross-check this. CANMAT and several reviews cite the same trial as venlafaxine 38%, sertraline 10%, bupropion 9% — a different metric drawn from the same dataset. The discrepancy is not an error in either source; it is the point made further down about how much the definition drives the number.

Four details that sharpen it

1. It concentrates in rapid cyclers. Among that group bupropion carried a significantly lower switch risk than venlafaxine (p < 0.01).
2. Switches came sooner. Time to switch was significantly shorter on venlafaxine than on either comparator.
3. It did not settle after the acute phase. The ratio of full-duration switches to brief hypomanias stayed higher for venlafaxine both acutely (3.60) and in continuation (3.75), where it fell for the other two.
4. A second trial pointed the same way. Paroxetine against venlafaxine, 60 patients over six weeks: 3% versus 13% — though it did not reach significance.

What about tricyclics?

Tricyclics come out high on point estimates — imipramine and desipramine ranked second and third highest — but neither reached significance, and amitriptyline paradoxically ranked among the safest. Individual tricyclics do not clearly differ from each other. The class-level figure, OR 2.57, carries a confidence interval from 0.48 to 13.8.

The frequently quoted desipramine switch rate of 43% comes from a small comparison summarised in a review article, not from a randomised trial powered to detect it.

The ISBD Task Force position — that SSRIs and bupropion may have lower switch rates than tricyclics, tetracyclics and SNRIs — is explicitly consensus rather than a conclusion from significant head-to-head data. Reasonable to avoid tricyclics in bipolar depression. Less reasonable to present the reasoning as settled.

Is bupropion actually the safest?

So: a reasonable first choice, with real evidence behind it in one specific comparison, and no evidence it is worse than the alternatives. That is weaker than "safest."

Which agents can't be ranked at all?

Their absence from the ordering is a data gap, not evidence of safety. If a colleague asks whether escitalopram is safer than sertraline for switch risk, the accurate answer is that nobody has looked.

Why do reported rates vary so much?

Three numbers, one trial. The definition does more work than the pharmacology.

Duration does similar work.

Those last two rows appeared the same year with similar-looking numbers, and they answer different questions. Acute trials under-detect throughout: the reassurance from the network analysis applies to a ten-week window, and the authors say so.

So what do you actually do?

1. Avoid venlafaxine in bipolar depression, particularly with rapid cycling or bipolar I. The one agent-level recommendation the randomised evidence genuinely supports — and you give up nothing in efficacy.
2. Be cautious with noradrenergic tricyclics on the same reasoning, acknowledging the evidence there is weaker.
3. Choose bupropion or a better-studied SSRI — sertraline, paroxetine, fluoxetine. Reasonable in direct comparisons.
4. Do not choose between those on switch risk. The differences among them are not statistically established, so a preference between sertraline and fluoxetine on this basis is not supported.
5. Use it as a short-term add-on to a mood stabiliser or antipsychotic rather than as monotherapy in bipolar I.
6. Screen for bipolarity before you prescribe. Patient factors — rapid cycling, bipolar I, earlier onset, prior switch, substance use — do more work than agent choice.

And would you recognise one?

The sleep row does most of the work, and it is easy to miss by asking the wrong question. “How are you sleeping?” gets the same answer — badly, not much — from a patient who is switching and one whose depression is still keeping them awake. “Are you tired during the day?” separates them.

On the family row: ask what has changed rather than whether they are worried. A relative relieved to see improvement will under-report.

Unlike everything above it, this section is clinical practice guidance rather than a finding from the comparative trials — the switch literature measures rates, not how to spot one in clinic.

The free chapter below covers the full differential — including how to separate a switch from activation syndrome and akathisia, why a positive screening questionnaire is usually a false positive at realistic base rates, and what to do in the first twenty-four hours.