Published on 19 September 2026

Analysis · Bipolar disorder · Psychopharmacology

Pramipexole in resistant bipolar depression: a trial closed at thirteen percent of target, with a signal on hypomanic symptoms

▬ Publication
Journal of Psychopharmacology · 2025;39(2):106-120 · McAllister-Williams et al.
DOI 10.1177/02698811241309622
PMID 39829389
Scientific 62
Editorial 69

Editorial collections PEB Academy A Study Under the Microscope

The essentials

The PAX-BD trial was designed to randomize 290 patients with treatment-resistant bipolar depression. It randomized 39, and 36 provided data for the primary outcome. At twelve weeks, pramipexole added to a mood stabilizer lowered the QIDS-SR score by 4.4 points against 2.1 under placebo, an adjusted difference of 2.9 points (95% CI −0.4 to 6.3), p = 0.0865, Cohen’s d 0.72. This is a medium-sized effect that falls short of significance, and the authors say so themselves. Statistically significant differences appear later, in roughly ten patients per arm: 6.3 points on the QIDS-SR and 5.36 points on psychosocial functioning at thirty-six weeks, 46% response against 6% at study exit. On tolerability, 44% of patients on pramipexole reported at least one hypomanic or manic adverse event against 29% on placebo, and a manic relapse with psychotic symptoms led to hospitalization in one of the eighteen exposed patients, all of whom were also taking a mood stabilizer. Not significant does not mean ineffective: the trial cannot conclude in either direction, and its authors themselves call it inconclusive.

Context

Treatment-resistant bipolar depression is a major therapeutic dead end. The authors note that the UK NICE guideline recognizes only three treatments for bipolar depression, lamotrigine, quetiapine, and olanzapine with or without fluoxetine, with lurasidone added in the British Association for Psychopharmacology guidelines. Once these options have failed, the clinician is left improvising. The dopaminergic hypothesis, and D3 agonism specifically, has fed the idea that pramipexole might have a place in bipolar depression.

Two small older randomized trials, of 22 and 21 patients, with a primary outcome at six weeks, had given positive results. PAX-BD was meant to settle the question. Recruitment ran into the pandemic, then into a funder requirement, present from the trial’s opening, to exclude patients taking an antipsychotic. This requirement, described by the authors as a major obstacle to recruitment, was relaxed midway through by amendment. The funder nonetheless closed the trial early because of slow recruitment, and the team published the trial as it stood, which was not a given.

The study at a glance

Question (PICO)
Population
Adults over 18 with bipolar I or II disorder by DSM-5 criteria, in a depressive episode with a QIDS-SR above 10, followed in secondary care in the United Kingdom. Resistance is defined pragmatically: non-response, intolerance, clinical contraindication or refusal of at least two treatments among quetiapine, olanzapine with or without fluoxetine, lamotrigine and lurasidone. Of 102 patients assessed, 51 entered the pre-randomization phase and 39 were randomized.
Intervention
Pramipexole titrated in 0.25 mg increments every three days over four weeks up to a target of 2.5 mg per day in salt weight, added to a mood stabilizer. Mean dose reached at the end of titration 2.18 mg per day. Eighteen patients.
Comparator
Identical-appearing placebo, manufactured and supplied under blind conditions by a contract manufacturer, added to a mood stabilizer. Twenty-one patients.
Primary outcome
Change in QIDS-SR at twelve weeks, analysis of covariance adjusted for the randomization score, intention to treat. Secondary outcomes: QIDS-SR at twenty-four, thirty-six and forty-eight weeks, response and remission, SHAPS, WSAS, GAD-7, ASRM, QUIP-RS, TSQM, MADRS, QIDS-C and YMRS, adverse events and serious adverse events. Follow-up planned to forty-eight weeks.
Design
Phase III randomized, multicenter, parallel-group, double-blind, placebo-controlled trial, 1:1 randomization, stopped early. Fourteen UK NHS sites recruited into pre-randomization, twelve randomized. Centralized minimization across nine factors with a 20% random component, with the first ten patients allocated at random without minimization. Recruitment from November 28, 2019 to April 5, 2022, suspended from March 18 to October 25, 2020, last visit on October 29, 2022. Analyses conducted in STATA 16. Oxford level of evidence 2b.

Quality control

Point checked	Judgment
Registration and analysis plan	Exemplary
FindingDual registration, ISRCTN72151939 and EudraCT 2018-2869-18, MHRA clinical trial authorization and favorable ethics committee opinion cited with their reference numbers. Protocol published in 2021 in BMC Psychiatry. The authors state that the analysis was governed by a statistical analysis plan established a priori and deposited in the supplementary material, which was not consulted here.
Randomization	Robust
FindingCentralized minimization through a secure web-based system across nine prognostic factors, with a 20% random element to prevent prediction of allocation.
Blinding	Never checked
FindingParticipants, clinicians and the research team were blinded, with an identical-appearing placebo manufactured under blind conditions. But the authors acknowledge among their limitations the absence of any check on blinding, for example by asking participants which arm they believed they were in, and raise the possibility that the drug’s adverse-effect profile unblinded some participants. In their favor, they note that the number of adverse events was comparable between arms, 128 under pramipexole and 162 under placebo.
Sample size reached	13% of target
FindingThirty-nine randomized and thirty-six analyzed on the primary outcome, against a target of 290 randomized and 414 recruited into pre-randomization. The authors judge their initial target perhaps somewhat too ambitious and publish a revised power calculation, based on a minimal clinically important difference raised from 3 to 4 points and on the standard deviation observed in PAX-BD: 34 patients per arm, that is 68 randomized and 95 recruited into pre-randomization, for 90% power.
Analysis time points	Added post hoc
FindingThe QIDS-SR, GAD-7 and ASRM analyses at twenty-four and thirty-six weeks were added after the weekly scores had been observed, as the authors explicitly state. At the same time points, by contrast, the WSAS and QUIP-RS analyses were indeed pre-specified in the analysis plan. The trial’s most widely cited depression result, the thirty-six-week one, therefore comes from a time point added after the data had been seen.
Multiplicity correction	None reported
FindingThe authors declare a two-sided threshold of 0.05 and specify that, outside the primary outcome and Fisher’s exact tests, they report confidence intervals rather than p-values. No correction for the multiplicity of secondary outcomes and time points is described in the publication.
Funding and conflicts of interest	Public, but one point to note
FindingPublic funding through the HTA programme of the UK National Institute for Health and Care Research, an old molecule with no manufacturer of the product among the declared funders. Conflicts of interest are declared in detail and at length. One point deserves flagging to the reader: the second-to-last author, Allan H Young, declares being editor-in-chief of the Journal of Psychopharmacology, the journal publishing the trial. The publication does not specify what editorial procedure was followed as a result.

Results

44%

of patients on pramipexole reported at least one hypomanic or manic adverse event, against 29% on placebo, with one hospitalized psychotic manic relapse among the eighteen exposed patients, all of whom were also taking a mood stabilizer.

Outcome	Result
QIDS-SR at twelve weeks	Adjusted difference 2.9 points (−0.4 to 6.3), p = 0.0865
ReadingA drop of 4.4 points (SD 4.8) against 2.1 (SD 5.1), roughly double under pramipexole. Cohen’s d effect size 0.72. Primary outcome not met. The confidence interval includes zero and spans clinically important effects: the trial does not settle the question.
MADRS and QIDS-C at twelve weeks	Not significant
ReadingMADRS 2.59 (−4.54 to 9.72), d = 0.65; QIDS-C 2.30 (−1.72 to 6.32), d = 0.34, in 16 against 19 patients. The primary outcome is self-rated, and the clinician-rated measure does not confirm it at the same time point. The article’s discussion does not return to this discrepancy.
Response and remission at twelve weeks	25% against 15% and 12.5% against 15%
ReadingNo significant difference. Remission is the only measure in the trial that does not tilt toward pramipexole.
QIDS-SR at thirty-six weeks	6.3 points (1.85 to 10.71), in nine against ten patients
ReadingA time point added after the data had been seen, nineteen patients in total, no multiplicity correction reported. At forty-eight weeks, the 5.5-point difference is no longer significant (−0.045 to 11.0), in eight against eight patients. This thirty-six-week figure should not be repeated without these caveats.
WSAS psychosocial functioning	5.36 points (0.38 to 10.34) at thirty-six weeks
ReadingAdvantage to pramipexole, d = 0.80, in eight against nine patients, then 9.63 points (0.25 to 19.02), d = 0.98, at forty-eight weeks in seven against five patients. Unlike the QIDS-SR, these time points were pre-specified in the analysis plan. No difference at six, twelve and twenty-four weeks. Very small samples, and the authors themselves flag possible attrition bias across all the late results.
Response and remission at study exit	46% against 6% and 31% against 0%
Readingp = 0.026 and p = 0.030 on Fisher’s exact test, in thirteen against sixteen patients, with one cell at zero. Study exit was redefined partway through as week forty-eight or the last available data point after week sixteen for patients affected by the early closure, producing follow-up durations ranging heterogeneously from sixteen to forty-eight weeks. The authors acknowledge possible attrition bias. Striking figures, and very fragile ones.
ASRM self-rated mania scale	+1.2 point (0.02 to 2.42)
ReadingAt twelve weeks, the only significant difference on mania. At later time points, the difference is under one point and not significant, and the clinician-rated YMRS shows no difference at twelve weeks, 1.16 point (−1.09 to 3.41). The proportion of time spent free of manic symptoms is 88% against 96%. Impulse-control-related adverse events are reported by 33% of patients against 19%, but the dedicated QUIP-RS scale shows no difference between arms. Finally, TSQM treatment satisfaction is significantly higher under pramipexole at six weeks, 19.17 points (5.21 to 33.13), and the authors describe tolerability overall as good.

Critical appraisal

Domain	Risk of bias
Randomization process	Low
FindingRobust centralized minimization. A caveat: baseline scores are unbalanced, with a starting QIDS-SR of 15.2 under pramipexole against 17.6 under placebo, and a difference in the same direction on the GAD-7, SHAPS and WSAS. The authors attribute this imbalance to the small sample size, adjust the primary analysis for it, and write that a concern remains about a possible bias, without specifying its direction.
Deviations from protocol	Low
FindingThree-way blinding, strict intention-to-treat analysis of the primary outcome, with patients later judged ineligible or in protocol violation remaining in their allocated arm. The supplementary primary-outcome analyses pre-specified in the analysis plan, adjusted for the minimization factors, were not conducted because of the small sample size, as the authors state.
Missing data	High
FindingSample sizes shrink over time: sixteen against twenty at twelve weeks, thirteen against seventeen at twenty-four, nine against ten at thirty-six, eight against eight at forty-eight, and seven against five for the WSAS at the end of follow-up. Fourteen patients had to exit before week forty-eight solely because of the early closure. Everything reported beyond twenty-four weeks rests on roughly ten patients per arm at most.
Outcome measurement	High
FindingA self-rated primary outcome, blinding never checked, and a clinician-rated measure that stays silent at the same time point. Together, these three elements make it impossible to attribute the observed difference to the drug alone with certainty. The authors themselves put forward the hypothesis of unblinding linked to adverse effects as a possible explanation for the early improvement in pleasure.
Selective outcome reporting	Some concerns
FindingQIDS-SR time points added after the data had been seen, and no multiplicity correction reported. In their favor, the authors state this explicitly, distinguish what was pre-specified from what was not, and describe their own trial as inconclusive.
Definition of resistance	Permissive
FindingIt includes intolerance, contraindication and refusal. The authors acknowledge this among their limitations and quantify it: 24% of randomized patients had failed none of the four recommended treatments in the current episode, and 41% had failed only one, meaning 65% of the sample had failed zero or one treatment, against 6% who had failed three or four. The population is therefore not the one the word resistance suggests, but it reflects a real clinical problem, as the authors themselves argue.

Level of evidence

Scientific62/100

Editorial69/100

Oxford level of evidence 2b. Confidence is high in the trial’s formal conduct and in the honesty of its reporting: the authors themselves describe their primary difference as not significant, their results as inconclusive, and they flag the post hoc origin of some analysis time points, the possible attrition bias, and the lack of any blinding check, concluding that larger randomized trials are needed. Their own reading is that PAX-BD adds further evidence suggesting pramipexole could be effective in this indication, and that larger trials are needed to settle the question. Confidence is low on efficacy, for lack of sample size. It is moderate on the tolerability signal, which rests on small numbers but is convergent: an excess of hypomanic or manic events, a significant rise on the self-rated mania scale at twelve weeks, one serious event judged treatment-related, and more impulse-control-related events. This signal is consistent with the known pharmacology of dopamine agonists, which lends it weight without proving it. Conversely, the clinician-rated mania scale and the impulsivity scale show no difference, and the authors judge tolerability overall to be good: the signal exists, it is not unequivocal.

The colleague test

What a psychiatrist familiar with clinical trials would say if shown PAX-BD between two consultations.

This trial was done well, and it was killed by context, not by the investigators. They had the nerve to publish it without a significant result on their primary outcome, which is rare enough on its own. But you don’t repeat 46% against 6% without saying it covers twenty-nine patients and an exit criterion that got redefined partway through. What sticks with me is the frequency of hypomanic symptoms.

Translation for practice: nothing changes in prescribing, and one more safety figure is added to the file on any dopamine agonist considered in a bipolar patient.

What you can do with this

A trial that does not reach significance with an effect size of 0.72 and a confidence interval running from −0.4 to 6.3 does not demonstrate that the drug is ineffective. It demonstrates that it lacked the power to answer the question. That distinction is worth keeping in mind, in consultation and in team meetings alike.
If you are considering a dopamine agonist in a bipolar patient, this work gives an order of magnitude for the frequency of hypomanic or manic symptoms under a mood stabilizer, and it is not negligible. The authors note that, in their trial, most patients who experienced these events under pramipexole were not taking an antipsychotic, though the sample size ruled out any statistical analysis of this point.
Impulse control disorders are part of the expected picture under a dopamine agonist. Here, they emerge from spontaneous adverse-event reports, 33% against 19%, while the dedicated scale does not differentiate the arms. They should be actively sought, including from family members, since patients rarely report them spontaneously.
On the product’s regulatory status, the publication states only one thing: pramipexole holds a marketing authorization for Parkinson’s disease in Europe and the United States, and no psychiatric indication is mentioned there. A clinical signal, such as the one from PAX-BD, is a distinct matter from a marketing authorization in psychiatry, and the latter could not be established from the documents consulted here. This is a status to check against the regulatory sources currently in force in your own jurisdiction before any prescription.
What you can tell a patient who asks whether there is anything else: the validated options for treatment-resistant bipolar depression have not changed, and this trial does not allow adding one.

Frequently asked questions

Does a p-value of 0.087 with an effect size of 0.72 mean pramipexole almost works?

It means the trial did not have the power to answer the question. With thirty-six patients analyzed instead of the 290 randomized patients targeted, the confidence interval is too wide to rule out either no effect or a large one. Neither efficacy nor lack of efficacy can be concluded. This is also the interpretation the authors themselves settle on.

Why not rely on the 46% response rate at study exit?

Because it covers twenty-nine patients, with one cell at zero, no multiplicity correction, an exit criterion redefined partway through the trial, and follow-up durations ranging from sixteen to forty-eight weeks, and because the authors themselves flag a possible attrition bias. A figure that is correctly calculated can still be uninterpretable in its context.

Is the 44% figure for hypomanic or manic events reliable?

It rests on eighteen exposed patients, which makes the estimate imprecise, and it reflects reported adverse events, not manic switches confirmed by a standardized assessment. Its internal consistency, however, makes it credible as a signal: it comes with a significant rise on the self-rated mania scale at twelve weeks and one documented serious case. Conversely, the clinician-rated YMRS shows no difference. This justifies caution without turning 44% into a reference value.

Should this trial be repeated?

The authors publish a revised power calculation that sets the definitive trial at 34 patients per arm, that is 68 randomized and 95 recruited beforehand, a target far more attainable than the original one. This may be the most lasting contribution of the publication.

Annotated bibliography

Source study. McAllister-Williams RH, Goudie N, Azim L, Bartle V, Berger M, Butcher C, Chadwick T, Clare E, Courtney P, Dixon L, Duffelen N, Fouweather T, Gann W, Geddes J, Gupta S, Hall B, Helter T, Hindmarch P, Holstein EM, Lawrence W, Mawson P, McKinnon I, Milne A, Molloy A, Moore A, Morriss R, Nakulan A, Simon J, Smith D, Stokes-Crossley B, Stokes PRA, Swain A, Taiwo A, Walmsley Z, Weetman C, Young AH, Watson S. A randomised double-blind, placebo-controlled trial of pramipexole in addition to mood stabilisers for patients with treatment-resistant bipolar depression (the PAX-BD study). Journal of Psychopharmacology. 2025;39(2):106-120. DOI 10.1177/02698811241309622. Thirty-seven authors. Trial registered under ISRCTN72151939 and EudraCT 2018-2869-18, MHRA clinical trial authorization CTA 31857/0003/001-0001, ethics approval 19/NE/0233, sponsor Cumbria, Northumberland, Tyne and Wear NHS Foundation Trust (reference RES-17-031). Funded by the HTA programme of the UK National Institute for Health and Care Research; the first author also declares support from the NIHR Newcastle Biomedical Research Centre (NIHR203309). Conflicts of interest declared in detail by numerous authors; worth flagging, the second-to-last author, Allan H Young, declares being editor-in-chief of the journal publishing the trial. PMID 39829389. No supplementary material accompanied the document consulted: the supplementary figures and tables of this publication were not consulted, and no data found only there is reproduced here.

Trial protocol. Azim L, Hindmarch P, Browne G, et al. Study protocol for a randomised placebo-controlled trial of pramipexole in addition to mood stabilisers for patients with treatment resistant bipolar depression (the PAX-BD study). BMC Psychiatry. 2021;21:334. Reference as cited by the source publication; DOI and PubMed identifier could not be verified from the documents consulted here.

Companion monograph. McAllister-Williams RH, et al. Health Technology Assessment, NIHR Journals Library. The authors cite it as in press, with no volume, pagination or identifier. The parallel economic evaluation is likewise announced as published separately, with no usable reference in the document consulted.

The two earlier randomized trials. Goldberg JF, Burdick KE, Endick CJ. Preliminary randomized, double-blind, placebo-controlled trial of pramipexole added to mood stabilizers for treatment-resistant bipolar depression. American Journal of Psychiatry. 2004;161:564-566. And Zarate CA, Payne JL, Singh J, et al. Pramipexole for bipolar II depression: a placebo-controlled proof of concept study. Biological Psychiatry. 2004;56:54-60. Twenty-two and twenty-one patients, primary outcome at six weeks, both positive. References as cited by the source publication, whose existence, journal, year, volume and pagination were confirmed against bibliographic registries on September 2, 2026.

The guidelines cited for treatment options. National Institute for Health and Care Excellence. Bipolar disorder: assessment and management. NICE Clinical Guideline CG185, 2014. And Goodwin GM, Haddad PM, Ferrier IN, et al. Evidence-based guidelines for treating bipolar disorder: revised third edition recommendations from the British Association for Psychopharmacology. Journal of Psychopharmacology. 2016;30:495-553. These are the two texts underlying the list of four treatments used to define resistance in PAX-BD. Reference as cited by the source publication, confirmed against bibliographic registries on September 2, 2026.

The pharmacological background of the impulsivity signal. Rizos A, Sauerbier A, Antonini A, et al. A European multicentre survey of impulse control behaviours in Parkinson’s disease patients treated with short- and long-acting dopamine agonists. European Journal of Neurology. 2016;23:1255-1261. Reference as cited by the source publication, confirmed against bibliographic registries on September 2, 2026.

Regulatory status. The publication states only that pramipexole holds a marketing authorization for Parkinson’s disease in Europe and the United States. The indications authorized in any given jurisdiction, and therefore the exact status of a psychiatric prescription with respect to that authorization, could not be established from the documents consulted here: this is a point to verify against the regulatory sources in force, and to date, before any prescription.

Editorial collections

PEB Academy A Study Under the Microscope