Benefit-risk arithmetic for monthly aripiprazole: the harm column was built from relapses

• Post hoc benefit-risk analysis of NCT01567527, a 52-week multicentre double-blind placebo-controlled randomised-withdrawal trial of aripiprazole once-monthly 400 mg in bipolar I disorder, published online on 1 July 2026 and in print in volume 413 dated 15 November 2026. Patients entered in a manic episode, were stabilised on oral aripiprazole and then on the monthly injection, and 266 who met stability criteria were randomised 1:1. The registry gives two site counts for the same trial: its location list holds 76 entries, while its own recruitment note reports 103 sites across the same seven countries. Both are given here, and the region claim rests instead on the patient breakdown, which is the firmer figure: of the 266 randomised, 203 were enrolled in the United States, 20 in Romania, 19 in Japan, 14 in South Korea, 5 in Poland, 3 in Taiwan and 2 in Canada. Registration was prospective, submitted on 26 March 2012 with enrolment beginning in August 2012. Masking was triple, covering participant, investigator and outcomes assessor.
• The registered primary outcome is time from randomisation to recurrence of any mood episode: hazard ratio 0.451 (95% CI 0.299 to 0.678). The 2026 paper prints no hazard ratio; a footnote to its Table 1 names the primary endpoint and refers to the head report (Calabrese et al., J Clin Psychiatry 2017;78(3):324–331, doi 10.4088/JCP.16m11201). Its own main efficacy outcome is the proportion free from recurrence, a key secondary endpoint of the trial, because NNT can only be computed from a proportion: 97 of 132 (73.5%) against 65 of 133 (48.9%), NNT 5 (95% CI 3 to 8).
• Harm was measured as discontinuation due to a treatment-emergent adverse event, 23 of 132 (17.4%) against 34 of 133 (25.6%), a calculated NNH of minus 13 printed in Table 2. Discontinuations by term were led by bipolar disorder (7 against 12), mania (3 against 11) and depression (5 against 6), which is 29 of the 34 on placebo, and the discussion attributes the higher placebo rate to disease worsening. The methods state in advance that a negative NNH is replaced by an imputed value of 1000. Table 3 carries four benefit-harm ratios, all of them built on discontinuation: 200 for any adverse event, 13.2 for akathisia, 26.4 for sedation and 26.4 for weight increased. The abstract carries the first of the four.
• Table 2 also prints NNH for the incidence of those events, and none of those values enters Table 3: akathisia 28 of 132 (21.2%) against 17 of 133 (12.8%), NNH 12; weight increased 31 (23.5%) against 24 (18.1%), NNH 18; sedation 8 (6.1%) against 1 (0.8%), NNH 19; weight gain of 7% or more, NNH 20. Against the same NNT of 5 these give 2.4, 3.6, 3.8 and 4.0. Separately, the serious events run the other way, 10 of 132 on drug against 25 of 133 on placebo, with mania recorded as a serious event twice against ten times; those counts sit in the registry and appear nowhere in this paper.
• The authors state their headline as follows: "Data suggest a favorable benefit-risk profile for AOM 400 versus placebo as a long-term maintenance treatment for BP-I." The Funding sources section names Otsuka Pharmaceutical Development & Commercialization Inc. and Lundbeck LLC, and states that the sponsors were involved in the design of the study, the collection, analysis and interpretation of data, the writing and reviewing of the article, and the decision to submit it for publication. Writing and editorial support came from the Prime Group of Companies in Knutsford, funded by the same two sponsors. In the CRediT statement all five authors are credited with writing, review and editing, and no author is credited with the original draft. Four of the five are employees or consultants of Otsuka or Lundbeck; the fifth declares consulting and speaking work for both.

The authors conclude that "Data suggest a favorable benefit-risk profile for AOM 400 versus placebo as a long-term maintenance treatment for BP-I", and all four of the paper's benefit-harm ratios are built on who stopped attending, while the akathisia in 21.2% of patients and the weight gain in 23.5% are printed one table earlier and never enter them.

The full text, its three tables and the supplementary material were read. This is a post hoc analysis of a trial run between August 2012 and April 2016, so the data are a decade old while the publication is recent. Three limitations that bear on the reading above are the authors' own: discontinuation as the safety outcome may underestimate the burden of adverse effects that did not lead to discontinuation; stabilising every participant on oral aripiprazole and then on the injection before randomisation enriched the sample for people who tolerate the drug, for which they cite an open-label comparison of 44 of 379 (11.6%) in aripiprazole-naive against 3 of 85 (3.5%) in previously treated patients; and the imputed value of 1000 is arbitrary. The analyses were descriptive and every value rests on a dichotomised outcome. Most confidence intervals in Table 2 are not significant, including those for akathisia and weight increased, so the incidence-based ratios are a comparison of point estimates. Denominators are 132 and 133 against 266 randomised, and 128 and 132 for weight gain of 7% or more. The registry lists 76 sites while its own recruitment note reports 103; the serious-event counts and the hazard ratio are in the registry and the head report, not in this paper.