Brexpiprazole for adolescent schizophrenia: what is a statistically positive PANSS result actually worth?

Context

Early-onset schizophrenia is a situation where therapeutic pressure is high and the evidence base is thin. Controlled trials dedicated to 13 to 17 year-olds are few, often small, and prescribing largely proceeds by extrapolation from adult data. The authors themselves recall that the US approval of brexpiprazole in adolescents rested on such extrapolation, supported by pharmacokinetic data and an interim open-label safety analysis. This is not a trivial shortcut: at this age, sensitivity to metabolic effects, sedation, hyperprolactinaemia, and motor effects differs from that of adults, and the consequences of an adverse effect on schooling, socialisation, and long-term adherence are heavier than in an established adult patient.

Brexpiprazole belongs to the family of D2 receptor partial agonists, alongside aripiprazole and cariprazine. This point of pharmacology sits outside the scope of the trial itself. The development case for this class rests less on a gain in efficacy than on a tolerability profile, in particular on akathisia and sedation. That is a plausible pharmacological promise, but a pharmacological promise is not a clinical result until a trial has measured it directly.

The question at hand is therefore easy to state and hard to settle: in adolescents, does brexpiprazole outperform placebo at six weeks, and by how much? A sufficiently large trial detects differences that no one would notice in a consultation. That is precisely the situation this publication puts on the table, and it is what makes it as much a teaching case as a piece of clinical data.

A note on method

What an active reference arm licenses you to say, and what it does not

The third arm, aripiprazole 10 to 20 mg per day, is not there to decide between two molecules. It serves a precise and limited purpose, called assay sensitivity: verifying that the trial, under its actual conditions of recruitment, assessment, and follow-up, is capable of detecting an effect where one is known to exist. If the reference drug fails to separate from placebo, the lesson is that the trial itself could not distinguish anything, and a null result on the tested drug becomes uninterpretable. It is a methodological safeguard, not a comparison. Here, the check works: aripiprazole separated from placebo by -6.53 points (95% CI -10.8 to -2.21), with a nominal p value of 0.0032.

The consequence is strict, and it is the most common trap in reading this type of trial. Any numerical comparison between brexpiprazole and aripiprazole, including on akathisia or somnolence, is descriptive. The authors state explicitly that the trial was neither designed nor powered for a direct head-to-head comparison between the two drugs, and that no testing hierarchy was planned. Writing that brexpiprazole causes less akathisia than aripiprazole would be an inferential error, even though the raw percentages point that way. The defensible statement is that, in this trial, observed rates of akathisia and somnolence were lower in the brexpiprazole arm than in the reference arm, without that difference having been tested.

Statistical significance and clinical relevance on the PANSS

The second methodological point structures the entire reading of this article. The PANSS is a 30-item scale whose total score ranges from 30 to 210. A difference of a few points on this scale has no perceptible translation in consultation. The authors themselves recall, in justifying the choice of primary endpoint, that anchor-based work using the Clinical Global Impression scale places the minimum clinically important difference at around 15 points, or a 34% reduction from baseline. Two caveats accompany this benchmark. It was estimated in adults with chronic schizophrenia, in the CATIE cohort, and its transposition to adolescents is not automatic. Above all, it characterises the improvement of a given patient, whereas -5.33 is a between-group difference: the two quantities do not map onto each other term for term.

With these caveats in mind, the finding holds, and is even reinforced by a benchmark internal to the trial itself. The sample size calculation targeted a difference of -7.4 points, with 105 patients per group and at least 80% power. The observed effect, -5.33 points, is smaller than the effect the investigators themselves expected, and roughly three times smaller than the clinical relevance benchmark they cite. Even the most favourable bound of the confidence interval, -9.55 points, remains below that benchmark. This is not a negative trial: the primary endpoint is met, the direction of effect is consistent, and the interval excludes zero. It is a positive trial with a small effect, and both statements need to hold in the same sentence.

The high placebo response, at -17.4 points, explains part of this compression. The authors place it at the upper end of the range expected from prior trials of atypical antipsychotics in adolescents, and suggest, as a hypothesis, that the three-arm design may have inflated it, since a participant had a two-in-three chance of receiving an active treatment and received two tablets. They also note that the idea of a stronger placebo response in children and adolescents than in adults rests on a scarce literature. This placebo response does not invalidate the result; it limits the measurable margin.

The study at a glance

Quality control

Results

The primary endpoint is met, and the confidence interval bounds the size of the possible gain more than it opens it up. On secondary endpoints, the response rate, the positive subscale, and the clinical global impression of improvement move in the direction of treatment, with nominal p values, while negative symptoms, remission, global functioning, and global clinical severity do not separate from placebo. The favourable result on general psychopathology comes from a post hoc analysis and can only carry exploratory weight.

This last set of findings calls for cautious wording. The trial was not powered for these endpoints, and the remission threshold is demanding at six weeks. What is established is that the trial did not demonstrate an effect on these dimensions, which is not the same as having demonstrated that there is none. The distinction determines what a longer trial, powered for functioning, might still show.

Critical appraisal

Authorisation, marketing, reimbursement

A preliminary point, outside the scope of this trial itself. The authors state that brexpiprazole is approved for schizophrenia in adults and in adolescents from age 13 in the United States and in a number of other countries, including Singapore, the United Arab Emirates, and Brazil. They report no European approval in this age group. Whatever the quality of a trial, it is not itself a marketing authorisation and it prejudges no decision by any medicines agency, wherever the reader practises. Three questions are logically distinct in any jurisdiction: whether a drug holds a marketing authorisation for the indication and age group in question, whether it is actually marketed there, and whether it is covered by collective reimbursement. These three planes do not automatically line up, and a positive answer on one says nothing about the other two. France offers a concrete illustration of this structure: verifying brexpiprazole’s marketing authorisation, actual commercial availability, and reimbursement status in French adolescents requires consulting the relevant official sources directly, since this trial, however well conducted, settles none of the three questions on its own. Readers practising outside France should check the equivalent three points against their own national regulator: the example is French, the structure of the problem is not.

Should the question of prescribing outside the marketing authorisation arise, most jurisdictions that regulate off-label use frame it as a matter of law, not of clinical judgement alone. In France, article L. 5121-12-1 of the public health code makes an off-label prescription conditional on the absence of an authorised, appropriate alternative, and attaches to it duties to inform the patient, to note the situation on the prescription, and to document the decision in the patient’s file. This wording should be checked in its version currently in force, as the text has been amended more than once. The mechanism it embodies, attaching accountability requirements to a prescription made outside its authorised indication, exists in some form in most jurisdictions that regulate off-label prescribing, even where the details differ. A trial demonstrating an effect discharges none of these obligations. Clinicians practising outside France should identify the equivalent requirement in their own regulatory framework before considering an off-label prescription on the strength of this trial alone.

On clinical substance, this trial does not shift the lines. It provides no formal comparison with the drugs used first-line in adolescents, and the size of the measured effect gives no argument for switching a stable patient. What it does contribute is a reassuring short-term tolerability signal, with no excess of adverse events over placebo across six weeks, in a drug class where akathisia and sedation weigh heavily on the adherence of a school-age patient. In PEB’s view, this finding can reasonably enter a second- or third-line discussion for a patient whose main problem is tolerability rather than efficacy, once the local regulatory and reimbursement status has been verified. This is an opinion, not a trial result: the study tested no treatment sequence.

A final word on monitoring. Six weeks says nothing about metabolic risk, the sensitive point in adolescents treated over the long term: monitoring of weight, waist circumference, and lipid and glucose profile remains the rule, and this trial provides no basis for relaxing it.

Level of evidence

PEB appraisal: confidence is high that brexpiprazole has an effect superior to placebo on PANSS total score at six weeks in adolescents, and that this effect is of small size. The design, the blinding, the sample size, and the successful assay-sensitivity check support this conclusion, as does the overlap of the adverse event profile with placebo over this duration.

Confidence is low, on the other hand, on three points. First, on any clinical translation of the measured effect: nothing indicates that a gap of this size is perceptible to the patient, the family, or the clinician. Second, on the favourable secondary endpoints, not adjusted for multiplicity, and even more so on the post hoc analysis of general psychopathology, which remain exploratory. Finally, on comparative tolerability with other partial agonists, which the trial did not test. The endpoints that do not separate from placebo, negative symptoms, remission, and functioning, are inconclusive and should not be presented as a demonstrated failure. The weak independence, with seven of nine authors employed or formerly employed by the two sponsors, does not alter the published data but justifies a careful reading of how the conclusions are framed.

Frequently asked questions

Editorial collections

Tags