Published on 21 September 2026

Analysis · Schizophrenia and psychotic disorders · Child and adolescent psychiatry

Early-onset psychosis: what thirty meta-analyses say about prognosis and antipsychotic choice

◆ Collection
Journal of the American Academy of Child and Adolescent Psychiatry · 2024;63(7):684-697 · Salazar de Pablo G et al.
DOI 10.1016/j.jaac.2023.10.016
PMID 38280414
Scientific 73
Editorial 78

Editorial collections PEB Academy The Monday Morning Patient

The essentials

This umbrella review brings together thirty meta-analyses on early-onset psychosis, defined as onset before age eighteen. The authors report 373 individual studies and 25,983 participants, a mean age of 15.1 years, and 38.3% female participants. Both totals are simple arithmetic sums of the sample sizes reported by the thirty meta-analyses, with no deduplication: the same pediatric trial can be counted more than once. The synthesis finds that 12.3% of psychotic disorders are identified before age eighteen, that 60.1% of subjects have a poor outcome, and that cognitive deficits are more pronounced than in healthy controls and than in subjects with adult- or late-onset psychosis. On treatment, clozapine is the antipsychotic with the highest efficacy on global, positive, and negative symptoms, with a standardized mean difference of 0.85 (95% confidence interval 0.75 to 0.95) against other antipsychotics in treatment-resistant forms. Olanzapine is the molecule associated with the greatest weight gain, a standardized mean difference of 1.2 versus placebo. Two caveats should guide the reading. The work recalculates no effect of its own; it narratively reports what others have already calculated, and it therefore depends entirely on the quality of the meta-analyses it draws on, which is uneven. Five of the fourteen authors report ties to antipsychotic manufacturers, which calls for particular caution in the chapter on comparative tolerability.

Context

An adolescent presenting with a first psychotic episode is not simply a younger adult. Schooling is still in progress, the family is directly involved in care, metabolic sensitivity to antipsychotics is higher, and the literature covering this population is markedly thinner than for adults. In practice, many decisions are extrapolated from adult data, for lack of anything better.

The question this work answers therefore comes before any clinical decision: what evidence actually exists for this population. The answer is a map, with its documented areas and its blank spaces. The authors present their work as the first overall meta-analytic synthesis on this population, a claim that appears verbatim in the publication’s abstract. That is what makes this a reference article worth keeping rather than a piece of news to comment on.

The study at a glance

Question, population, measure
Population
Children and adolescents with early-onset psychosis, defined as onset before age eighteen. Mean age 15.1 years, 38.3% female, affective and non-affective psychoses pooled together across the meta-analyses reviewed.
Scope covered
Epidemiological and risk factors; biological markers including neuroimaging, biochemistry, and neurophysiology; cognition; prognosis; response to interventions; adverse effects and acceptability.
Comparisons
Between antipsychotics, against placebo or a control condition, and against subjects with adult- or late-onset psychosis.
Outcomes
Prognosis; overall, positive, and negative symptomatic efficacy; tolerability including weight gain and prolactin levels; cognitive performance.
Design
Umbrella review of thirty meta-analyses, reporting 373 individual studies and 25,983 participants, both totals being non-deduplicated sums. Protocol prospectively registered on PROSPERO under number CRD42022350868, stated compliance with PRISMA reporting guidelines, quality of included reviews assessed with the AMSTAR-2 tool. Results reported narratively, grouped by domain.

Quality control

Point checked	Judgment
Protocol registered before analysis	Yes
FindingProspective registration on PROSPERO, number CRD42022350868, systematic literature search across multiple databases and registers, with selection, extraction, and quality assessment all conducted independently.
Quality assessment of included reviews	Conducted with AMSTAR-2
FindingThe authors score each meta-analysis on the tool’s sixteen items. Individual scores appear in the supplementary tables, but no overall score is given in the publication’s abstract. Recalculated from those tables, scores range from 4 of 16 to 16 of 16, with a mean of 11.4 and a median of 11.5. Nine of the thirty reviews score 9 of 16 or below. The mean therefore conceals a subset of clearly weak reviews, two of which supply results that are frequently cited on tolerability.
New calculation of effects	Absent
FindingNo new statistical pooling, no recalculated confidence interval; results are reported narratively. The figures cited inherit the strengths and weaknesses of their original sources.
Authors’ conflicts of interest	Five of fourteen authors
FindingFive authors report industry ties, two of them with very extensive lists of companies, and nine report no ties at all. The chapter most sensitive to this configuration is the tolerability ranking between molecules. The detailed list appears in the annotated bibliography below.
Funding of the work	None declared
FindingThe authors state they received no funding for this work. However, the funding of the primary studies pooled by the meta-analyses is not summarized either in the version reviewed or in the supplementary tables, so the upstream chain of possible influence cannot be traced.
Literature search window	Closing date not specified here
FindingThe date of the last database search does not appear in the version reviewed. Protocol registration dates to 2022 and publication to 2024, which brackets the search without dating it precisely. The most recent work is not covered, which is inherent to the format.

Results

60.1%

Proportion of subjects with early-onset psychosis classified as having a poor outcome. The definition of poor outcome used is not given in the version reviewed. The prognosis domain rests on a single meta-analysis, with a mean follow-up of 14.4 years, drawing on cohorts largely predating current models of care.

Result	Value
Share of psychotic disorders identified before age eighteen	12.3%
ReadingFigure taken from a meta-analysis of 36 studies on the age of onset of mental disorders, which also gives 3% before age fourteen. This is an age of detection, not a biological age of onset. A minority but not a marginal population, which partly explains the thinness of the literature devoted to it.
Long-term functional outcome	60.1% poor outcome
ReadingThe prognosis domain of this umbrella review rests on a single meta-analysis, 21 longitudinal studies and 422 subjects, scored 8 of 16 on AMSTAR-2, with follow-up ranging from 1.5 to 42 years, a mean of 14.4 years, standard deviation 11.4, and loss to follow-up ranging from 0% to 59%. A weighty figure, to be handled with care in front of a family: it pools old cohorts and does not predict any individual patient’s trajectory.
Cognition, comparison with healthy controls	Standardized mean differences from 0.58 to 1.3 depending on domain
ReadingValues taken from a meta-analysis of 12 studies, ranging from motor function, 0.58, to processing speed, 1.3. Deficits are broad and diffuse, affecting every domain examined.
Cognition, comparison with adult- or late-onset psychosis	Difference reported, without a quantified effect size
ReadingThe meta-analysis concerned reports more pronounced deficits in arithmetic, IQ, psychomotor speed, and verbal memory, but only as heterogeneity test statistics, along with comparable deficits in visual memory and attention. No standardized mean difference is provided for this comparison in the material reviewed. Causal interpretation is not possible, since age of onset and severity are intertwined.
Clozapine in treatment-resistant forms	Standardized mean difference of 0.85, 95% CI 0.75 to 0.95
ReadingValue taken from a meta-analysis of 15 trials comparing clozapine with other antipsychotics in resistant subjects, with no detected heterogeneity. Two other meta-analyses point the same way with different values: 1.6 versus placebo in a network meta-analysis of 28 trials, and 0.48 (95% CI 0.04 to 0.91) across 3 trials against olanzapine and haloperidol. The agreement concerns the direction of the effect, not its exact magnitude.
Weight gain on antipsychotics	Olanzapine 1.2, clozapine 0.92, quetiapine 0.85 versus placebo
ReadingStandardized mean differences from the most extensive network meta-analysis. Only olanzapine exceeds one standard deviation. Next come paliperidone at 0.73, risperidone at 0.61, asenapine at 0.43, and aripiprazole at 0.29; no significant weight gain is found for molindone, ziprasidone, lurasidone, fluphenazine, or haloperidol. In absolute terms, another meta-analysis finds an average of 4.5 kg under second-generation antipsychotics versus 1.4 kg under first-generation agents. A major magnitude in subjects who are still growing, and one of the most directly usable findings of this work.
Lurasidone’s weight profile	Less weight gain than five comparators
ReadingMean differences of 3.62 kg versus olanzapine, 2.13 kg versus quetiapine, 1.16 kg versus risperidone, 0.98 kg versus asenapine, and 0.85 kg versus paliperidone. This result comes from a network meta-analysis scored 9 of 16 on AMSTAR-2, two of whose authors are affiliated with the company that markets the molecule. It should be read with that caveat in mind, and weighed against a finding from another meta-analysis included here, a fivefold increase in the risk of akathisia versus placebo under lurasidone.
Non-pharmacological interventions	No meta-analyzed efficacy data
ReadingA single meta-analysis, 2 trials and 91 participants, addressing cognitive remediation, and its only reported result concerns dropout, not efficacy. This is an absence of analysis, not an absence of effect. The authors themselves conclude that further meta-analytic work is still needed on this point.
Oxidative stress and inflammation	No difference found, across 4 studies
ReadingA negative result for catalase, glutathione, glutathione peroxidase, superoxide dismutase, total antioxidant status, and cellular oxidative damage. It rests on 4 studies and 261 subjects, with high heterogeneity and meta-regressions that were planned but not carried out for lack of studies. This absence of signal does not amount to proof that no abnormality exists.

Critical appraisal

Domain	Judgment
Usefulness of the synthesis	High
FindingThe format delivers, in a single read, the state of knowledge on a poorly documented population, and flags the domains where nothing is established.
Dependence on the sources it draws on	Total
FindingAn umbrella review never improves the quality of the work it assembles. An error present in an included meta-analysis carries through here with no possibility of correction, all the more so because results are reported narratively.
Overlap of primary studies	Non-deduplicated sums
FindingThe total of 373 studies and of 25,983 participants matches exactly the sum of the sample sizes reported by each of the thirty meta-analyses in the supplementary tables. No deduplication is performed. Ten meta-analyses, moreover, appear in both the efficacy table and the tolerability table, and several draw on the same small pool of pediatric randomized trials. The apparent convergence is therefore overstated, and the exact degree of overlap is not quantified by the authors.
Conflicts of interest	Worth flagging
FindingThey do not disqualify the work; they point vigilance toward the passages where a given molecule comes out comparatively favored. The most exposed point is lurasidone’s weight profile, whose primary source is partly authored by the company concerned.
Scope of the prognostic figure	Descriptive
FindingIt rests on a single meta-analysis of low methodological quality, describing cohorts followed for more than a decade on average, under models of care that no longer apply today. It cannot ground any individual prognosis.
Internal consistency on tolerability	Partial
FindingThe abstract concludes that the risk of treatment discontinuation, for any cause or for adverse effects, is low or comparable to placebo. Yet one of the included meta-analyses reports, for discontinuation due to adverse effects, a relative risk of 12.8 for clozapine and 5.1 for haloperidol. The abstract’s general wording smooths over this spread.

Level of evidence

Scientific73

Editorial78

An umbrella review with no new statistical pooling, drawing on meta-analyses of both randomized trials and observational studies. Standard levels-of-evidence scales have no dedicated rank for this format: assigning one is an editorial judgment, not a datum from the source itself. What is demonstrated here is the mapping of the field and the identification of its gaps. What is suggested is the hierarchy of efficacy between molecules, consistent in direction across sources but variable in magnitude. What belongs to expert opinion is applying the prognostic figure to a patient seen today, something nothing in this work supports.

The colleague test

What an experienced colleague might say about this study in two minutes, between two consultations.

“What I take from this is that clozapine comes out on top in these kids once two other drugs have failed, and that we underestimate what olanzapine makes them gain. The prognosis figure, I don’t bring that one out in front of parents.”

The first point echoes the authors’ own conclusion, who recommend considering clozapine in subjects resistant to two non-clozapine antipsychotics. The idea of a typical delay before starting it, by contrast, belongs to expert opinion: this work measures no prescribing delay at all. Translated for practice: factor in metabolic risk from the very choice of the first antipsychotic, and use the overall prognosis figure to calibrate the intensity of follow-up, not to forecast a future.

Regulatory status of the molecules discussed

A molecule that ranks well in an international synthesis is not necessarily an option available to every prescriber, and the age range covered by its marketing authorization often differs from the age range actually studied. Regulatory status varies from one country to another and should always be checked before prescribing; three points verified on August 13, 2026, using France as one illustrative example, are worth flagging here. In France, aripiprazole holds a marketing authorization in adolescent schizophrenia from age fifteen, its summary of product characteristics stating that use below fifteen is not recommended for lack of sufficient data; the age threshold and indication should be checked against the equivalent authorization in your own country. Across Europe, asenapine holds authorization only for moderate to severe manic episodes of bipolar I disorder in adults, so neither in schizophrenia nor in adolescents, a restriction that applies broadly, not only in France. Molindone, which comes out favorably on weight in two of the meta-analyses reviewed here, is not an accessible option anywhere in Europe: production was discontinued in 2010 by its sole United States distributor, and it holds no European marketing authorization.

Lurasidone has held a European marketing authorization since March 21, 2014, later extended to adolescents from age thirteen. In France, the Haute Autorité de Santé’s transparency committee issued an opinion on the adult indication on November 19, 2014, rating the added clinical benefit as significant but with no improvement over available alternatives. Its actual availability and reimbursement conditions, in France as in any other country, should be checked at the time of prescribing.

Clozapine monitoring rules illustrate how much these details can change, and how recently. In France, since April 1, 2025, initial prescription is no longer restricted to hospital settings: it can be started in the community by a specialist in psychiatry, neurology, or geriatrics. On September 8, 2025, the French national medicines safety agency (ANSM) issued updated hematologic monitoring rules centered on absolute neutrophil count, with total white cell count no longer required. The revised schedule is weekly from week 1 to week 18, monthly from week 19 to week 52, every twelve weeks during the second year, then annually beyond two years in patients with no history of neutropenia. Starting treatment requires an absolute neutrophil count of at least 1,500 per cubic millimeter, with the threshold lowered to 1,000 in confirmed benign ethnic neutropenia, and discontinuation below 500. Whatever your jurisdiction, the indication, age limits, and current hematologic monitoring protocol should be checked against local guidelines and the relevant summary of product characteristics before prescribing to a minor; the French schedule above illustrates the level of detail to expect, not a universal standard.

What you can do with this

After two adequately conducted trials of non-clozapine antipsychotics without sufficient response, raise the question of clozapine rather than moving on to a third molecule: this is the authors’ explicit recommendation.
Put weight, glucose, and lipid monitoring in place from the start of treatment, alongside a growth curve: in an adolescent, weight gained is not easily lost.
Factor in weight risk from the very choice of the first antipsychotic; olanzapine is the molecule most heavily associated with this risk across these syntheses.
For every molecule considered, verify the indication, the authorized age range, and the current monitoring requirements in your own jurisdiction, including the absolute neutrophil count monitoring schedule for clozapine, revised in France in September 2025.
With a family, say plainly what is known and what is not: individual variability is large, and the prognostic data come from old cohorts.
Include a cognitive component in follow-up, since schooling is the major functional stake at this age, while keeping in mind that no meta-analyzed efficacy data yet exist for non-pharmacological interventions in this population.
The course of action is set out in the NICE decision tree for schizophrenia in adults.

Frequently asked questions

Does the 60.1% poor-outcome figure apply to an adolescent seen today?

Not directly. It comes from a single meta-analysis of cohorts followed for a mean of 14.4 years, under earlier models of care, and the definition of poor outcome used is not given in the version reviewed. It signals a stake, not an individual probability.

Should clozapine be offered earlier than in adult practice?

The authors recommend considering it after resistance to two non-clozapine antipsychotics, without specifying a timeline particular to adolescents. It carries the strongest efficacy signal in the synthesis, but one of the included meta-analyses also places clozapine first for discontinuation due to adverse effects. The decision remains bound by locally applicable prescribing and monitoring requirements, of the kind described above for France.

Can the tolerability ranking between molecules be trusted?

With reservation. The work recalculates no effect of its own, the rankings come from meta-analyses of uneven quality, ranging from 4 to 16 of 16, and five of the fourteen authors report ties to manufacturers. The broad orders of magnitude are useful; the fine-grained ranking is less so.

Do the more pronounced cognitive deficits mean early-onset psychosis damages the brain more?

This cannot be demonstrated here. Early onset is associated with more severe forms of illness, and the observed association does not allow the two to be separated. The comparison with adult- or late-onset psychosis is, in fact, reported without any quantified effect size.

What about molecules discussed here that are not available where you practice?

Check local status before drawing on any of these figures. Molindone is no longer produced, asenapine has no European authorization in schizophrenia or in adolescents, and lurasidone’s actual availability should be checked in your own country, France included. A molecule that ranks well in an international synthesis may not be an option you can actually prescribe.

Does the absence of a result on inflammation mean there is nothing to look for there?

No. The negative result rests on 4 studies and 261 subjects, with high heterogeneity and planned complementary analyses that were abandoned for lack of data. This is insufficient power, not a refutation.

Annotated bibliography

Source study. Salazar de Pablo G, Rodriguez V, Besana F, Civardi SC, Arienti V, Maraña Garceo L, Andrés-Camazón P, Catalan A, Rogdaki M, Abbott C, Kyriakopoulos M, Fusar-Poli P, Correll CU, Arango C. Umbrella Review: Atlas of the Meta-Analytical Evidence of Early-Onset Psychosis. Journal of the American Academy of Child and Adolescent Psychiatry. 2024 Jul;63(7):684-697. Epub 2024 Jan 25. DOI 10.1016/j.jaac.2023.10.016. PMID 38280414. Open-access publication.

Protocol registration. Prospectively registered on PROSPERO, number CRD42022350868, under the title Early Onset Psychosis: Umbrella Review on Diagnosis, Prognosis and Treatment factors.

Funding. The authors declare they received no funding for this work. Two authors additionally report personal institutional support: Instituto de Salud Carlos III and the Spanish Ministry of Economy and Competitiveness for one author, and Italy’s National Recovery and Resilience Plan together with the MNESYS project for another.

Declared conflicts of interest. Five of the fourteen authors report ties. Salazar de Pablo: personal fees from Janssen Cilag and Menarini. Catalan: personal fees from Janssen, support from Instituto de Salud Carlos III and the Spanish Ministry of Economy and Competitiveness. Fusar-Poli: research funding or personal fees from Lundbeck, Angelini, Menarini, and Boehringer Ingelheim, support from the NextGenerationEU program and the Italian Ministry of University and Research. Correll: consultant, advisor, or honoraria recipient for AbbVie, Acadia, Alkermes, Allergan, Angelini, Aristo, Boehringer Ingelheim, Cardio Diagnostics, Cerevel, CNX Therapeutics, Compass Pathways, Darnitsa, Gedeon Richter, Hikma, Holmusk, IntraCellular Therapies, Janssen and Johnson & Johnson, Karuna, LB Pharma, Lundbeck, MedAvante-ProPhase, MedInCell, Merck, Mindpax, Mitsubishi Tanabe Pharma, Mylan, Neurocrine, Newron, Noven, Otsuka, Pharmabrain, PPD Biotech, Recordati, Relmada, Reviva, Rovi, Seqirus, SK Life Science, Sunovion, Sun Pharma, Supernus, Takeda, Teva, and Viatris; expert testimony for Janssen and Otsuka; data safety monitoring board membership for Lundbeck, Relmada, Reviva, Rovi, Supernus, and Teva; grants from Janssen and Takeda; royalties from UpToDate; stock options in Cardio Diagnostics, Mindpax, LB Pharma, and Quantic. Arango: Spanish and European public funding, consultant or recipient of honoraria or grants from Acadia, Angelini, Boehringer, Gedeon Richter, Janssen Cilag, Lundbeck, Minerva, Otsuka, Pfizer, Roche, Sage, Servier, Shire, Schering Plough, Sumitomo Dainippon Pharma, Sunovion, and Takeda. The remaining nine authors, Rodriguez, Besana, Civardi, Arienti, Maraña Garceo, Andrés-Camazón, Rogdaki, Abbott, and Kyriakopoulos, report no biomedical financial interest or potential conflict of interest.

Regulatory sources cited. Summary of product characteristics for oral aripiprazole, indication in schizophrenia from age fifteen. European Medicines Agency registry entries for lurasidone, authorized March 21, 2014, and for asenapine, indication limited to manic episodes of bipolar I disorder in adults. Opinion of the Haute Autorité de Santé’s transparency committee on lurasidone, November 19, 2014. Communication from France’s National Order of Pharmacists on the opening of community-based initial clozapine prescription from April 1, 2025. Safety communication from the French national medicines safety agency (ANSM) on absolute neutrophil count monitoring under clozapine, September 8, 2025. Discontinuation notice for molindone in the United States, 2010. All of these sources were consulted on August 13, 2026.

What was reviewed. Verification was carried out on the published version of the article, the publisher page giving the title, the complete author list, the bibliographic reference, the structured abstract, the plain-language summary, author contributions, protocol registration, the funding statement, and the conflict-of-interest declarations. The supplementary material provided by the authors was reviewed in full: it includes the PRISMA 2020 compliance checklist, six tables of characteristics and results covering epidemiology, biological markers, cognition, prognosis, response to interventions, and adverse effects, the list of the sixteen AMSTAR-2 items, meta-analysis-by-meta-analysis quality scores, and the references for the thirty included meta-analyses. The totals of 373 studies and 25,983 participants were recalculated from these tables, as were the mean and median AMSTAR-2 scores. The full text of the publication, its figures, and its detailed methods section were not reviewed: the closing date of the literature search, the definition of poor outcome used, and the funding of the primary studies cannot therefore be reported here. The French and European regulatory points were verified against the sources cited above on August 13, 2026. The article underwent independent double reading.

Editorial collections

PEB Academy The Monday Morning Patient