Published on 12 September 2026, updated on 14 September 2026
Schizophrenia non-response: what comes after the first antipsychotic?
In brief
What do you do when the first antipsychotic is not enough? This network meta-analysis pools 59 randomised trials and 5,409 patients to compare the available next steps. Two of them show a signal, antipsychotic combination therapy and electroconvulsive therapy as augmentation, but at very low certainty and, for the first, with a clear rise in adverse events. The others are inconclusive, dose escalation and switching to clozapine included. The authors do not say that nothing works. They say the evidence is missing for almost everything that is done in practice.
The context
The sequence is familiar. An antipsychotic is prescribed at an adequate dose for several weeks, the response falls short, and a decision has to be made. The options are well known and appear in the guidelines: raise the dose, switch molecule, combine two antipsychotics, move to clozapine, add cognitive behavioural therapy for psychosis, or turn to electroconvulsive therapy or transcranial magnetic stimulation. They are laid out in sequence in the NICE schizophrenia decision tree, from first episode to clozapine.
What was missing was a comparison of all these strategies inside a single evidence network, with an explicit judgement of the confidence each one deserves.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Patients with persistent symptoms after at least one adequate four-week antipsychotic trial. 59 trials, 5,409 participants, mean age 39.8 years, 3,416 men and 1,441 women. 5,013 patients contributed to the primary analysis | |
| Interventions | |
| Dose escalation, switching antipsychotic, switching to clozapine, antipsychotic combination therapy, augmentation with xanomeline-trospium, with cognitive behavioural therapy for psychosis, with electroconvulsive therapy or with transcranial magnetic stimulation | |
| Comparator | |
| Continuing the same antipsychotic | |
| Primary outcome | |
| Change in overall symptoms, as a standardised mean difference | |
| Design | |
| Systematic review and random-effects network meta-analysis, masked raters required, certainty rated with CINeMA, protocol pre-registered on OSF · CEBM 1a |
The search covers the Cochrane Schizophrenia Group registry up to 13 January 2025 and PubMed up to 8 January 2026.
The findings
| Strategy | Result |
|---|---|
| Antipsychotic combination therapy | 18 trials, 575 patients. SMD −0.25 (95% CI −0.46 to −0.04), very low certainty |
| PEB readingWeak signal adverse events increased, odds ratio 1.93 (1.26 to 3.00) | |
| Electroconvulsive therapy as augmentation | 5 trials, 97 patients. SMD −0.51 (−0.96 to −0.06), very low certainty |
| PEB readingLargest apparent effect, on the smallest sample | |
| Cognitive behavioural therapy for psychosis as augmentation | 5 trials, 340 patients. SMD −0.23 (−0.58 to 0.11), very low certainty |
| PEB readingWeak evidence, interval including no effect | |
| Dose escalation | Inconclusive |
| PEB readingNo evidence in support which is not evidence against | |
| Switching to clozapine | Inconclusive |
| PEB readingNo evidence in support in this setting which is not evidence against | |
| Switching antipsychotic, transcranial magnetic stimulation, xanomeline-trospium | Inconclusive |
| PEB readingInsufficient data | |
One secondary result deserves attention: no evidence of a subgroup difference appears between patients who fail to respond to a non-clozapine antipsychotic and those who fail to respond to clozapine. On a subgroup analysis that is necessarily underpowered, this does not demonstrate that no difference exists.
The clozapine result, stated correctly
The study concludes that no evidence supports switching to clozapine in this population. Two readings would be equally wrong.
Concluding that clozapine is ineffective: nothing here tests it conclusively in this setting, and absence of evidence is not evidence of absence. Concluding the reverse, that the reflex of an early switch to clozapine is now vindicated: the population of this meta-analysis, defined by a single non-response after four weeks, is not treatment-resistant schizophrenia in the sense of the consensus criteria, which require at least two documented and adequate failures.
Critical appraisal
| Domain | Judgement |
|---|---|
| Certainty of the evidence | Very low |
| FindingVery low certainty for the three quantified comparisons, and results the authors themselves describe as very uncertain overall. No conclusion can rise above the status of a possible trend | |
| Sample size per arm | Reservation |
| Finding97 patients for electroconvulsive therapy, 340 for cognitive behavioural therapy for psychosis, 575 for antipsychotic combination. The only signals in the network rest on thin branches | |
| Definition of the population | Reservation |
| FindingA single four-week non-response is enough: a broader criterion than treatment resistance, so the results do not apply directly to patients with established resistance | |
| Head-to-head comparisons | Reservation |
| FindingThe authors themselves call for more trials comparing the strategies directly against one another | |
| Conduct | Sound |
| FindingPre-registered protocol, masked raters required, certainty rated systematically and stated without euphemism | |
| Funding | Sound |
| FindingGerman public funding (the German Research Foundation and the federal ministry), with additional support from a Japanese private foundation and two doctoral fellowships | |
| Declared interests | Worth flagging |
| FindingTwo authors out of eleven declare consulting or lecture honoraria, the last author from sixteen entities and one co-author from two. Three declare a grant or a research fellowship, the remaining six declare none. The direction of the conclusions does not follow that interest: no evidence for dose escalation, none for switching to clozapine, and an excess of adverse events reported for antipsychotic combination | |
Level of evidence
PEB appraisal: high-grade method, very low certainty on the results. The contradiction is only apparent. The work is rigorous, but it pools a literature that is thin and rarely comparative. The scientific score rewards conduct and transparency, not the strength of the clinical conclusions, which the authors themselves call very uncertain.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ What strikes me is not the result, it is the emptiness. We do this every day and nobody has run the trials. I will stop pushing the dose out of reflex, but I am not going to give up on clozapine in a genuinely resistant patient on the grounds that this meta-analysis does not test it. ”
What this means in practice: the message is about the state of the evidence, not about the treatments being ineffective. It moves the decision towards an explicit shared choice with the patient.
What you can do with this on Monday morning.
- Stop raising the dose by reflex when a patient does not respond. The authors find no evidence supporting this practice, exactly as for switching to clozapine, and in both cases that signals a gap in the data rather than a demonstrated lack of effect. The difference is that dose escalation exposes the patient without an established benefit.
- Document the non-response before any change. The included trials require at least four weeks at an adequate dose. The consensus criteria for treatment resistance are more demanding, with at least six weeks per trial and verified adherence, and those are the criteria that count before clozapine is considered.
- Keep clozapine for genuine resistance, after at least two documented failures, rather than as a reflex after the first.
- Discuss cognitive behavioural therapy for psychosis as an add-on. The signal is weak and its confidence interval includes no effect, but it is the only strategy in the network for which the study reports no excess of adverse events.
- Tell the patient what is true. At this precise step the evidence is limited for every option, which makes the decision a shared choice rather than an algorithm.
- The course of action is set out in the NICE decision tree for schizophrenia in adults.
Frequently asked questions
Does this study weaken the case for clozapine in treatment-resistant schizophrenia?
No. It finds that no conclusive evidence exists for switching to clozapine after a single four-week non-response. Its place in established resistance, defined by at least two adequate trials, rests on a separate literature and on different populations, which this meta-analysis does not cover.
Should two antipsychotics be combined?
The combination shows a statistically significant but modest effect, at very low certainty, with an odds ratio of 1.93 for adverse events. The authors write that combination therapy and ECT augmentation “might be considered, but with substantial caution”. This is not a first-line strategy.
Why does electroconvulsive therapy show the largest effect without being recommended here?
Because that effect rests on five trials and 97 patients, at very low certainty. A large apparent effect on a thin sample is not proof.
What does very low certainty actually mean?
That the true effect may be substantially different from the estimated effect. This judgement is not about the quality of the meta-analysis: it aggregates the risk of bias in the trials, the imprecision of the estimates, heterogeneity, and the coherence of the network.
Annotated bibliography
Furukawa Y, Salahuddin NH, Wei Y, et al. (2026). Next-step treatment for schizophrenia non-responsive to antipsychotics: a systematic review and network meta-analysis. eClinicalMedicine, 96, 103988. DOI 10.1016/j.eclinm.2026.103988 · PMID 42232685. Source study analysed here. Protocol pre-registered on OSF. Funded by the German Research Foundation and the German federal research ministry, with additional support from a Japanese private foundation and two doctoral fellowships.