The Cloz Supremacy

Is clozapine superiority for treatment-resistant schizophrenia real?

Clozapine has a special place in the psychopharmacologist toolkit. It is the only medication recommended specifically for ‘treatment-resistant schizophrenia’, defined as lack of response to at least two other antipsychotics at adequate doses. Whole clinical services are geared towards initiating patients on clozapine, patients with the severest forms of mental illness, in whom all else has failed.

I have written about the history of clozapine. Of the landmark trial led by John Kane which established its superiority in treatment-resistant illness. I have noted the serious side-effects - agranulocytosis, myocarditis, pneumonia, constipation - but emphasised that clozapine prescription is associated with reduced mortality.

I have rallied against scare-mongering headlines (Britain’s most dangerous prescription drug). I’ve applauded the loosening of monitoring restrictions, allowing greater access. I’ve seen first-hand the stability it can give patients with chronic psychosis, allowing them to live meaningful lives.

But a new meta-analysis published in The Lancet Psychiatry has called into question whether clozapine really is superior to other antipsychotics in treatment-resistant schizophrenia. If correct, it could overturn a central tenant in psychiatry, widely accepted as a ‘fact’: clozapine supremacy.

Beginning to see the Leucht

Efficacy of clozapine versus second-generation antipsychotics in people with treatment-resistant schizophrenia: a systematic review and individual patient data meta-analysis, is from the group of Professor Stefan Leucht, a top tier meta-analyst. He has already brought us seminal studies comparing efficacy and tolerability across antipsychotics, demonstrating the protective effect against relapse, showing that psychiatric medication has comparable effect sizes to the rest of medicine.

I’ve written about the motivations of researchers conducting meta-analyses before’ well it seems that unusually, Leucht’s personal motivation is to do high quality meta-analyses! His group is exactly who you want to conduct this study.

As you might expect, the analysis was pre-registered on PROSPERO and a detailed protocol published in advance. There were some deviations from this protocol, which are explained in their Appendix. They mostly relate to additional analyses, or analyses which were not possible due to a lack of data from the included studies.

The team sought to answer a specific question, is clozapine more effective than other second generation antipsychotics in treatment-resistant schizophrenia. We we can describe in a PICO (Population Intervention Comparator Outcome) format:

• Population: individuals with treatment-resistant schizophrenia

• Intervention: clozapine

• Comparator: any second generation antipsychotic

• Outcome: PANSS score at 6-8 weeks of treatment

They included any randomised trial with at least 80% of the trial population having schizophrenia (as opposed to other similar conditions like schizoaffective disorder). The definition of ‘treatment-resistance’ was taken from each trial and ranged from tight (insufficient response to two antipsychotics, confirmed prospectively) to relaxed (insufficient response to one antipsychotic). Interestingly, they blanketly excluded all trials from China due to concerns about their quality.

Before going further, it’s worth considering what this analysis will and will not be able to answer. It won’t answer whether clozapine is superior to other antipsychotics in schizophrenia in general (another study by the same group suggests it is). It won’t tell us whether clozapine is superior to first generation antipsychotics in treatment-resistance (this is taken as read). It will tell us whether clozapine is more effective in reducing symptoms in the short-term. It won’t tell us whether clozapine is superior to other second generation antipsychotics in outcomes like relapse prevention, overall functioning or mortality.

Let me say something else about the study question. It is important for clinical practice. We gear all our pathways of treatment-resistant schizophrenia to getting patients on clozapine. It is recommended by all the guidelines. This isn’t a trivial medication, even compared with other antipsychotics. It requires careful initiation to monitor for serious adverse cardiac or haematological effects then ongoing monitoring and treatment of longer term side-effects like constipation and hyper-salivation. Where I work, this means that patients are enrolled into a ‘clozapine clinic’ which provides the medication and monitoring.

If the superiority of clozapine is not proven, this whole treatment model is called into question.

Return to the meta-verse

As previously noted, meta-analyses are tricky and vulnerable to researcher-degrees of freedom. We need reassurance that the researchers haven’t simply picked the best methodology and models to support their own viewpoint. The two aspects which provide this reassurance are pre-registering intended analyses (so you can’t simply change these when the results are known) and running different analyses to see whether they converge on a similar set of results. That helps reassure us that these results are consistent across the meta-verse of analyses that could possibly be run.

Leucht’s team did just that, being transparent about the types of analysis they intended to run (publishing these ahead of time) and providing multiple alternative analyses. Their main methodology was Bayesian but they additionally ran more familiar frequentist models.

They attempted to get individual patient level data from included trials, rather than relying on aggregate measures like means and standard deviation. This provides the ability to examine factors which might modify the results. The authors did other clever things to increase power, such as inputing missing data and adding two studies that had to be analysed within protected research environments as informative priors. A nice touch is that they involved patients with lived experience in the design and interpretation of the study.

They identified 12 trials in which they were able to obtain individual patient level data and an additional 7 trials that only had aggregate data. The results of the primary analysis are shown below:

Reproduced from Schneider-Thoma et al. 2025 under CC-BY

These suggest clozapine did slightly less well than other antipsychotics, though the 95% credible interval crosses the line of no effect (I think Bayesians will beat me up if I say ‘did not reach statistical significance’). As might be expected, duration of illness had a worse prognosis, while more severe symptoms at baseline was a good prognostic factor (think regression to the mean). Olanzapine performed best but again did not reach statistical significance had wide credible intervals.

There were negligible differences between groups for the secondary outcomes. There was also no difference when using all published (aggregate data) rather than individual patient data.

For the frequentists in the audience, here is a more traditional forest plot using individual patient data, provided in the Supplementary material.

Reproduced from Schneider-Thoma et al. 2025 under CC-BY

Again the overall effect (diamond at the bottom) crosses the middle line of no effect, meaning there was not a statistically significant difference between the groups. If anything, it looks like the results are leaning against clozapine.

The authors conclude:

It is noteworthy that an IPD meta-analysis of randomised controlled trials, which is methodologically considered a high standard of evidence, did not provide evidence for the preferential use of clozapine in treatment-resistant schizophrenia and therefore uncertainty remains regarding superiority of clozapine.

Cope-azine, stat dose

If the results of this analysis are true and there is truly no superiority of clozapine in treatment-resistant schizophrenia, have we all been wasting our time? Have we been exposing patients to unnecessary side-effects? Are the supposed benefits of clozapine from observational studies perhaps explainable by more the frequent follow-ups with health professionals which are required for monitoring?

Overturning guidelines is not something that will be done by a single meta-analysis, no matter the credibility of the authors. The only immediate change in practice might be cementing that patients have an adequate trial of a second generation antipsychotic (ideally olanzapine) before considering clozapine.

An accompanying editorial from three leading clozapine experts - Dan Siskind, Korinne Northwood and Rob McCutcheon - describes reservations. Individual patient data was only available for 1052 of 1599 eligible participants (small in the world of meta-analyses). They note that lower doses of clozapine were used in trials run by drug companies. Unfortunately this type of shenanigans is not uncommon, pharma companies may give their new drug the best chance of success by comparing it to clozapine at a lower than optimal dose. The trials did not report plasma levels of clozapine, which is standard in clinical practice to ensure the drug is within a therapeutic range.

These experts point to the overwhelming observational evidence in favour of clozapine, ending with the statement:

We believe clozapine remains the most effective medication for treatment-resistant schizophrenia.

My ideal trial

Leucht and his team don’t call for the position of clozapine to be down-graded. Instead, in highlighting the uncertainty of evidence in favour of clozapine (we some times say clinical equipoise), they call for the science to settle the evidence in the only way we know how, by experiment.

You’re asking me how I would design a trial for clozapine in treatment-resistant schizophrenia, given unlimited funding? OK I’ll tell you.

Firstly, the trial should prospective confirm non-response to at least two antipsychotics, ideally with plasma levels (to rule out non-adherence), one of which should be risperidone. Participants should then be randomised to either clozapine or olanzapine. Study investigators and patients would have to be blinded, meaning that the additional physical health monitoring would have to be done for both groups and participants unblinded only in medical emergencies. Dosing would have to be guided by plasma levels, so should be independent from the study investigators. The trial should be adequately powered to detected a meaningful difference between groups; in reality this would mean hundreds of participants in each arm, perhaps as big as the sample size in Leucht’s meta-analysis.

Observational evidence can only take us so far.

An adequately-powered, well-conducted randomised trial is the only way to confirm (or refute) clozapine supremacy.