Model fOr cloZApine tReaTment, MOZART

Dr Emanuele Osimo | Department of Psychiatry University of Cambridge, School of Medicine Imperial College

A risk prediction model for Treatment Resistant Schizophrenia

Impact Summary

The practice of medicine frequently requires to make decisions regarding treatments based on how likely one thinks the potential consequences are (often called “outcomes”). “Risk prediction models” are applications that can help to predict the probability of future events, such as a specific diagnosis, by considering multiple “predictors” – the generic name given to factors that you can use to calculate the risk of the specific diagnosis, such as age, sex or ethnicity.

Although such models for several outcomes are being developed in the field of psychiatry, few are clinically useful. This is because they either require research-grade information, or they require information not available when a prediction would be most useful, when the condition starts, or else because they have not been developed following best practice, for example because they have not been shown to work in the wider population (they have not been “externally validated”).

MOZART is a risk prediction model that can predict the risk of developing treatment resistant schizophrenia – a form of psychosis that does not respond well to normal medications – and that can be used at the beginning of psychosis. MOZART was developed precisely to help clinicians and patients make treatment decisions. By using information readily available at the beginning of psychosis, it would be cheap to use, as it is based on information that is routinely collected. Further, its development followed risk prediction modelling best-practice guidelines, including “external validation”, a step that can show the model to work in a separate population from the one it was developed in. It is important to stress that MOZART is not yet ready for the clinic: it will require further steps of development and external validation, as well as regulatory approvals.

Underpinning Research

About a quarter of individuals with a first episode of psychosis will develop treatment resistant schizophrenia. Regrettably, we currently lack the ability to predict who will respond well to standard antipsychotic treatments and who will not.

The late prescription of the only licensed medication for treatment resistant schizophrenia, clozapine, reduces its effectiveness, extending patient suffering and causing additional costs to society. For this reason, there is increasing focus on predicting who is treatment resistant to monitor them more closely, and to start them on an effective treatment as soon as possible.

Using routinely collected information at the beginning of psychosis, such as the age, sex, ethnicity, and inflammatory and metabolic blood markers of a patient, MOZART (Osimo et al., 2023) can give an approximation of the true risk of the patient being treatment resistant. This calculator was trained on data from specialist early intervention teams for first episode psychosis in Cambridgeshire and Birmingham, and validated on a separate similar South London-based sample. Showing good performance in internal and external validation, decision-curve analysis – a specific statistical method to study how useful a method is a different risk thresholds – shows that MOZART can be useful in identifying higher risk patients for closer monitoring, but not yet in starting medication. This will require further research work.

Subject to further development and validation steps – such as including more variables to produce more powerful estimates – and then clinical approvals, MOZART will be useful to help clinicians to select high-risk patients for close monitoring and early clozapine adoption. Improving the model will increase its usefulness, for example by facilitating early treatment commencement, and hence, exemplifying the promise of precision health in psychiatry.


Dr Emanuele Osimo extends his sincere gratitude to all individuals who have collaborated collectively on this project. They are Dr Benjamin Perry, Dr Pavan Mallikarjun, Ms Megan Pritchard, Mr Jonathan Lewis, Dr Asia Katunda, Dr Graham Murray,  Prof Jesus Perez, Prof Peter Jones, Dr Rudolf Cardinal, Prof Oliver Howes, Prof Rachel Upthegrove, and Prof Golam Khandaker.