Psychosis and metabolic risk: PsyMetRiC 2.0 reaches the clinic

Compared to the general population, individuals with psychosis have a 15-year reduced life expectancy on average, primarily due to preventable physical health comorbidities such as type 2 diabetes and obesity (Plana-Ripoll et al., 2019). Further, some common treatments (e.g. antipsychotics) can worsen cardiometabolic function (Pillinger et al., 2020). Early detection of individuals at-risk of future cardiometabolic disorders is paramount to move clinical care from reactive management to proactive prevention, thereby reducing long-term morbidity and mortality.

To this end, the Psychosis Metabolic Risk Calculator (PsyMetRiC) was developed. PsyMetRiC is a clinical prediction model that predicts the 6-year risk of metabolic syndrome in young individuals following a first episode of psychosis (Perry et al., 2021). It has been extensively validated in different settings. However, very few clinical prediction models in psychiatry have successfully been implemented for routine use in clinic (Salazar de Pablo et al., 2021) due to numerous barriers including poor model accuracy, insufficient consideration of stakeholder acceptability and utility, and the need for substantial infrastructure for their translation and regulated use in clinic.

Perry et al. (2026) set out to address this gap by refining and externally validating the PsyMetRiC prediction models with a focus on clinical utility and acceptability, and implementing and registering the models in a web-application as regulated, clinically available medical devices.

Individuals with severe mental illness are 1.5-2.5 times more likely to develop cardiovascular disease compared to the general population.

Methods

This retrospective, multicohort study used routinely collected data from two primary care (CPRD; QResearch) and one secondary care (CRIS) electronic databases, to identify individuals aged 16-35 years with either a first-recorded diagnosis of a psychosis-spectrum disorder (for primary care) or enrolment in an early intervention service for psychosis (for secondary care).

The authors refined the previous PsyMetRiC model, adding new predictors (e.g., family history of cardiometabolic disorder), to develop and externally validate three main prediction models:

PsyMetRiC2-MetS: to predict metabolic syndrome within 1-6 years.
PsyMetRiC2-T2D: to predict the time-to-event outcome of type 2 diabetes within 10 years.
PsyMetRiC2-WG: to predict clinically significant weight gain (increase to less healthy BMI category) within 1 year.

Predictive performance of these models was primarily assessed by measures of discrimination, calibration, and clinical usefulness. Importantly, the prediction models were collaboratively developed with stakeholders including clinicians, carers, and a lived experience advisory panel of young people with psychosis.

Results

Sample

Overall, 25,850 individuals were included across the three databases. Using primary care data, the PsyMetRiC2-MetS model was developed on 3,989 individuals in CPRD and externally validated on 4,347 individuals in QResearch, and similarly for the PsyMetRiC2-T2D model with 9,181 individuals and 7,487 individuals respectively. Using secondary care data (CRIS), the PsyMetRiC2-WG model was developed and internally validated on 846 individuals (with no external validation due to an insufficient validation sample).

Prediction model performance

1. PsyMetRiC2-MetS

In external validation, the full model (with biochemical predictors) discriminated well between people with higher and lower risk of metabolic syndrome, with a C-index of 0.81. Calibration metrics (slope = 1.22; intercept = -0.04) indicated acceptable agreement between predicted and observed risk, and decision curve analysis suggested greater net benefit (i.e. clinical utility) of using the model compared to the clinical alternatives of treating all/none at a threshold above 0.05. The full model generally outperformed the partial model (C-index = 0.79; calibration slope = 1.14; calibration intercept = -0.11; comparatively lower net benefit at higher risk thresholds).

2. PsyMetRiC2-T2D

In external validation, the model distinguished well between people with higher and lower risk of type diabetes (C-index of 0.81), with calibration plot showing good agreement between predicted risk and observed proportion. Decision curve analysis indicated greater clinical utility of the model compared to treating all/none at thresholds above 0.03.

3. PsyMetRiC2-WG

As aforementioned, only internal validation was conducted for the weight gain model. Both the full model and the partial model performed similarly in both discrimination and calibration metrics (respectively: C-index = 0.78, C-index = 0.77; calibration slope = 0.88, calibration slope = 0.87), both demonstrating greater clinical utility at thresholds above 0.03 in decision curve analyses.

Web application

The PsyMetRiC2-MetS and PsyMetRiC2-T2D models have been registered as class 1 software as a medical device in compliance with the UK Medical Devices Regulations 2002, thereby enabling them for clinical use. These are available on an accompanying web application (https://psymetric.app).

The inclusion of biochemical predictors improved performance, highlighting their importance in the assessment of individuals with early psychosis.

Conclusions

The authors have developed and validated several cardiometabolic prediction models for young people with psychosis spectrum disorders, which appear to show good performance and clinical usefulness. The registration of the models as regulated medical devices in Great Britain makes them some of the first clinical prediction models to be available for routine clinical use in psychiatry.

The authors say that their models:

can help shift cardiometabolic care in early psychosis from reactive management—which is associated with persistently poor outcomes—to earlier, proactive prevention supported by shared decision making.

The PsyMetRiC models are some of the first clinical prediction models to be available for routine clinical use in psychiatry, taking “care in early psychosis from reactive management to earlier, proactive prevention”.

Strengths and limitations

Key strengths of this study include:

Collaborative development of the prediction model with stakeholders including clinicians, carers, and a lived experience advisory panel of young people with psychosis (as highlighted in Haynes et al., 2026). This not only brings novel insights and empowers this population, but also helps to improve the clinical relevance and utility of PsyMetRiC to patients (e.g., operationalising the outcomes with stakeholder feedback).
Use of routine predictor data to prioritise scalability to other settings.
Assessment of equity in performance across sex and ethnic background through subgroup analyses (not reported above, but minimal meaningful differences observed).
Clear and thorough reporting of methodology.

Some limitations include:

Larger samples are required to be able to assess further subgroups, and also for the PsyMetRiC2-WG model, given the lack of a suitable external validation sample for this model and its subsequent exclusion from the web application.
The analysis relies on recorded data from electronic health records which cannot capture more fine-grained information such as medication adherence, which is an important consideration in the association between psychosis and cardiometabolic outcomes.
The authors chose to reduce model complexity by not incorporating non-linear relationships and/or interactions, which may improve performance, given known difference (e.g. across ethnicities).
Whilst these prediction models can help to identify at-risk individuals (and do highlight predictors of greater importance for each outcome), they do not tackle the question of causality and how to prevent the development of these outcomes. A causal modelling approach would be beneficial here (e.g., Leighton et al. (2026), as recently blogged about by Dominic Oliver).

Stakeholder involvement, routine clinical data and transparent reporting strengthen PsyMetRiC 2.0, but limitations in available data and the inability of prediction models to identify causal mechanisms remain important considerations.

Implications for practice

The PsyMetRiC models have important implications for clinical practice as some of the first prediction models in psychiatry to be registered as class 1 medical devices, ready for clinical use in Great Britain. Using these models would facilitate a more personalised and proactive approach to management of cardiometabolic function, and may improve morbidity and mortality in this population.

Whilst no specific interventions are currently recommended by the web application, future work will soon be underway to assess how different interventions, guided by risk stratification from the models, may improve outcomes. The absence of treatment recommendations highlights the importance of incorporating risk estimates with clinician judgement and each individual’s circumstances and needs (as opposed to a blanket approach).

The superiority of the full models over the partial models indicates the importance of capturing these biochemical factors for a better understanding of cardiometabolic function. Given this, the successful implementation of the PsyMetRiC models will rely on regular and comprehensive cardiovascular screening to also ensure that predictor data are both timely and available. However, a recent study examining long-term screening patterns in primary care in the United Kingdom found that approximately only half of adults with a severe mental illness had been screened for six key cardiovascular risk factors (Launders et al., 2025, as blogged about by Jingyi Wang). This highlights the clear need for more targeted strategies for this high-risk group experiencing substantial cardiovascular health inequalities.

PsyMetRiC 2.0 provides a practical tool for identifying young people with psychosis at increased cardiometabolic risk, but prediction alone is not enough without effective screening and intervention pathways.

Statement of interests

One of Yanakan Logeswaran’s PhD supervisors (Dr Dominic Oliver) is part of the PsyMetRiC Operating Division in partnership with University of Birmingham Enterprise but derives no financial benefit. Yan did not use AI to write this blog post.

Editor

Edited by Éimear Foley. ChatGPT assisted with language refinement and formatting during the editorial phase.

Links

Primary paper

Benjamin Perry, Emanuele Osimo, Shuqing Si, Karla Hitchins, Clara Lewis, Ben Laws, Simon Griffin, Golam Khandaker, Graham Murray, David Shiers, Carolyn Chew-Graham, Peter Jones, Alastair Denniston, Marco Bardus, Sue Jowett, Annabel Walsh, Shizana Arshad, Tomas Formanek, Toby Pillinger, Robert McCutcheon, Richard Holt, Silke Heyse, Magaly Rambousek, Khadija Whiteley, Rachel Upthegrove, Joie Ensor (2026) Cardiometabolic prediction models for young people with psychosis spectrum disorders in the UK (PsyMetRiC 2.0): a retrospective, multicohort clinical prediction model study. The Lancet Psychiatry, 13(4), 291-303.