On average, people with psychosis die 15 years earlier than the general population. This is largely due to preventable physical illnesses, with obesity playing a central role (Plana-Ripoll O et al, 2019). Obesity is around three times more common in people with psychosis compared to the general population (Afzal M et al, 2021). Previous research has shown that while antipsychotic medications are essential to aid recovery, they are a key contributor to this issue (Johnston G, 2025; Pillinger T et al, 2020).
Prediction models like PsyMetRiC (Perry BI et al, 2026) are helpful to identify risk of future physical health problems, which can inform young people with psychosis along with their clinical teams. However, these models are not able to unravel what actually causes these issues and how to prevent them.
This new study by Leighton et al. (2026) attempts to bridge this gap. The authors explored whether early weight gain in the first 12 weeks of antipsychotic treatment causes obesity after 1 year, and whether we can use causal prediction models to understand what would happen for an individual if we intervened.
Understanding early weight gain in the first weeks of antipsychotic treatment could help prevent long-term physical health inequalities in psychosis.
Methods
This study used data from a randomised controlled trial comparing olanzapine and haloperidol in people with first-episode psychosis.
The authors conducted two main analyses:
- Causal mediation analysis (n=97) to test whether weight gain at 1 year is mediated by early (0-12 weeks) versus later (12-52 weeks) weight change.
- Prediction modelling, developing two sequential logistic regression models:
- A baseline model predicting ≥7% weight gain at 12 weeks (n=172)
- A 12-week model predicting obesity at 1 year (n=97)
They also used counterfactual modelling to estimate how changing treatment or early weight might alter outcomes.
Results
Sample and outcomes
The baseline model included 172 participants. In this group, 57% experienced clinically significant weight gain (≥7%) at 12 weeks. The 12-week model included 97 participants, of whom 36% developed obesity at 1 year.
Causal mediation
The total effect of olanzapine compared to haloperidol on 1-year weight was 6.62 kg (95% CI 1.44 to 12.3). Most of this effect was explained by weight gain in the first 12 weeks (5.70 kg; 95% CI 2.83 to 8.66). There was no strong evidence that later weight gain between 12-52 weeks played a meaningful mediating role.
Overall, long-term differences in weight appear to be driven largely by early changes rather than later weight gain.
Prediction model performance
The models distinguished well between people with higher and lower risk of clinically significant weight gain/obesity. However, their predictions are overly optimistic, showcasing patterns of under-estimating patients’ risk. Specifically:
- Baseline model (12-week clinically significant weight gain):
- C-index: 0.728 (good discrimination)
- Calibration slope: 0.768 (some overfitting)
- 12-week model (1-year obesity):
- C-index: 0.904 (outstanding discrimination)
- Calibration slope: 0.601 (notably overfitted)
Counterfactual predictions
Switching from olanzapine to haloperidol reduced risk of early weight gain. Reducing weight gain within the first 12 weeks lowered predicted risk of developing obesity within 1 year.
This means that if a patient gains a lot of weight early into treatment, intervening quickly could change their long-term trajectory.
Using mediation analysis and prediction modelling, the study showed that early weight gain in the first 12 weeks of treatment explained most of the difference in longer-term weight outcomes.
Conclusions
The authors concluded that:
Early weight gain in the first 12 weeks of antipsychotic treatment may causally mediate weight outcomes at 1 year.
They also demonstrate the feasibility of causal prediction models, which can estimate how treatment changes or changes in early weight trajectories might alter outcomes for individuals.
However, these models are not yet ready for clinical usedue to small sample sizes in model development and a lack of external validation. The findings are best viewed as hypothesis-generating, pointing toward the potential of causal prediction modelling rather than tools for current clinical decision-making.
Early weight trajectories may influence 1-year outcomes, but although causal prediction models are feasible, they are not yet ready for practice.
Strengths and limitations
This is an ambitious and methodologically innovative study using randomised controlled trial data. This reduces the impact of confounding by indication, where clinicians in routine practice choose treatments based on patient characteristics. As the same characteristics could also influence outcomes, confounding by indication makes it difficult to separate true treatment effects from differences between patients. The use of randomised trial data in this study helps balance these characteristics across treatment groups and therefore strengthens claims of causality, meaning differences in outcomes are more likely to reflect the treatment itself rather than pre-existing differences between patients.
However, several issues limit the impact of these findings:
- The sample is small, especially for the 12-week and mediation analyses (n=97)
- There is substantial, unbalanced dropout across the two arms.
- Miscalibration is seen in both models.
Together, these limitations raise concerns about attrition bias and overfitting, making it less likely that this model will generalise to other patients.
More generally, while causal prediction models offer a step forward from traditional risk prediction, they themselves come with important limitations. They rely on strong assumptions about the underlying causal structure: if these are incorrect, the resulting predictions and suggested interventions may be misleading. As such, these approaches require large, high-quality datasets to accurately model this causal structure. However, this then means that these datasets may not reflect real-world clinical samples. Randomised controlled trial data only represent around 20% of people with psychosis, as many have physical health comorbidities that lead to trial exclusion (Taipale H et al., 2022).
Causal prediction models also remain vulnerable to unmeasured confounding (e.g. lifestyle or biological factors), which can bias estimates even in well-designed analyses (though the authors of the current paper mitigated this with sensitivity analyses).
A key challenge is that counterfactual predictions, which are central to these models, cannot be directly validated at the individual level. This limits confidence in individual-level “what if” clinical scenarios.
This study benefits from randomised data for causal inference, but practical and methodological limitations in modelling, generalisability, and unverifiable counterfactuals mean findings remain exploratory.
Implications for practice
So, what does this mean for real-world care?
1. Timing is everything
Starting antipsychotic treatment can quickly lead to substantial weight gain. This echoes previous studies showing that people with first episode psychosis gained 3.5kg on average within the first 35 days of treatment (Vochoskova K et al. 2023).
* The first 12 weeks of antipsychotic treatment are crucial *
If early weight gain drives later obesity, then waiting months or years to intervene is too late. Services should prioritise:
- Early weight monitoring (weekly/fortnightly)
- Early access to dietary or lifestyle support
- Proactive discussions about weight risk before starting medication
2. Rethinking treatment decisions
The findings also raise questions about antipsychotic choice. While olanzapine may be more effective or better tolerated in some cases, its metabolic risks are substantial.
This study suggests:
- Treatment decisions should explicitly weigh psychiatric benefit vs physical health risks
- Shared decision-making is key: patients should understand these trade-offs and have input in which side effects they view as most important to avoid.
3. Toward precision psychiatry
Perhaps the most exciting implication is the potential for causal, individualised prediction models. Through use of causal prediction models, we can predict not just risk, but what would happen if we intervened. This framework does not need to act in opposition with associative prediction models (e.g. PsyMetRiC), but could inform clinical decision making to reduce the impact of identified risks and optimising care for people with psychosis.
Overall, while the study uses novel and appropriate methods, the findings should be interpreted cautiously, instead demonstrating the feasibility and the potential of causal prediction models in psychiatry.
Early antipsychotic treatment can rapidly increase weight, highlighting the need for early monitoring, informed prescribing decisions, and the potential role of prediction models in guiding personalised risk and treatment in psychosis care.
Statement of interests
Dominic Oliver is part of the PsyMetRiC Operating Division in partnership with University of Birmingham Enterprise but derives no financial benefit. Generative AI was used for editing purposes.
Editor
Edited by Éimear Foley. AI tools assisted with language refinement and formatting during the editorial phase.
Links
Primary paper
Samuel Leighton, I Lam Leong, Damian Machlanski, Benjamin Perry, Sotirios Tsaftaris, Fani Deligianni, Stephen Lawrie, Rajeev Krishnadas (2026) Antipsychotic-induced weight gain in psychosis: causal mediation analysis and feasibility study of causal actionable prediction model development using counterfactuals to target obesity. British Journal of Psychiatry. Published online 2026:1-10. doi:10.1192/bjp.2026.10561
Other references
Afzal M, Siddiqi N, Ahmad B et al. (2021) Prevalence of Overweight and Obesity in People With Severe Mental Illness: Systematic Review and Meta-Analysis. Front. Endocrinol. 12:769309. doi: 10.3389/fendo.2021.769309
Johnston G (2025) Rapid weight gain after SMI diagnosis, but why so few referrals for support? The Mental Elf, 2 Dec 2025.
Perry BI, Osimo EF, Si S. et al. (2026) Cardiometabolic prediction models for young people with psychosis spectrum disorders in the UK (PsyMetRiC 2.0): a retrospective, multicohort clinical prediction model study. Lancet Psychiatry 2026 13 (4) 291-303 https://doi.org/10.1016/S2215-0366(25)00398-0
Pillinger T, McCutcheon RA, Vano L. et al. (2020) Comparative effects of 18 antipsychotics on metabolic function in patients with schizophrenia, predictors of metabolic dysregulation, and association with psychopathology: a systematic review and network meta-analysis. Lancet Psychiatry 2020 7 (1) 64-77. https://doi.org/10.1016/S2215-0366(19)30416-X
Plana-Ripoll O, Pedersen CB, Agerbo E, et al. A comprehensive analysis of mortality-related health metrics associated with mental disorders: a nationwide, register-based cohort study. Lancet 2019; 394: 1827–35; DOI: 10.1016/S0140-6736(19)32316-5
Taipale H, Schneider-Thoma J, Pinzón-Espinosa J. et al. (2022) Representation and outcomes of individuals with schizophrenia seen in everyday practice who are ineligible for randomized clinical trials. JAMA Psychiatry 2022 79 (3) 210-218. https://doi.org/10.1001/jamapsychiatry.2021.3990
Vochoskova K, McWhinney SR, Fialova M, et al. Weight and metabolic changes in early psychosis-association with daily quantification of medication exposure during the first hospitalization. Acta Psychiatr Scand. 2023 148 265–276. https://doi.org/10.1111/acps.13594
