It’s my first day back in the community mental health clinic working as a psychiatric resident doctor. Brompton in the corner, Kelston Roundhill out the window. Jim’s telling me that the threatening voices have started coming back. After some discussion, I’m telling Jim I think he should restart an antipsychotic. He explains that aripiprazole made him restless and only olanzapine worked well last time, but he gained a lot of weight. What should we do?
The 2020 Maudsley Advanced Prescribing in Psychosis Guidelines tell me prevention of weight gain is better than cure, and that diet, exercise, switching antipsychotics or adding aripiprazole, metformin, topiramate, orlistat or Glucagon-Like Peptide-1 (GLP-1) receptor agonists can help (Morrison et al., 2020). A 2022 Cochrane Review tells me that there is low certainty evidence from four randomised controlled trials (RCTs) that metformin may be effective in preventing weight gain (Agarwal et al., 2022). A Mental Elf blog from last month highlights a 2025 Lancet Commission report, which recommends adding metformin if starting clozapine or olanzapine (Lawson, 2025).
Well, these resources are suggesting metformin, but I can’t remember ever seeing a patient starting metformin at the same time as an antipsychotic. What is going on? Am I a bad doctor? Adding to my confusion, the 2025 Maudsley Prescribing Guidelines in Psychiatry say that medications to treat weight gain should only be considered if lifestyle interventions or antipsychotic switching have not worked or obesity is an immediate physical risk (Taylor et al., 2025). It would be great to see what clinicians are doing around the UK. Thankfully, some excellent psychiatric epidemiologists have already had that same thought (Trajano et al., 2025).
Despite strong guideline support for using metformin to prevent weight gain caused by antipsychotic treatment, its use in practice remains uncommon, raising questions about clinical habits and real-world prescribing.
Methods
Trajano and colleagues used two primary care databases from the Clinical Practice Research Datalink to identify 26,537 patients who started aripiprazole, olanzapine, quetiapine or risperidone between 2005 and 2019, and had previously had schizophrenia, bipolar disorder or other non-organic psychoses noted in their records.
In Part 1, they measured whether these people were ever prescribed metformin and when. Then, for those that weren’t initially prescribed metformin, they measured the cumulative incidence of being prescribed metformin 1 and 2 years down the line. They also measured the proportion of antipsychotic users prescribed metformin each year.
In Part 2, they compared the characteristics of those prescribed metformin within 2 years of starting the antipsychotic compared to those not ever prescribed any metformin.
In Part 3, they compared the changes in weight between patients who started metformin when starting the antipsychotic and those that never took metformin. They also stratified by substance as well as using multivariate linear regression models to estimate the effect of starting metformin. Model 1 was adjusted for baseline weight alone and model 2 included baseline weight plus antipsychotic medication, age at index date, ethnicity, social deprivation, prior diagnosis of diabetes and prior diagnosis of poly-cystic ovarian syndrome (PCOS).
Results
Part 1
Of the 26,537 patients who started an antipsychotic, 2,873 had started metformin before starting the antipsychotic. Among the remaining 23,664, there was a cumulative incidence of 1.9% starting metformin within a year and 3.3% starting it within 2 years. This equated to 696 people starting metformin within 2 years. 1,083 started metformin more than 2 years after starting an antipsychotic, and 21,885 never took metformin.
The proportion of people prescribed metformin increased from 1.31% in 2005 to 5.84% in 2017.
Part 2
Compared to the 21,885 that never took metformin, the 696 that took metformin within 2 years of starting the antipsychotic were more likely to have diabetes, have hypertension, be obese, and take lipid-regulating medications. There were also more subtle differences in recorded ethnicity and local relative deprivation and the proportion with schizophrenia, dyslipidaemia, PCOS (polycystic ovarian syndrome), insulin-prescriptions, as well as mean HbA1c and glucose levels. Distributions of sex, ages, geographical regions and other health conditions and medications appeared similar between the two groups. It seemed like 51% of those starting metformin had documented diabetes or PCOS as probable indications for its use.
Part 3
The 21,885 patients who never started metformin were 76.1kg on average when starting the antipsychotic and 78.7kg two years later. So, their weight increased 4.2% on average. 212 patients started metformin between 1 month before and 3 months after starting the antipsychotic. On average they were 90.4kg at the start and 89.3kg at 2 years, thus losing 0.7% on average. There were no clear differences between different antipsychotics in the stratified analyses.
The authors then estimated weight at 2 years using a linear regression model that included only metformin use and baseline weight as independent variables. Females that took metformin were predicted to be 2.04 kg lighter (95%CI: -0.47 to 4.55) than females that didn’t, and males 3.02 kg lighter (95%CI: 0.20 to 5.84). When using a model that also included antipsychotic medication, age at index date, ethnicity, social deprivation, prior diagnosis of diabetes and prior diagnosis of PCOS, the values were 1.48 kg (95%CI: -1.07 to 4.03) among females and 1.84 kg (95%CI: -0.98 to 4.67) among males.
Among over 26,000 patients starting antipsychotics, metformin was rarely prescribed, typically reserved for those with diabetes or polycystic ovarian syndrome, and its use was associated with modest weight loss compared to those who never received it.
Conclusions
The authors concluded that:
- There is low utilisation of metformin, despite guidelines supporting its use,
- Further studies are required to reliably estimate the effect of metformin outside of trial settings and who might benefit most, and
- Clinicians need clear guidance on implementing current guidance around cardiometabolic care when using second generation antipsychotics.
Metformin remains underused despite guideline support, and clearer implementation strategies and further real-world research are needed to guide its role in cardiometabolic care alongside second generation antipsychotics.
Strengths and limitations
It is exciting to see the researchers harness large population-representative datasets to assess clinically important questions:
- Are we prescribing metformin?
- Who are we prescribing it to?
- Are we seeing the expected effects on weight gain?
Strengths of the study include the large study population and detailed comparison between those starting metformin and not. Being based on routinely collected data in primary care, it should be at low risk of the healthy volunteer bias and Hawthorne effect (i.e., behaviour change in response to being observed) and the sharing of analytical code is invaluable for understanding and building upon this work.
The data source and broad inclusion criteria mean that study population should also be representative of the broader population starting antipsychotics. However, with the average age of patients entering this study being over 50 years old, the results may not be generalisable to younger adults of the ages at which severe mental illness (SMI) is most likely to be first diagnosed. The exclusion of over 120,000 patients due being registered at the GP for less than 6 months or not having a record of SMI before the antipsychotic, potentially contributed to underrepresentation of young adults. While those exclusion criteria were necessary for study validity and relevance, the exclusion of 36,800 patients (because there were no lipids or HbA1c recorded in prior 2 years) was potentially unnecessary and a contributor to selection bias.
Another limitation is that at each time point after antipsychotic initiation only 33 to 41% of participants had weight recorded (Richards-Belle et al., 2025). The authors filled in the missing data based on the assumption that the probability of a variable being missing can be calculated from observed data. However, this may not be valid as people who gain weight are perhaps more likely to get their weight measured. Sensitivity analyses could be used to estimate the plausible impact of this (Madley-Dowd et al., 2025), but it would be extra challenging because the relationship between weight gain and whether weight is measured is probably affected by whether someone is taking metformin.
Ideally, for a fuller picture of healthcare practices this study would also include information on secondary care prescribing or other weight management interventions. Additional limitations relating to part 3 of the study are the inherent difficulty in causal analyses of weight changes (Tennant et al., 2023) and the inevitable presence of unmeasured confounding. Finally, it is a shame that there was a lack of power to compare outcomes between antipsychotics: we would expect the usefulness of metformin to be greater for those starting olanzapine than aripiprazole.
This large study uses representative data to explore metformin prescribing and weight outcomes, but exclusions, missing records, and confounding limit its relevance to younger patients and broader prescribing decisions.
Implications for practice
Part 1 of this study should be a wake-up call that we are probably not doing enough to prevent or manage antipsychotic induced weight gain.
Part 2 of this study indicates that even among the few people co-prescribed metformin, the reason was mostly diabetes or PCOS, again reinforcing the message that we don’t seem to be focusing on the prevention of weight gain enough.
Part 3 of the study serves as a contribution to triangulating the evidence around the effect of metformin in preventing antipsychotic-induced weight gain. We had evidence from RCTs that metformin helps prevent weight gain, but those studies may not have been very generalisable and may have been affected by volunteer bias or the Hawthorne effect. That this study of primary care records draws similar conclusions, despite having different sources of bias, strengthens our confidence in the conclusions (Lawlor et al., 2016).
What other forms of observational evidence may reinforce the conclusion that metformin helps prevent weight gain? We could perform cross-context comparisons using other countries with different healthcare systems and confounding structures. Alternatively, we could compare the results to those for a negative control exposure that we don’t expect to cause large weight loss, such as statins, or a negative control outcome that we don’t expect to be affected by metformin, for example blood pressure in non-diabetic patients. Ultimately, there remains enough uncertainty about the pros, cons and patient experiences around starting metformin with antipsychotics that inclusive RCTs with substantial follow-up times appear warranted.
So, should Jim start metformin? If he’s going to start olanzapine, probably yes. But first it will be important to discuss the effectiveness of lifestyle interventions, the potential side effects of metformin, including lactic acidosis and vitamin B12 deficiency, and be clear about the ongoing monitoring of this medication use, whether in primary or secondary care. Will I recommend metformin for everyone starting second-generation antipsychotics based on the results of this study? No, we’ll make shared decisions based on individual risk factors and preferences, with the help of the latest guidelines (e.g. Carolan et al., 2025). Will the guidelines look the same in 5 years’ time? GLP-1 agonists will surely feature more prominently, won’t they?
Clinicians should consider metformin for patients starting olanzapine or clozapine, but remain mindful of individual risks, the value of lifestyle interventions, and the evolving role of newer treatments like GLP-1 agonists.
Statement of interests
I’ve never met the authors, but also use CPRD for psychiatric epidemiology research, and have taken much inspiration from their previous work.
Links
Primary paper
Farache Trajano L, Hayes JF, Launders N, Davies NM, Osborn DPJ, Richards-Belle A. (2025) Co-prescription of metformin and antipsychotics in severe mental illness: a UK primary care cohort study. BMJ Mental Health. 2025;28:e301505. https://doi.org/10.1136/bmjment-2024-301505
Other references
Agarwal, S. M., Stogios, N., Ahsan, Z. A., Lockwood, J. T., Duncan, M. J., Takeuchi, H., Cohn, T., Taylor, V. H., Remington, G., Faulkner, G. E. J., & Hahn, M. (2022). Pharmacological interventions for prevention of weight gain in people with schizophrenia. The Cochrane Database of Systematic Reviews, 10(10), CD013337. https://doi.org/10.1002/14651858.CD013337.pub2
Carolan, A., Hynes-Ryan, C., Agarwal, S. M., Bourke, R., Cullen, W., Gaughran, F., Hahn, M. K., Krivoy, A., Lally, J., Leucht, S., Lyne, J., McCutcheon, R. A., Norton, M. J., O’Connor, K., Perry, B. I., Pillinger, T., Shiers, D., Siskind, D., Thompson, A., … O’Donoghue, B. (2025). Metformin for the Prevention of Antipsychotic-Induced Weight Gain: Guideline Development and Consensus Validation. Schizophrenia Bulletin, 51(5), 1193–1205. https://doi.org/10.1093/schbul/sbae205
Lawlor, D. A., Tilling, K., & Davey Smith, G. (2016). Triangulation in aetiological epidemiology. International Journal of Epidemiology, 45(6), 1866–1886. https://doi.org/10.1093/ije/dyw314
Lawson, K. (2025, August 13). Physical health side effects of psychotropic medication. National Elf Service. https://www.nationalelfservice.net/treatment/antipsychotics/physical-health-side-effects-of-psychotropic-medication-holistic-prevention-and-management/
Madley-Dowd, P., Hughes, R. A., Mathur, M. B., Heron, J., & Tilling, K. (2025). Using directed acyclic graphs to determine whether multiple imputation or subsample multiple imputation estimates of an exposure-outcome association are unbiased (No. arXiv:2503.24035). arXiv. https://doi.org/10.48550/arXiv.2503.24035
Morrison, P., Taylor, D. M., & McGuire, P. (2020). The Maudsley guidelines on advanced prescribing in psychosis (pp. xii, 93). Wiley Blackwell.
Richards-Belle, A., Launders, N., Hardoon, S., Richards, A., Man, K. K. C., Davies, N. M., Bramon, E., Hayes, J. F., & Osborn, D. P. J. (2025). Comparative cardiometabolic safety and effectiveness of aripiprazole in people with severe mental illness: A target trial emulation. PLOS Medicine, 22(1), e1004520. https://doi.org/10.1371/journal.pmed.1004520
Taylor, D. M., Barnes, T. R. E., & Young, A. H. (2025). The Maudsley Prescribing Guidelines in Psychiatry. John Wiley & Sons.
Tennant, P. W. G., Tomova, G. D., Murray, E. J., Arnold, K. F., Fox, M. P., & Gilthorpe, M. S. (2023). Lord’s ‘paradox’ explained: The 50-year warning on the use of ‘change scores’ in observational data (No. arXiv:2302.01822). arXiv. https://doi.org/10.48550/arXiv.2302.01822
Trajano, L. F., Hayes, J. F., Launders, N., Davies, N. M., Osborn, D. P. J., & Richards-Belle, A. (2025). Co-prescription of metformin and antipsychotics in severe mental illness: A UK primary care cohort study. BMJ Mental Health, 28(1). https://doi.org/10.1136/bmjment-2024-301505