Published on 15 September 2026
Diabetes and mental disorders: who gets monitored, who gets treated?
In brief
Forty-nine studies, more than 5.5 million people with diabetes, one question: at equal diabetes, do people with a psychiatric disorder receive the same monitoring and the same treatments as everyone else? For monitoring, the answer is no. The primary outcome, receiving at least one recommended monitoring procedure, gives an odds ratio of 0.81. Retinal screening sits at 0.77, HbA1c measurement at 0.81, renal investigation at 0.78, lipid profile at 0.83, foot examination at 0.85. For treatment the picture differs: no overall gap, more insulin, at 1.52, and far fewer GLP-1 receptor agonists, at 0.26, but on two studies only. Against what one would expect, the gaps do not widen with the severity of the disorder: they are clearest for the category of any mental disorder and for dementia, and they narrow in the schizophrenia and bipolar disorder subgroups. The data are observational and do not identify the cause of the gap.
The context
People with psychiatric disorders carry a higher cardiometabolic mortality than the general population, and diabetes is up to three times as frequent among them. Diabetes holds a particular place in that chain, because the psychiatrist intervenes twice: they prescribe drugs associated with a raised risk of metabolic syndrome and of type 2 diabetes, and they are often the doctor the patient sees most regularly.
The question asked here is therefore not whether psychiatric patients have more diabetes, which has been established for a long time. It is whether, once diabetes is present and diagnosed, they receive the same care. The authors note that no meta-analysis had so far quantified that gap against an established standard, in this case the nine NICE diabetes monitoring indicators.
The study at a glance
| Item | Content |
|---|---|
| Population | Patients with diabetes, type 1 or type 2, who also have a psychiatric disorder |
| Detail5,503,712 people with diabetes, of whom 838,366, or 15.2%, had a diagnosed mental disorder, defined by ICD-9 or ICD-10 criteria in 40 studies. Mean age 61.4 years, SD 8.7, range 47 to 82 years. Diagnostic subgroups actually analysed: mood disorders, major depressive disorder, schizophrenia, bipolar disorder, substance use disorders, dementia, anxiety disorders, personality disorders. No ADHD and no post-traumatic stress disorder subgroup is reported. | |
| Exposure studied | Diabetes care actually received, in monitoring and in treatment |
| DetailMonitoring: the nine NICE indicators, brought down to eight after serum creatinine and the urine albumin-creatinine ratio were merged into a single renal indicator. Treatment: any antidiabetic drug, insulin, non-insulin antidiabetic drug, GLP-1 receptor agonist, lipid-lowering drugs, antihypertensives, dietary advice, referral to structured diabetes education, influenza vaccination, smoking cessation support. Metformin, SGLT2 inhibitors and statins are not the subject of any reported numerical estimate. | |
| Comparator | Patients with diabetes and no psychiatric disorder |
| DetailThe comparison is made once diabetes is diagnosed, which sets aside the question of incidence and isolates the question of care. | |
| Outcome | Binary composite primary outcome, receiving at least one recommended monitoring procedure or treatment |
| DetailSecondary outcomes: each indicator taken on its own and each drug class. Random-effects model with restricted maximum likelihood estimation, Hartung, Knapp, Sidik and Jonkman method when between three and ten studies contributed with non-zero between-study variance, heterogeneity measured by I², subgroup analyses by diagnosis and by country, meta-regressions. | |
| Design | Systematic review and meta-analysis of observational studies, compliant with MOOSE and PRISMA 2020, protocol registered on OSF |
| Detail49 studies, 42 cohort and 7 case-control, searches in Scopus, Embase, MEDLINE and PsycINFO from database inception to 8 February 2025, with no language restriction. Data collected between 1990 and 2020. Quality assessed with the Newcastle-Ottawa Scale. Work signed collectively by the ECNP PAN-Health Group. | |
| Funding and declared interests | No funding. Numerous declared interests |
| DetailThe authors declare no funding source for this work. Several co-authors declare honoraria, advisory activity or holdings with the pharmaceutical industry, outside the submitted work. The data are available from the corresponding author on reasonable request. | |
Quality control
| Point checked | Judgement |
|---|---|
| Registered protocol | Registered on OSF |
| FindingPrior registration limits after-the-fact selection of the outcomes reported. Departures from the protocol are documented and justified in the supplementary material, in particular the alignment of the indicators on the NICE standard and the addition of moderators after reading the corpus. | |
| Standardised reporting | PRISMA 2020 and MOOSE |
| FindingFlow diagram, complete compliance checklists and a list of excluded studies with reasons, which makes the work reproducible. | |
| Quality of the included studies | 41 of 49 studies high quality |
| FindingNewcastle-Ottawa Scale, high-quality threshold set at 7 out of 9. Median score 8, interquartile range 7 to 9. The study-by-study rating appears in the supplementary material, with scores running from 4 to 9. | |
| Power | Very high |
| Finding5,503,712 people with diabetes. A corpus of that size makes modest gaps detectable, but the confidence intervals stay wide on several indicators, because the random-effects model absorbs major heterogeneity. | |
| Heterogeneity | Extreme |
| FindingI² between 92 and 100% across the main monitoring analyses. Health systems, payment arrangements and the definitions of psychiatric disorders vary from one country to the next. The average aggregates different realities, and the authors themselves urge caution on this point. | |
| Nature of the data | Observational |
| FindingNo causal inference is possible. The odds ratio measures a gap, it does not name its origin. Only 19 of the 49 studies provide adjusted estimates. | |
| Publication bias | Sought, not detected |
| FindingFunnel plots and Egger’s test performed where at least ten studies were available for an outcome. No sign of asymmetry on monitoring, on treatment, or on somatic health service use. | |
| Transfer to one country | Country-level evidence is thin |
| FindingCountry by country, the evidence is thin. France, to take one example, contributes a single study, on dementia, with a monitoring odds ratio of 0.61, interval 0.59 to 0.62. It points the same way as the whole, but one study represents neither a national psychiatric population nor a current care system. | |
The findings
| Care procedure | Odds ratio |
|---|---|
| Monitoring, primary composite outcome | 0.81 (0.70 to 0.94) · 29 studies · I² 99% · p=0.0049 |
| ReadingThe central result of the study. It stays significant after low-quality studies, case-control studies, unadjusted estimates and veteran populations are excluded. It becomes non-significant once inpatient or mixed populations are removed, 11 studies, p=0.12. | |
| Retinal screening | 0.77 (0.63 to 0.95) · 21 studies · I² 99% · p=0.013 |
| ReadingThe widest gap among the monitoring procedures. It is a scheduled procedure, which presupposes a referral and an appointment kept. | |
| Renal investigation | 0.78 (0.63 to 0.96) · 16 studies · I² 100% · p=0.022 |
| ReadingSerum creatinine and the urine albumin-creatinine ratio pooled into a single indicator, for want of any distinction in the source studies. | |
| HbA1c measurement | 0.81 (0.68 to 0.97) · 24 studies · I² 100% · p=0.024 |
| ReadingThe most elementary monitoring in diabetes, the one on which every later decision depends, is done less often. | |
| Lipid profile | 0.83 (0.69 to 0.99) · 20 studies · I² 99% · p=0.043 |
| ReadingA gap of small size, whose upper interval bound comes close to unity. | |
| Foot examination | 0.85 (0.76 to 0.95) · 11 studies · I² 92% · p=0.0044 |
| ReadingWith retinal screening, this is one of the two procedures purely specific to diabetes, with no overlap with the monitoring of the metabolic effects of psychotropic drugs. | |
| Blood pressure, body mass index, smoking status | 1.08 (0.77 to 1.52), 1.01 (0.86 to 1.19) and 1.09 (1.02 to 1.17) |
| ReadingNo gap for the first two. Smoking status is recorded slightly more often in psychiatric patients, on two studies only. | |
| Diabetes treatment, composite outcome | 1.02 (0.84 to 1.23) · 22 studies · I² 97% · p=0.87 |
| ReadingA decisive point. On treatment taken as a whole, no gap is shown. The gap this study documents concerns monitoring, not whether the patient is treated. | |
| Insulin | 1.52 (1.16 to 1.99) · 10 studies · I² 99% · p=0.0022 |
| ReadingMore insulin, not less. The authors read this as a possible marker of more severe presentations at diagnosis, so of diabetes left untreated for longer, or of a higher proportion of type 1 diabetes. Table 2 of the article prints 1.50 for that same analysis, while the text, the abstract and the supplementary material give 1.52. | |
| Non-insulin antidiabetic drugs | 1.00 (0.83 to 1.20) · 11 studies · I² 99% · p=0.99 |
| ReadingStrictly no gap. | |
| GLP-1 receptor agonists | 0.26 (0.13 to 0.49) · 2 studies · I² 0% · p<0.0001 |
| ReadingThis is the widest gap in the study, and it falls on the class most useful in a population exposed to drug-induced weight gain. Two studies only, so a fragile estimate despite zero heterogeneity. The authors also report that the quantitative data were too sparse to analyse this class by diagnosis. | |
| Lipid-lowering drugs | 0.93 (0.73 to 1.19) · 7 studies · I² 96% · p=0.50 |
| ReadingNo significant gap across all disorders taken together. A gap appears in the schizophrenia subgroup alone, 0.90 (0.84 to 0.96), 4 studies, p=0.003. No analysis addresses statins specifically. | |
| Antihypertensives | 0.72 (0.52 to 0.98) · 5 studies · I² 81% · p=0.044 |
| ReadingCardiovascular prevention through blood pressure control is started less often, on a small number of studies. | |
| Referral to structured diabetes education | 0.39 (0.27 to 0.58) · 2 studies · I² 0% · p<0.0001 |
| ReadingTwo studies, but a considerable gap, and it falls on the procedure that governs how far the patient can manage the disease themselves. | |
| Somatic health service use | 1.59 (1.30 to 1.94) · 17 studies · I² 99% · p<0.0001 |
| ReadingA counter-intuitive and central result. Psychiatric patients attend somatic care more often, without this translating into better diabetes monitoring. The deficit is therefore not explained simply by an absence of contact with the health system. | |
The gradient by diagnosis does not follow the severity of the psychiatric disorder. It even runs against clinical intuition: the clearest gap is seen for the broad category of any mental disorder and for dementia, while it narrows and stops being significant in the schizophrenia and bipolar disorder subgroups. The authors note this explicitly and put forward an untested hypothesis, that prescribing psychotropic drugs brings with it additional metabolic monitoring which partly offsets the disadvantage.
| Diagnostic subgroup | Monitoring, composite outcome |
|---|---|
| Dementia | 0.60 (0.30 to 1.21) · 4 studies · p=0.10 |
| ReadingThe lowest point estimate in the corpus, non-significant on the composite but significant procedure by procedure: retinal screening 0.60 (0.52 to 0.69), HbA1c 0.63 (0.43 to 0.91), renal investigation 0.69 (0.67 to 0.71). | |
| Substance use disorders | 0.75 (0.50 to 1.12) · 8 studies · p=0.13 |
| ReadingNon-significant. A positive association even appears on renal investigation, 1.13 (1.07 to 1.20), 3 studies. | |
| Schizophrenia | 0.85 (0.70 to 1.04) · 10 studies · p=0.12 |
| ReadingSchizophrenia is not significantly associated with any of the monitoring indicators taken on its own. The significant gaps that remain concern treatment: non-insulin antidiabetic drugs 0.60 (0.46 to 0.78), lipid-lowering drugs 0.90 (0.84 to 0.96), referral to structured diabetes education 0.42 (0.25 to 0.69), with conversely more antidiabetic drugs of all types, 1.27 (1.19 to 1.35). | |
| Major depressive disorder | 0.86 (0.74 to 1.00) · 7 studies · p=0.049 |
| ReadingThe only subgroup that reaches the significance threshold on the monitoring composite, and it does so narrowly. On the individual procedures, only foot examination is significant, 0.84 (0.72 to 0.98). | |
| Bipolar disorder | 0.87 (0.67 to 1.12) · 6 studies · p=0.22 |
| ReadingNo significant association with any indicator of diabetes care quality, in monitoring or in treatment. | |
| Severe mental illness, grouped category | 0.94 (0.82 to 1.07) · 20 studies · p=0.33 |
| ReadingSchizophrenia, bipolar disorder and major depressive disorder grouped together. Non-significant on the composite. Two procedures remain significantly less often performed, retinal screening 0.76 (0.62 to 0.94) and foot examination 0.82 (0.70 to 0.95). | |
| Anxiety disorders and personality disorders | A single study per category, no composite |
| ReadingAnxiety disorders, one study: HbA1c 0.81 (0.79 to 0.83), lipid profile 0.92 (0.89 to 0.95), but retinal screening 1.09 (1.04 to 1.13), a gap in the patients’ favour. Personality disorders, one study: retinal screening 0.89 (0.84 to 0.93), HbA1c 0.73 (0.68 to 0.79), lipid profile 0.75 (0.69 to 0.81). These results rest on a single source and must be held as preliminary. | |
One reading is possible and another must be set aside. The one that is possible: access to the newer drug classes may be more unequal than access to the older ones. The authors themselves put it no more strongly than as disparities in access to newer treatments, and the argument rests on two studies for GLP-1 receptor agonists, against eleven for non-insulin antidiabetic drugs, where the gap is nil. The one that must be set aside: nothing in these data designates anyone as responsible. Monitoring procedures done less often may reflect a prescriber’s reluctance, difficulty in reaching a specialist, adherence expected to be poorer, care forgone by the patient, or the absence of a general practitioner. The odds ratio does not tell these mechanisms apart.
Critical appraisal
| Domain | Judgement |
|---|---|
| Design and level of evidence | Best available on the question |
| FindingSystematic review with meta-analysis, registered protocol, two standardised reporting frameworks, formal assessment of the quality of the included studies. The base stays observational: the level of evidence is capped by that of its sources, whatever the scale of the aggregation. The authors did not carry out a formal certainty assessment of the GRADE type, and they say so. | |
| Relevance of the question | Directly clinical |
| FindingThe question arises at the interface between psychiatry and somatic medicine, exactly where the excess mortality is built. | |
| Internal consistency | Steady across monitoring |
| FindingThe five significant monitoring indicators all point the same way, across procedures of very different kinds, from a blood test to retinal screening. There is, on the other hand, no regular gradient by severity of disorder: the direction of variation between subgroups runs against intuition and the authors do not explain it. | |
| Heterogeneity | Unexplained |
| FindingNo meta-regression explains the variability of the monitoring outcome: not the proportion of men, not mean age, not the type of diabetes, not ethnicity, not the observation period. A single moderator emerges, and on treatment only, the proportion of men, coefficient minus 0.011, p=0.021. The heterogeneity persists in the sensitivity analyses and in the subgroup analyses. | |
| Robustness | One exclusion overturns the result |
| FindingThe primary result withstands five sensitivity analyses out of six. It collapses when the analysis is restricted to strictly outpatient populations, 11 studies, p=0.12. Significance therefore depends in part on the care setting studied, which weighs on any transfer to a community caseload. | |
| Causality | Not established |
| FindingResidual confounding is likely: deprivation, smoking, density of doctors, competing comorbidities. Only 19 of the 49 studies provide adjusted estimates, and the degree of adjustment varies from one to another. | |
| Representativeness | Corpus dominated by the USA |
| FindingAfrica, South America and Asia are absent or under-represented. The authors write that their results mainly reflect care processes in the United States. In the analysis by country, the monitoring gap is in fact not significant in the United States, 15 studies, p=0.15, nor in the United Kingdom, 5 studies, p=0.61. | |
| Publication bias | Sought, not detected |
| FindingFunnel plots and Egger’s test on the outcomes carrying at least ten studies. No asymmetry shown. The authors do report, on the other hand, that no adjustment was made for the multiplicity of tests. | |
Level of evidence
Confidence is high on the existence of a monitoring gap, lower on its exact size. The power is considerable, the gap is found again on five procedures of very different kinds, no publication bias is detected, and the result withstands most of the sensitivity analyses. It is tempered by an extreme heterogeneity that no meta-regression explains, by the loss of significance in a strictly outpatient population, and by the fact that several interval bounds come close to unity. Confidence is low on the origin of that gap, and nil on whether any given intervention would change it: no intervention study is included here, and the authors explicitly call for randomised trials.
What is established most solidly: people followed in psychiatry who have diabetes receive the recommended monitoring procedures less often, even though they attend somatic care more often. What is not established: a gap in treatment taken as a whole, which is nil in this meta-analysis. What is suggested, on two studies: much lower access to GLP-1 receptor agonists. What the data contradict: the idea of a gradient that rises with the severity of the psychiatric disorder. What belongs to interpretation: the mechanisms the authors invoke, fragmentation of health systems, stigma, diagnostic overshadowing, patient-related barriers, and the scarcity of integrated guidelines.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ It is no surprise, but seeing the figure written down does something. The most troubling part is not that they are treated less, it is that they attend more and are monitored less. Retinal screening, creatinine, those get lost along the way. That one I cannot put down to the patient. ”
What this means in practice: the consequence is not that the psychiatrist takes over the diabetes, that is not their job. It is that they stop taking the monitoring for granted. Checking that a recent test exists, knowing who follows the patient metabolically, and writing to the general practitioner when nobody does, are three consultation-room acts.
What you can do with this
- Record the date of the last HbA1c in the notes of every patient on an antipsychotic, in the same way as the date of the last electrocardiogram.
- Name the person who provides the metabolic follow-up. If the answer is uncertain, take it that nobody provides it. That the patient attends often does not guarantee that the monitoring is done.
- Write to the general practitioner when starting or changing an antipsychotic, stating the metabolic risk expected from the molecule chosen.
- Check that retinal screening, foot examination and renal investigation do appear in the patient’s care pathway, as these are the procedures most often lost when follow-up is fragmented.
- Bear in mind that referral to structured diabetes education is, in the few studies that measure it, the procedure most unequally offered. Asking the colleague who manages the diabetes for it is coordination, not prescription.
- The course of action is set out in the NICE decision tree for schizophrenia in adults.
Frequently asked questions
Is it simply that patients do not turn up for appointments?
That is one of the possible explanations, and these data neither confirm it nor rule it out. It sits awkwardly with one of the study’s results: somatic health service use is higher in psychiatric patients, odds ratio 1.59, without diabetes monitoring benefiting from it.
Are psychiatric patients treated less for their diabetes?
Not according to this study. On the composite treatment outcome no gap appears, 22 studies, odds ratio 1.02, p=0.87. They even receive insulin more often, 1.52. The documented deficit concerns the monitoring procedures and, on two studies only, GLP-1 receptor agonists.
Should the psychiatrist prescribe the diabetes treatments?
No. The study does not address the division of roles and recommends nothing on the point. What falls to the psychiatrist is monitoring the metabolic effects of the treatments they prescribe, and coordinating with the doctor who manages the diabetes.
Do these figures transfer to a given health system?
Not as they stand. The corpus is international and dominated by the United States, and the heterogeneity between health systems is extreme. France, to take one example, contributes a single study, on older people with dementia, with a monitoring odds ratio of 0.61. The gap points the same way there, but its size in any national psychiatric population remains to be established.
Is the disadvantage greater in schizophrenia?
No, and this is one of the most unexpected results. Schizophrenia is not significantly associated with any monitoring indicator, and neither is bipolar disorder. The widest gaps are seen for dementia and for the broad category of any mental disorder. The authors put forward an untested hypothesis, that prescribing psychotropic drugs comes with a metabolic monitoring that partly offsets the disadvantage.
Annotated bibliography
Source study. Wagner E, Højlund M, Fiedorowicz JG, Nielsen RE, Østergaard SD, Høye A, Heiberg IH, Poddighe L, Delogu M, Holt RIG, Correll CU, Cortese S, Carvalho AF, Boyer L, Dragioti E, Du Rietz E, Firth J, Fusar-Poli P, Hartman CA, Larsson H, De Giorgi R, Lehto K, Lindgren P, Manchia M, Nordentoft M, Skonieczna-Żydecka K, Veroniki AA, Marx W, Campana M, Mortazavi M, Hasan A, Stubbs B, Taipale H, Vancampfort D, Vieta E, Solmi M ; ECNP PAN-Health Group. Disparities in diabetes treatment and monitoring for people with and without mental disorders: a systematic review and meta-analysis. The Lancet Psychiatry, 2026 ; 13(2) : 112-124. Published online on 5 January 2026. DOI 10.1016/S2215-0366(25)00332-3 · PMID 41506273. Open access under a Creative Commons licence. No funding declared for this work. Protocol and methodological appendices registered on OSF.