Summary of findings
We set out to identify the determinants of diabetes, to explore variation in health outcomes and mortality, and to investigate access to, and experiences of, diabetes health care for people with SMI and diabetes.
Underpinned by a social inequalities framework, we conducted a mixed-methods study, using a concurrent triangulation design, comprising:
- a quantitative longitudinal observational study, using the health records of 32,781 people with SMI (2761 of whom had diabetes) and of 9573 ‘controls’ with diabetes and no mental illness, matched on age, sex and practice from the CPRD (GOLD) database, with individual linkages to HES, ONS and IMD data
- a qualitative interview study with 39 adults with SMI and diabetes, nine family members and 30 health-care staff.
Emerging findings from each investigation iteratively informed the design and analysis of the other. Quantitative and qualitative results were integrated using charting and co-design methods to develop a more comprehensive and in-depth understanding of the relationship between SMI and diabetes.
Diabetes in people with severe mental illness
We found that the prevalence of T2DM was 10% in people with SMI. Sociodemographic factors associated with increased risk of diabetes in SMI included older age, older age at diagnosis of SMI, Asian or black ethnicity, living in areas of socioeconomic deprivation and family history of diabetes. Baseline cardiometabolic risks of overweight, cardiovascular disease, hypertension and higher baseline blood pressure, cholesterol and HbA1c levels were also associated with diabetes, as was being on antipsychotic medication. People with SMI who were prescribed atypical antipsychotics were 40% (OR 1.39) more likely to develop diabetes.
We observed a general trend of improved control of HbA1c, cholesterol and blood pressure from 2000 to 2016 in people with SMI and diabetes. These apparent improvements might reflect improvements in care, influencing better control over time, but may also be a function of changes in the composition of the study population. Recorded prevalence of diabetes increased throughout the period and practices were financially incentivised for case-finding from 2004 onwards. Less severe cases at earlier stages of disease may, therefore, have constituted an increasing proportion of the diabetes study population over time, as detection (and therefore correct categorisation) of diabetes ‘caseness’ improved.
Diabetes and mental health outcomes in people with severe mental illness and diabetes
Factors associated with poorer physical or mental health outcomes were complex and differed by outcome. For macrovascular complications, older age, longer diabetes duration, comorbidity, baseline cardiovascular disease, hypertension and prescription of antihypertensive agents were all risk predictors. A more deprived socioeconomic status was also weakly associated with a higher risk. Current smokers had almost double the risk compared with non-smokers, although missing data limit confidence in this finding.
Microvascular complications were predicted by older age, baseline cardiovascular disease and baseline HbA1c level exceeding the recommended threshold of 58 mmol/mol (7.5%). There was also weak evidence that more deprived socioeconomic status, longer duration of SMI and baseline prescription of antidiabetes medication increased the risk of developing these complications.
In contrast to the association with diabetes risk, ethnicity was not identified as a risk predictor for poor physical health outcomes in people with SMI and diabetes. This may be as a result of missing ethnicity data and small numbers recorded as being from ethnic minority populations in our data set.
For mental health outcomes, older age had a small but statistically significant protective effect. Women were more likely to have recorded depression and anxiety than men (which could indicate increased risk and/or increased recognition). Asian ethnicity increased the risk of depression, even after adjustment for baseline antidepressant prescriptions. People with bipolar disorder were more likely to have a relapse of SMI in the follow-up period than people with schizophrenia, and also had an increased risk of depression and anxiety. Clinical diagnosis of depression with psychosis, unsurprisingly, considerably increased the risk of depression and anxiety. An additional Charlson Index comorbidity was associated with reduced risk of SMI relapse but increased risk of depression and anxiety. Baseline prescription of antidiabetes medication was a significant predictor for SMI relapse, increasing risk by 34%. A possible explanation may be that being on such medication is associated with more severe/poorly controlled diabetes, which acts as a trigger for SMI relapse.
All-cause mortality in SMI and diabetes increased with older age, male sex, white ethnicity, Charlson Index comorbidity and socioeconomic deprivation; longer duration of SMI slightly reduced the effect on mortality. Schizoaffective disorder was associated with an increased risk of all-cause mortality, compared with schizophrenia. Baseline antidiabetes drug prescription increased all-cause mortality by 35%. Antipsychotics were also associated with higher mortality. Typical antipsychotics increased mortality risk by 27% and atypicals by 31%. These risks are likely to be confounded by higher severity of conditions (diabetes and SMI) requiring medication. Although missing data preclude definitive conclusions, people who used substances at baseline had a three times higher mortality risk than people with no substance use.
The impact of having severe mental illness for people with diabetes
We were interested in understanding the impact of having SMI for people with diabetes, and the impact of having diabetes for people with SMI.
We found evidence that people with SMI alongside diabetes had better control of diabetes and blood pressure and were more likely to receive physical health checks than people with diabetes alone. Despite this, we found that the risk of depression and all-cause mortality was increased, compared with people without SMI. We think this may be because of ‘service gaps’ in accessing elective or planned care that could improve health outcomes. This is supported by the findings on macrovascular complications, for which elective admissions were lower but emergency admissions higher, for people with SMI alongside diabetes than for people with diabetes alone. Similarly, as previously reported in the literature,124 there are likely to be gaps in recognition and reporting of health problems for this group. We found that the rate of macrovascular complications was similar in people with and people without SMI, but people with SMI were less likely to have a record of less acute conditions, such as chronic heart disease. Service gaps were also suggested by qualitative findings, which highlighted the additional barriers to receiving physical health interventions, structural inequalities and the sense of mental illness ‘overshadowing’ physical health needs from the perspective of both health-care providers and people with SMI. Challenges distinguishing between mental and physical illness symptoms, and the impact of low mood and anxiety on the importance given to, and the ability to manage, physical health may compound these difficulties.
The impact of having diabetes for people with severe mental illness
Diabetes alongside SMI increased the probability of receiving physical health checks, but was a significant risk factor for poorer physical and mental health outcomes than SMI alone. The risk of depression and anxiety was increased. From interviews, we found that worries about diabetes could make people feel depressed and anxious; this was also found to lower motivation and intentions to manage diabetes and physical health more generally in SMI. It could be challenging to distinguish between symptoms of mental illness relapse and symptoms of poor glycaemic control. Multimorbidity was common and posed a significant challenge; people with SMI frequently found it overwhelming to deal with several long-term conditions.
Ordering of severe mental illness and type 2 diabetes diagnosis
A surprising initial finding was the larger than expected proportion of people who had been diagnosed with T2DM before their SMI diagnosis. Although further investigation established that this was, in part, a function of using the necessary cut-off points to include only more reliable quality data (which had the unintended effect of selectively excluding people diagnosed with SMI at an earlier age), it was clear that the population diagnosed with T2DM before SMI constituted a significant proportion of those with comorbid diabetes and SMI. The NDA reported an even higher proportion of people with diabetes diagnosed before SMI (65%).144
The approach to restrict analyses to UTS data is one that is widely used in studies interrogating large health-care data sets. The findings about how this affected the make-up of the study population, differentially excluding more people diagnosed with SMI at an earlier age, have wider applicability. Despite its potential to significantly affect study results, to our knowledge, this limitation of applying UTS criteria has not been reported elsewhere.
Although not part of the initial objectives, we conducted a limited exploration of how diagnosis order affected demographics, baseline characteristics and outcomes for the study population. People diagnosed with SMI before T2DM were more likely to be diagnosed at an earlier age for both conditions, be male, be overweight, live in the most deprived neighbourhoods and be current smokers. Baseline HbA1c, cholesterol and blood pressure levels were higher. They also had, in general, better physical and mental health outcomes, with lower risk of macrovascular complications and hospital admissions. Rates of SMI relapse, depression and anxiety, and mortality were all lower. The baseline prevalence of microvascular complications was lower, but they were more likely to develop new complications during the follow-up period than people diagnosed with T2DM before SMI.
People diagnosed with T2DM before SMI were more likely to have depression with psychosis, and a higher prevalence of comorbidities and prescribed medication other than antipsychotics. This group had had diabetes for, on average, 5 or 6 years; therefore, they had the ‘opportunity’ to develop complications in the pre-baseline period. A higher baseline prevalence of microvascular complications, and fewer new microvascular complications during follow-up, would, therefore, be expected (as found in the data), compared with those diagnosed with SMI first.
Health-care use and costs for people with severe mental illness and diabetes
To our knowledge, this is the first study to compare health-care use and costs across sectors for people with SMI and diabetes and people with diabetes alone using patient-level data. We found that use of primary care services, hospital admissions and duration of hospitalisation were all significantly higher for people with SMI alongside diabetes. Health-care costs were correspondingly higher (around double) than costs for diabetes alone. These costs are likely to be an underestimate as they exclude hospital admissions to specialist mental health facilities (which are likely to be higher for people with SMI).
Health-care interventions and outcomes for people with severe mental illness and diabetes
Our investigation of health-care interventions and how these influence health outcomes using routine health-care records was constrained by limited availability of information about these interventions in the data set, and by confounding between intervention and outcomes variables. However, interviews and co-design workshops provided important insights about the experience of negotiating health care for people with SMI and diabetes. These indicated that people accessed care separately for their conditions: diabetes care from their general practice and mental health care from specialist mental health services.
Formal and informal relationships with others, including social (family and friends) and health-care professionals, were highly valued as an important source of support for health management, although they were notably absent for many people. Such relationships and continuity of care were seen as essential to support health-care needs, particularly for dealing with the complex and overwhelming challenges of living with multimorbidity. These relationships helped with motivation, as well as providing more practical support to navigate health care. A case management approach to help people navigate across health service silos, sharing health records across different services and increasing staff knowledge about mental and physical conditions and management of multimorbidity, could be key to improving care and outcomes for people living with the double burden of SMI and diabetes.
Comparison with other research
These findings confirm results of previous research and also address some important evidence gaps.
The prevalence of diabetes in people with SMI has been consistently estimated to be two to three times higher than for the general population. Our estimate of 10.2% is similar to previously reported rates from meta-analyses (10.2%)44 and from interrogation of routine health-care data (8.0%).145
A range of risk factors (which are similar to those in the general population) for developing diabetes in people with SMI have been previously described,25,146 but studies have been inconsistent. Stubbs et al.122 reported only increasing age to be associated with risk of T2DM. A later meta-analysis implicated older age, longer illness duration and multiepisode (SMI) status, but not sex, ethnicity or smoking status;44 however, multivariable meta-regression analysis found that multiepisode (vs. first-episode) status was the only significant predictor for T2DM. These findings diverge from our results, which found a number of additional demographic and health variables, including Asian or black ethnicity, to be T2DM risk factors in SMI. Another health-care database study with a high minority ethnic population reported that increased risk of T2DM in SMI was exacerbated by Asian or Afro-Caribbean ethnicity;147 neighbourhood socioeconomic disadvantage has been reported as a risk factor in a previous study.148 Antipsychotics, particularly atypical antipsychotics, are widely reported to be associated with increased risk of T2DM,64 which was confirmed by our results.
Most previous research has reported only on the population of people diagnosed with SMI before T2DM, has been cross-sectional or has not considered diagnosis order. We found that there were important differences between the two populations, people diagnosed with SMI before T2DM and people diagnosed with T2DM before SMI, and these differences warrant further exploration. Diabetes appears to increase the risk of poor mental illness outcomes. A cross-sectional study reported elevated risk for all mental illness in people with diabetes, with the greatest excess risk (around threefold) for schizophrenia.149
Although relatively limited, there have been qualitative studies using focus groups, individual interviews and surveys to explore the experience of people living with comorbid SMI and diabetes.150–153 A common theme reported in these studies is that mental illness care significantly overshadows diabetes care. Another strong theme across studies is that of feeling overwhelmed by multiple chronic comorbidities with differing self-care recommendations. Our study also found that mental illness symptoms and care needs dominated people’s day-to-day lives, casting a ‘shadow’ over everything else. Multimorbidity was also a consistent feature, making self-management and accessing health care extremely challenging. Furthermore, mood influenced the ability to manage, and importance given to managing, physical health. A further barrier to managing diabetes in SMI was that health care was experienced as being provided in condition-specific silos.
Previous studies on diabetes health-care interventions received by people with SMI (compared with people without SMI) have given conflicting results. The NDA144 found that, compared with the general diabetes population, people with SMI and T2DM were slightly less likely to receive NICE-recommended care checks, but there was no difference in treatment target achievement. Overall evidence suggests that levels of health checks and cardiovascular and metabolic control (e.g. cholesterol, blood pressure, HbA1c) are comparable (except for retinopathy checks).154–156 However, concerningly, despite similar achievement of treatment targets as the general diabetes population, morbidity and mortality are increased for people with SMI.154
Potential explanations for the NDA144 findings and the present study include limitations of data recording (see Limitations of this study). However, they can potentially also indicate problems with accessing health-care interventions for the same level of need. We found evidence suggesting service gaps, with similar rates of recorded macrovascular complications for people with SMI and those without SMI, but differing patterns of acute versus elective care between the two groups. A study investigating treatment of acute myocardial infarction in people with SMI157 found that they were less likely to receive revascularisation therapies than the general population, despite changes in care to improve this situation.
Strengths and limitations
Strengths of this study
We accessed a large, comprehensive longitudinal data set, CPRD (GOLD), with individual data linkages to HES, ONS and IMD data, allowing us to address important evidence gaps. Patient characteristics in the CPRD database have been shown to be broadly representative of the general UK population in terms of age, sex and ethnicity,98 meaning that our findings are likely to have wider generalisability.
There are a growing number of studies using routinely collected health-care data to explore mental and physical health, health care and outcomes for people with SMI.145,147,155 The large numbers of records that are usually involved mean that such studies can address important applied health research questions. However, none, to date, has reported using longitudinal data combined with qualitative inquiry to develop a more in-depth understanding than can be achieved by cross-sectional designs or quantitative analyses alone.
Our mixed-methods design means we have been able to fill in ‘gaps’ from the quantitative study, which allows us to theorise further on the meaning of the findings. Moreover, an integrated approach in which emerging findings from both quantitative and qualitative inquiry informed further questions, allowed us to conduct a more detailed, in-depth exploration to address study objectives. The large qualitative sample size and inclusion of people with SMI and diabetes, their families and a variety of health-care staff ensured representation of a wide range of views in our data. Co-design workshops also ensured that, in addition to our multidisciplinary research team, people with SMI and diabetes, their families and health-care providers helped to interpret study findings. This strengthens the reliability and relevance of our findings, grounding them in the day-to-day experience of people living with and managing SMI and diabetes. Strong PPI and clinical input have also helped strengthen interpretation of findings and their relevance and application to practice.
Limitations of this study
As with all studies using routine health-care data, the quality of the data, in terms of accuracy and completeness, will influence the questions that can be addressed and the reliability of findings. As with all observational studies, it is not possible to control for unobserved confounders, and systematic biases in measurement errors might lead us to over- or underestimate associations of risk factors with outcomes for people with SMI. We also did not have access to contextual data on lifestyle, or to the environmental and social determinants of health that were not available in the health-care record or through data linkages.
Despite being representative of the UK population, the CPRD data set has some key limitations. Geographically, practices with large list sizes in urban areas in the south and north-west of England are over-represented, whereas practices in the north-east are under-represented.98
Missing data in the CPRD limited some of the analyses. We found large numbers of missing data for lifestyle factors (e.g. smoking), which are coded less systematically than information on clinical diagnosis and prescriptions. It cannot be assumed that these data are missing at random, as missingness may be influenced by various factors, including the coding behaviour of primary care providers, a patient’s health status and the implementation of financial incentive schemes. Changes in QOF indicators, for example, are likely to change recording behaviour, not just practice. Therefore, it is difficult to adjust for missing data, as the potential impact of all of these factors is unknown.
Data for health-care interventions and pathways were limited in the CPRD (because of unreliability, e.g. of referral data for diabetes education and specialist services, and a lack of linked outpatient data in our data set). The ability to explore the influence of health-care interventions on health outcomes was, therefore, restricted. Related to this, when we found differences in provision of interventions for people with SMI, it was not possible to identify the level of the ‘service gap’ (referral or access) from the CPRD. We were also unable to measure severity of mental illness or diabetes, as this information is not routinely recorded, and cannot be reliably inferred from other parameters.
The study was also limited by the accuracy and completeness of the recording of diagnoses and activities by practices. The validity of the Read coding system used in practices’ electronic medical records varies over time and by diagnosis: in a 2010 review, estimates of the positive predictive value of recorded diagnostic codes in the General Practice Research Database system (precursor to the CPRD) were 98.6 (95% CI 92.2 to 100.0) for diabetes and 81.0 (95% CI 87.0 to 94.0) for schizophrenia.158 However, completeness of recording in the CPRD has improved over time, particularly for conditions included in the QOF, for which standard reporting requirements (codes specified under QOF business rules) were introduced. This includes both diabetes and SMI. Although we based our diagnostic criteria on QOF business rules, we also extended our definitions to include patients for whom there was a reasonable presumption of having the condition (e.g. patients without a recorded diagnosis of diabetes receiving regular prescriptions of insulin or antidiabetic medication).
We estimated hospital use and costs based on the HES data set, which contains detailed information about admissions to general hospitals (including non-specialist mental health providers), but not admission to specialist mental health facilities. Use of community mental health care, outpatient attendances and A&E attendances was also missing from our data (because of limitations on how many data sets can be linked to avoid inadvertent identification of practices). A further limitation was that costs for diabetes care could not be separated from other costs.
The qualitative findings indicated fluctuating diabetes management because of changes in mental health or emotional states. However, this was not captured in the CPRD data because of large gaps between diabetes checks (often 6 or 12 months). Eligibility criteria for the interview study necessarily excluded people who were too unwell to participate and probably selected people with less severe SMI and better diabetes control. They were also less likely to be homeless or have unstable housing.
We were not able to achieve the target sample size for family members, as many participants with SMI and diabetes were not able to identify a family supporter. However, family members participated in the PPI group and co-design workshops, providing valuable insights.
Conclusions
We believe that this is the first study to combine interrogation of a large health-care data set with in-depth interviews to identify the factors associated with developing T2DM and those influencing mental and physical health outcomes for people with SMI and diabetes, and to estimate health-care costs and explore the experience of living with SMI and diabetes.
Implications for practice
The findings from this research indicate that the following five key messages are important for clinical practice.
Regular diabetes checks for people with severe mental illness
Check all people with SMI regularly to identify those with diabetes and those at high risk of developing diabetes and offer appropriate intervention.
Address the metabolic side effects of psychotropic medication
Although psychotropic medication forms a key part of the management of SMI, many of these medications are linked to weight gain and diabetes. Clinicians should therefore carefully consider the choice of psychotropics, discuss the side effects with patients, monitor for metabolic side effects and intervene when necessary.
Respond to the challenges of multimorbidity
Health-care professionals and services must take account of multimorbidity in the assessment and management of diabetes.159 Possible solutions include a keyworker or case management approach to better integrate support for both mental illness and diabetes, and improved training for health-care staff on managing multimorbidity.
Provide tailored support for self-management
Provide a bespoke diabetes self-management education package for people with SMI that addresses the unique challenges to having these conditions together. These include managing side effects of antipsychotic medications, managing the impact of fluctuating mental health and mood on diabetes self-management, and distinguishing between mental and physical symptoms.
Address social concerns
Health-care services and professionals should identify difficulties related to diabetes management that stem from social issues, such as housing, discrimination, transport, poverty, access to benefits and social capital, given the complex interactions between socioeconomic factors and diabetes and SMI. When these are identified, patients should be signposted to appropriate services.
Implications for research
This study has implications for future research on SMI and diabetes, and also for research using similar methods.
Understand the relationship between health checks and outcomes
This study identified a paradoxical association between the number of health checks and poorer outcomes. There are a number of potential explanations for this finding that require further investigation, for instance reverse causality or lack of effective intervention. Exploration of this association using advanced statistical methods (such as instrumental variables regressions) and dynamic models is a research priority.
The relationship between severe mental illness and other long-term physical conditions
For this study, we constructed a large number of variables and will make the codes available for other researchers, reducing future resource needs for conducting research using these data sets. There is an opportunity, therefore, to replicate our methods to efficiently investigate other health conditions, such as cardiovascular disease, chronic lung disease and other chronic disorders.
Is diabetes a risk factor for older-age onset of psychosis?
An unanticipated finding of the study was that the diagnosis of diabetes often antedated the onset of SMI. We were able to conduct only limited analyses because of the nature of the data set. Further work is needed using a cohort of people with T2DM to fully understand this relationship.
Are tailored diabetes interventions clinically effective and cost-effective compared with generic interventions for people with severe mental illness?
Common across all of the recommendations for practice is the need for more intensive support, requiring additional resources for this high-risk and vulnerable group. This study design could not assess the effectiveness and cost-effectiveness of such approaches. Research is needed to develop and to determine the effectiveness and cost-effectiveness of such interventions (e.g. tailored diabetes self-management education, or changes to organisation of diabetes care for people with SMI) for improving diabetes and other outcomes.