Interrogation of patient health records

Introduction

The interrogation of patient health records study addressed the following research questions:

• What are the sociodemographic and illness-related risk factors associated with –
  • diabetes developing in people with SMI?
  • variation in diabetes and mental health outcomes in people with SMI and diabetes?
• How do physical and mental health outcomes differ between people with SMI and diabetes and people with –
  • SMI without diabetes?
  • diabetes but no SMI?
• What factors are associated with access to, and receipt of, diabetes care for people with SMI?
• How and at what cost is diabetes monitored and managed in people with SMI, compared with those without SMI?
• What health-care interventions (e.g. medication, referrals and care pathways) are associated with better diabetes outcomes for people with SMI and diabetes?

These correspond to study objectives 1–4, 6 and 7 (see Chapter 2, and the objective sections later in this chapter). For all objectives, we conducted analyses in line with the inequalities framework to quantify the relative effect of social inequalities on quality of care and outcomes. Specifically, we used deprivation and disadvantage markers, such as the IMD, as independent variables to estimate gap or gradient effects. We planned to stratify analyses by ethnicity, but were limited by the small numbers of people in minority ethnic categories for most analyses.

Data and methods

Ethics approval

A data-use agreement for CPRD records and linked HES and ONS mortality data was granted by the International Scientific Advisory Committee (reference: 17_161R).

Data sets

Three data sets were extracted from the CPRD for the quantitative analyses; the relationships between these data sets are shown in Figure 2. Data set A contains anonymised longitudinal health records of a cohort of 32,759 people with SMI. Records were extracted from the CPRD GOLD population cohort if they:

• had at least one clinical or referral event with a SMI diagnostic code in their medical history (see Report Supplementary Material 2, Table S1, for the code list)
• were actively registered with one of the contributing general practices in England in the study period from 1 April 2000 to 31 March 2016
• were eligible for the linkages required (HES, IMD and ONS death)
• were aged ≥ 18 years when SMI was diagnosed
• had continuous health data up to research standard [up to standard (UTS)] in the study period.98

FIGURE 2 Venn diagram illustrating the relationship between CPRD data sets. (Figure not reproduced here; see the original article.)

Data set B is a subset of data set A and contains records of 2761 people with SMI who also had T2DM. T2DM was defined as the presence of at least one clinical diagnostic code in a patient’s clinical or referral records (see Report Supplementary Material 2, Tables S2 and S3). We excluded people who had a record of type 1 diabetes mellitus after their diagnosis of T2DM.

Data set C comprises a cohort of 9573 control patients with a diagnosis of T2DM and no record of SMI who were matched to people in data set B (cases). These controls, with a diagnosis of T2DM and no record of SMI, were matched to cases based on age, sex and practice in a ratio of 4 : 1.

Variables

Using the longlist identified, as described in Chapter 4, Results, we constructed a range of candidate variables associated with the development of T2DM and with physical and mental health outcomes based on whether or not (1) the risk factor could be measured using the information in the CPRD and its linked data sets, (2) the data had been collected in the CPRD and its linked data sets and the quality of data recording was systematic enough for analysis and (3) it was likely to have an impact on the outcomes of interest.

Table 3 summarises the data sources and construction methods for the candidate risk factors that were used to measure patients’ baseline characteristics.

It should be noted that patients’ follow-up periods varied for each objective, depending on the patient samples and research questions. We provide a detailed definition of follow-up period for each objective in the sections that follow.

Table 4 summarises the outcome variables under examination, along with their associated data sources, analysis methods and related objective(s).

Methods

To improve data quality, we developed and applied four common inclusion criteria to the extracted data set before carrying out analyses. These criteria were designed to account for the nature of SMI and diabetes, as well as for the features and limitations of this longitudinal data set:

• First, we identified the diagnosis dates of SMI and diabetes. Diagnosis date is commonly identified as the date when the earliest diagnostic code was recorded in primary care. Because the first contact for (and diagnosis of) SMI can be in secondary care, we identified the diagnosis date for SMI as the earliest of the date of first GP diagnosis or the date of the first hospital admission for the condition. We also applied this method to the identification of diabetes diagnosis, although this condition is more commonly diagnosed in a primary care setting. This increased the number of people with SMI (data set A) to 32,781 and people with both SMI and T2DM (data set B) to 3448. Our matched controls (data set C), therefore, decreased in number to 9551 people.
• Second, we excluded patients whose diagnosis of SMI and diabetes was recorded in the 90 days following registration with their current practice. As patients’ primary care records could be transferred between practices, we used this step to exclude those patients whose earliest primary care records of SMI and diabetes were likely to be updated from medical history because of changes of practices, rather than a new diagnosis.
• Third, we excluded patients who were diagnosed with SMI and diabetes before the age of 18 years, using our modified diagnosis dates.
• Fourth, we identified the patients with diagnosis codes of both type 1 diabetes and T2DM and removed people with potential type 1. For the patients with codes of both types, we followed an existing identification algorithm (from study NIHR 14_168R113), which categorised patients as potentially having type 1 diabetes if (1) their first diagnostic code of diabetes was recorded before 18 years of age or (2) there was a diagnostic code of type 1 recorded before 18 years of age or (3) they had been treated with insulin only.

Most analyses were conducted on patient-level data sets that typically had two levels of hierarchy: (1) patient-level observations nested within (2) practices. We accounted for the data structure by applying multilevel mixed-effects models with practice-level random intercepts.

Although a rich medical information source, these routine primary care data have been collected for administrative purposes, and the data quality depends on various factors such as the accurate and complete utilisation of the computerised recording system, standardised coding and the introduction of financial incentive schemes (such as the QOF) that reward general practices for performing incentivised Read-coded activities. We therefore identified an UTS follow-up period for each patient, which included comprehensive medical records for research. The start of the UTS period was defined as the later date between patient current registration and practice UTS dates; the end of this period was identified as the earliest among patient death, transfer-out and practice last collection dates.

We took a step-by-step forward approach to build the models from the most basic form (including only exposure variables) to expanded models with a wide range of adjustments. This approach was adopted to detect the potential confounding effect one variable might have on another, such as the association between SMI status and health outcomes attributed to older age. We adjusted for patient characteristics and time in all the final models as reported in this report (see Variables and statistical methods and Report Supplementary Material 3). Other candidate explanatory variables were included in the final models if they had significant association (p < 0.1) with the outcome of interest or their inclusion had improved the goodness of fit of models. We used c-statistics to compare logistic models and the Akaike information criterion and Bayesian information criterion (BIC) for negative binomial and Cox proportional hazards models. In the results tables (see Tables 6, 9, 13, 16, 22 and 23, and Report Supplementary Materials 4–12), we report a core model including only risk factors with good data quality and an extended model that also included covariates with less complete data, such as lifestyle and biometric measures.

Objective 1: factors associated with the development of diabetes in people with severe mental illness

Objective

The objective was to identify which sociodemographic, illness, family history and lifestyle factors are associated with the development of diabetes in people with SMI.

Study population

We used data set A (see Figure 2) containing records of 32,781 adults with SMI and applied the four common inclusion criteria, as outlined in Methods. From the remaining sample, people were included in the analysis if they (1) were diagnosed with SMI in their UTS continuous data period, (2) either had no record of T2DM in their medical history or had T2DM diagnosed only after SMI and (3) had a follow-up period of at least 1 day.

The follow-up period of eligible patients started from the later date between SMI diagnosis and UTS period start plus 15 months, to ensure that at least 15 months’ UTS continuous data were available for extracting baseline patient characteristics. The end of the follow-up period was identified as the end of the UTS continuous data period or the end of the study period, whichever was earlier. The final sample eligible for analysis included 14,838 people with SMI and without T2DM at baseline.

Variables and statistical methods

Descriptions of the outcome and candidate explanatory variables used in analyses for this objective, and their associated data sources and analysis methods, can be found in Tables 3 and 4. Further details on the statistical model specification have been provided in Report Supplementary Material 3.

For the status of diabetes, we adjusted for the length of follow-up (in years) and financial years to account for the effect of time in the multilevel logistic regressions. Robust standard errors for correlations by practices were specified in these regressions.

Results

Descriptive statistics

The distribution of patient characteristics at baseline was summarised according to diabetes diagnosis (Table 5).

Of the 14,838 people with SMI at baseline, 707 (4.8%) were diagnosed with T2DM in the follow-up period. This rate is lower than estimates from previous studies,44,122 and led us to further explore the sample characteristics.

The estimate above excludes those who developed diabetes before the onset of SMI. We found that, if we removed the criterion to have the recorded T2DM diagnosis after the diagnosis of SMI, in the 15,984 eligible patients whose SMI diagnosis was within the UTS data period, 1452 (9.1%) had a diagnosis of T2DM; of these, 782 (53.9%) had SMI diagnosed before T2DM and 670 (46.1%) had SMI diagnosed after or on the same day as T2DM. These figures are comparable to findings from the 2016–17 National Diabetes Audit (NDA),123 which reported that 34.9% of people with SMI and T2DM were diagnosed with SMI first, but are unexpected, as it is generally thought, from clinical experience, that most people with comorbid SMI and diabetes develop SMI first.

In the total sample of 29,281 people with SMI (not restricted to the UTS period), we found that 2984 (10.2%) patients had T2DM, and that 2209 (74.0%) of these had SMI before T2DM, which is more consistent with reported prevalence in the literature and with experience of diagnosis order from clinical practice. This demonstrated that, by restricting SMI diagnosis within the UTS data period in our initial analysis, we selectively excluded patients diagnosed with SMI at an earlier age; therefore, the patients remaining in the sample were more likely to have late-onset SMI.

Using the restricted UTS sample of 14,838, people who developed T2DM were diagnosed with SMI at an older age were more likely to be of Asian or black ethnicity and were more likely to live in the most deprived neighbourhoods than patients with no record of T2DM. Furthermore, people with T2DM had a higher baseline prevalence of cardiovascular disease and hypertension and were more likely to have been prescribed the medications under investigation.

Table 5 also shows that a large proportion of the sample had no baseline records for lifestyle factors and biometric measures. Based only on available data, the descriptive statistics showed that people with T2DM were more likely to be overweight; to have higher baseline levels of blood pressure, cholesterol and HbA_1c; and to have a family history of diabetes.

Regression analyses results for the risk factors for development of type 2 diabetes

Regression results are summarised in Table 6, and Appendix 3, Figure 10, illustrates the non-linear impact of age on the risk of diabetes.

The key risk factors for T2DM identified in the regression analyses were consistent across the four models, as were their estimated associations with the risk of T2DM. In the core model, results showed that older age appeared to be a risk factor for developing diabetes until around age 60 years [odds ratio (OR) 1.17, 95% CI 1.14 to 1.21], after which this impact would decrease, as suggested by the impact of quadratic age (see Appendix 3, Figure 10). For patients from an Asian minority background, the risk of T2DM increased by 147% (OR 2.47, 95% CI 1.78 to 3.42); for those from a black minority background, the risk of T2DM increased by 85% (OR 1.85, 95% CI 1.25 to 2.75). People who lived in the most deprived neighbourhoods were more likely, by around 86%, to develop diabetes (OR 1.86, 95% CI 1.39 to 2.48) than people from the most affluent areas.

After adjusting for patient sociodemographic characteristics and SMI status (e.g. type and duration), having comorbid hypertension and cardiovascular disease at baseline increased the risk of diabetes by 80% (OR 1.80, 95% CI 1.42 to 2.28) and 41% (OR 1.41, 95% CI 1.03 to 1.93), respectively. Patients who had been prescribed antipsychotics, particularly atypical antipsychotics, were more likely to have diabetes than people without these medications. All of these findings are consistent with expectations.

The extended model contained variables with higher proportions of missing values. The estimated results of this model, therefore, should be interpreted with caution. There seemed to be a strong association between BMI above the normal range and increased diabetes risk. This risk was doubled for patients who were overweight (OR 2.27, 95% CI 1.53 to 3.37), four times higher for those who were obese (OR 4.22, 95% CI 2.86 to 6.20) and 10 times higher for those who were severely obese (OR 10.94, 95% CI 6.61 to 18.10). Results from the Cox proportional hazards regressions generated a similar set of predictors for T2DM, the estimated impact of which was also comparable to that of the multilevel logistic regressions (see Table 6).

Summary of findings

Type 2 diabetes prevalence was 10.2% in the total sample of 29,281 people with SMI and 9.1% in the UTS-restricted sample (n = 15,984). The key predictors for developing T2DM in people with SMI included older age, ethnicity (Asian and black), socioeconomic deprivation, physical comorbidities (hypertension and cardiovascular disease) and antipsychotics. Potential risk factors may also include being overweight or obese, and glucose dysregulation.

Objective 2: factors associated with variation in diabetes and mental health outcomes in people with severe mental illness and diabetes

Objective

The objective was to identify which sociodemographic, illness, family history and lifestyle factors are associated with variations in diabetes and mental health outcomes among people with SMI and diabetes.

Study population

We used data set B containing records of 3448 patients with a clinical diagnosis of both SMI and T2DM, identified from either primary care records or hospital admissions (see Figure 2 for details of relationships between data sets). After applying the common eligibility criteria regarding registration, diagnosis age and diabetes type, there were 2984 patients in the sample.

The start of the follow-up period was identified as the latest date of the recorded SMI diagnosis, T2DM diagnosis or the start of UTS data period plus 15 months, so that patients were followed up after the second diagnosis and there was a 15-month window to obtain their baseline characteristics. The end of follow-up was identified as the end of the UTS data period or the end of the study period date, whichever was earlier. A total of 2754 patients with a follow-up length of at least 1 day were included for analysis.

Variables and statistical methods

Candidate risk factors are summarised in Table 3. The physical and mental health outcomes under examination included:

• glycaemic and cardiovascular control (HbA_1c, blood pressure and cholesterol levels)
• diabetes complications (hyperglycaemia, hypoglycaemia, microvascular complications and macrovascular complications)
• hospital admissions for macrovascular complications
• mental health outcomes (SMI relapses and depression and anxiety)
• mortality.

Details on the construction and statistical methods of these outcome variables are provided in Table 4, and further details on the methodology can be found in Report Supplementary Material 1.

Results

Descriptive statistics

Patient baseline characteristics are summarised in Table 7 and health outcomes are summarised in Figures 3 and 4, and Table 8. These descriptive statistics are reported for the whole eligible sample, as well as by diagnosis order.

FIGURE 3 Average cholesterol and HbA1c, 2000/1–2015/16. (a) Average cholesterol (mmol/l) and 95% CI; and (b) average HbA1c (%) and 95% CI. (Figure not reproduced here; see the original article.)

FIGURE 4 Average blood pressure (mmHg), 2000/1–2015/16. (a) Systolic blood pressure and 95% CIs; and (b) diastolic blood pressure and 95% CIs. (Figure not reproduced here; see the original article.)

Table 7 shows that 73.3% (2019/2754) of eligible patients had SMI diagnosed before T2DM, and 26.7% had T2DM diagnosed first (including same-day diagnosis). Patient characteristics differed between the two groups. Patients who had SMI before T2DM were diagnosed at an earlier age for both conditions: average diagnosis age was 41.9 years for SMI and 55.0 years for T2DM. In comparison, patients who had T2DM first were diagnosed with T2DM at around 57.1 years of age and with SMI at around 63.3 years of age. Patients who had SMI before T2DM were also more likely to be male and to live in the most deprived neighbourhoods.

Patients who had T2DM before SMI were more likely to have depression and psychosis as the type of SMI. The prevalence in this group was more than double the prevalence in people diagnosed with SMI first. These patients also had a higher prevalence of comorbidities and were more likely to be prescribed medications other than antipsychotics.

There were fewer missing data for baseline lifestyle factors and biometric measures in patients who had T2DM diagnosed first than in people diagnosed with SMI first. The latter were more likely to smoke cigarettes (although there were a large number of missing data in both groups, limiting definitive conclusions). People diagnosed with SMI first also had higher levels of HbA_1c, cholesterol and blood pressure at baseline, and were more likely to be obese than people diagnosed with T2DM first.

For physical health outcomes, we examined the variations in glycaemic and cardiovascular control by plotting time trends against financial years (see Figures 3 and 4). There was a general trend for improved control among patients in both groups between 2000 and 2016. On average, people who had T2DM before SMI had better baseline control, and continued to have similar, or better, control for cholesterol and blood pressure in their follow-up periods. However, the average levels of HbA_1c in this group were consistently above the NICE-recommended threshold of 58 mmol/mol (7.5%) in this period, whereas the average levels of HbA_1c among people diagnosed with SMI first dropped (and remained) below this threshold from around 2004/5.

The apparent improvements in cholesterol, HbA_1c and blood pressure might reflect improvements in care and control over time, but they might also reflect 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. As case-finding improved, less severe cases at earlier stages of the condition may have constituted an increasing proportion of the study population over time.

As summarised in Table 8, people who had SMI before T2DM had, in general, better physical and mental health outcomes. Patients in this group were less likely to develop macrovascular complications and had fewer hospital admissions for these complications in the follow-up period. These patients also had lower rates of SMI relapses and depression and anxiety, and had a lower mortality rate. The SMI before T2DM group had a lower baseline prevalence of microvascular complications, but were more likely to develop these conditions during the follow-up period than the other group. Those diagnosed with diabetes before SMI will have had diabetes for 5 or 6 years before entering the data set for this analysis; therefore, they would have had the ‘opportunity’ to develop complications in the pre-baseline period. As a result, they would be expected to have a higher baseline prevalence of microvascular complications (and fewer new microvascular complications during follow-up) than those diagnosed with SMI first.

Regression analyses results

The adjusted risk factors for physical health outcomes are reported in Table 9. Among the patient sociodemographic characteristics, older age was a significant predictor of macrovascular complications and more deprived socioeconomic status was also weakly associated with a higher risk of complications.

Diabetes duration of 1 additional year increased the odds for primary care diagnosis of macrovascular complications by around 3% (OR 1.03, 95% CI 1.00 to 1.06) and for hospital admissions by 4% [incidence rate ratio (IRR) 1.04, 95% CI 1.01 to 1.07]. The risk of these complications in patients with cardiovascular disease at baseline was more than double the risk in patients without this comorbidity; this increased by 198% (OR 2.98, 95% CI 1.97 to 4.49) for primary care diagnosis and by 171% (IRR 2.71, 95% CI 1.87 to 3.92) for hospital admissions. Risk predictors of macrovascular complications also include comorbid hypertension and the baseline prescription of antihypertensives. These two predictors are correlated because of their nature, and we included them separately in our regressions. We chose between alternative models using the c-statistic for the logistic regressions and BIC for negative binomial regressions. Furthermore, there was evidence suggesting that the presence of additional Charlson Index comorbidities was associated with increased hospital admissions.

The adjusted risk of microvascular complications was predicted by older age and baseline cardiovascular disease. There was also weak evidence that more deprived socioeconomic status, longer duration of SMI and baseline prescription of antidiabetes medications may also be associated with an increased risk of developing these complications. Ethnicity was not associated with risk of either macrovascular or microvascular complications.

All-cause mortality increased with demographic characteristics, including older age, male sex, white ethnicity and socioeconomic deprivation. Longer duration of SMI was associated with slightly reduced risk of mortality, and certain types of SMI, such as schizoaffective disorder, were associated with an increased risk of mortality, compared with having schizophrenia. One additional comorbidity of the Charlson Index increased the risk by 16% [hazard ratio (HR) 1.16, 95% CI 1.04 to 1.28]. Antipsychotics were associated with higher risk of mortality: those on typical antipsychotics had an increased risk of mortality of 27% (HR 1.27, 95% CI 1.02 to 1.59) and those on atypical antipsychotics had an increased risk of mortality of 31% (HR 1.31, 95% CI 1.08 to 1.60). The baseline prescription of antidiabetes medications also increased the risk by 35% (HR 1.35, 95% CI 1.10 to 1.65). We were not able to control for severity of mental illness or diabetes.

The regression results for mental health outcomes (SMI relapse and depression or anxiety) are summarised in Table 10. After adjustment, older age had a small but statistically significant impact on mental health outcomes; the risk reduced by ≈ 2% for each additional year of age [OR 0.98 (95% CI 0.98 to 0.99) for SMI relapses and OR 0.98 (95% CI 0.97 to 0.99) for depression and anxiety]. Female patients were more likely to have depression and anxiety (OR 1.27, 95% CI 1.03 to 1.58). Both SMI relapse and diagnosis of depression or anxiety were affected by SMI type. Patients with bipolar disorder were more likely to have a relapse of SMI in the follow-up period than patients with schizophrenia (OR 1.36, 95% CI 1.12 to 1.66) and had a similar increased risk for depression and anxiety (OR 1.37, 95% CI 1.07 to 1.77). Patients with a clinical diagnosis of depression and psychosis had an increased risk of depression and anxiety (OR 2.89, 95% CI 2.03 to 4.13) during follow-up. In terms of comorbidities, we found that the presence of dementia and learning disability was associated with reduced risk of SMI relapses. An additional Charlson Index comorbidity was associated with a lower risk of SMI relapses (OR 0.85, 95% CI 0.75 to 0.95) but a higher risk of clinical events of depression and anxiety (OR 1.20, 95% CI 1.05 to 1.37). Furthermore, baseline prescription of antidiabetes medications was a significant predictor for SMI relapses (OR 1.34, 95% CI 1.08 to 1.66).

In sensitivity analyses to investigate whether or not health outcomes and their association with risk factors differed by diagnosis order, we included diagnosis order in the regression models as an explanatory variable and tested its impact on outcomes as well as its interacting relationship with other risk factors. Adjusting for diagnosis order had a similar effect to adjustment for the duration of conditions. For instance, macrovascular complications were associated with longer duration and T2DM. When conducting regression analyses on a restricted sample that included only those patients who had SMI diagnosed first, the results were consistent with those obtained from the unrestricted models.

Results of the extended models can be found in Report Supplementary Material 4. In these models, we investigated the impact of potential risk factors, such as lifestyle, biometric measures and family history, on physical and mental health outcomes. The data on these variables were less complete and the patterns of missing data were unlikely to be random, meaning that these results should be interpreted with caution. The results showed that current smoker status increased the risk of macrovascular complications (primary care diagnosis), compared with non-smokers (OR 1.91, 95% CI 1.19 to 3.06); the risk of mortality was more than three times higher for patients who used substances at baseline (HR 3.30, 95% CI 1.58 to 6.90); and a baseline HbA_1c level exceeding the recommended threshold was associated with a higher risk of developing microvascular complications (HR 1.66, 95% CI 1.33 to 2.08).

Summary of findings

The descriptive statistics showed that people diagnosed with SMI before T2DM differed from those diagnosed with T2DM first in characteristics including diagnosis age, types of SMI, socioeconomic status, comorbidities, medication and lifestyle, as well as baseline diabetes and cardiovascular controls.

After adjustment using regression analyses, we found that older age, socioeconomic deprivation and the presence of multimorbidity were the common risk predictors for physical health outcomes. The duration of T2DM and the use of medications were also associated with poorer status of some of these physical health outcomes. All-cause mortality risk increased with older age, male sex, white ethnicity and socioeconomic deprivation. The risk of poorer mental health outcomes, however, was primarily predicted by younger age, type of SMI and comorbidities.

Objective 3: comparing health-care interventions and health outcomes in people with severe mental illness and diabetes with those in people with severe mental illness without diabetes

Objective

The objective was to compare health-care interventions and physical and mental health outcomes in people with SMI and diabetes with those of people with SMI without diabetes.

Study population

We used data set A, containing records of 32,781 people with a clinical diagnosis of SMI (see Figure 2). After applying the common inclusion criteria regarding diagnosis date and age, practice registration, and type, 29,281 patients remained in the sample.

The start of follow-up was calculated as the later date of SMI diagnosis or the start of UTS data plus 15 months, so that each patient was followed up from (or after) the diagnosis of SMI, with a 15-month window for collecting baseline characteristics. The end of follow-up was identified as either the end of UTS data or the end of the study period, whichever occurred first. A total of 26,977 patients with a follow-up length of at least 1 day were included in the analyses.

Variables and statistical methods

The exposure variable for this objective was the status of T2DM. The other candidate explanatory variables are summarised in Table 3. The outcome variables under investigation were as follows (see Table 4 for further details):

• glycaemic and cardiovascular control (HbA_1c, blood pressure and cholesterol levels)
• diabetes complications (hyperglycaemia, hypoglycaemia, microvascular complications and macrovascular complications)
• hospital admissions for macrovascular complications
• mental health outcomes (SMI relapses)
• all-cause mortality
• health checks (glucose, cholesterol, blood pressure and BMI).

For this objective, we used SMI relapse as the mental health outcome. Markers of depression and anxiety were not examined as the clinical Read codes for these overlapped with the diagnostic codes of depression and psychosis (a type of SMI).

Variation in receiving health-care interventions was explored using levels of health checks. The QOF61 rewards primary care providers for performing regular health checks and monitoring the physical health status for people with a wide range of long-term conditions, including SMI. We therefore selected four QOF-incentivised physical health checks (glucose, cholesterol, blood pressure and BMI) to examine their provision and association with T2DM status. For further details on the methodology, see Report Supplementary Material 3.

Results

Descriptive statistics

The baseline characteristics for eligible patients are summarised in Table 11. Of the 26,977 people with SMI, 2755 (10.2%) also had a diagnosis of T2DM. Patients with T2DM were diagnosed with SMI at an older age; the mean diagnosis age was 40.7 years for people without T2DM and 47.6 years for people with T2DM. People with T2DM were more likely to be female, to be of Asian or black ethnicity and to live in socioeconomically deprived neighbourhoods than people without T2DM. Furthermore, there was a higher probability of having a family history of diabetes in people with T2DM.

The baseline prevalence of cardiovascular disease and hypertension was substantially higher in people with both SMI and T2DM than in people without T2DM, consistent with the patterns of baseline prescriptions; the proportions of people receiving antihypertensive and lipid-lowering drugs were much higher in those with T2DM than in those without T2DM. People with both diagnoses were also more likely to receive other medications, including antidepressants, antipsychotics and antidiabetes medications.

People with T2DM, on average, had a higher baseline BMI and poorer glycaemic and blood pressure controls. The average cholesterol level, however, was lower in people with T2DM than in people without T2DM. In general, there were large numbers of missing data for lifestyle factors and biometric measures in people without T2DM, and the patterns of missing values for these variables differed between the two groups. Therefore, these comparisons should be interpreted with caution.

The patterns for the health outcomes of diabetes and cardiovascular control are shown in Figures 5 and 6. These figures show a common trend: that diabetes and cardiovascular control measures improved from 2000/1 through to 2015/16 for both people with and people without T2DM. The gap between these two groups, however, widened for average cholesterol and HbA_1c levels.

FIGURE 5 Objective 3: average cholesterol and HbA1c, 2000/1–2015/16 for people with SMI with T2DM and for people with SMI without T2DM. (a) Average cholesterol (mmol/l) and 95% CIs; and (b) average HbA1c (%) and 95% CIs. (Figure not reproduced here; see the original article.)

FIGURE 6 Objective 3: average blood pressure (mmHg) and 95% CI, 2000/1–2015/16 for people with SMI with T2DM and for people with SMI without T2DM. (a) Systolic blood pressure and 95% CIs; and (b) diastolic blood pressure and 95% CIs. (Figure not reproduced here; see the original article.)

People with T2DM who had a lower cholesterol level at baseline continued to have better cholesterol control in the follow-up period than people without T2DM. The average level dropped below 5 mmol/l, the NICE-recommended threshold, after 2003/4 for people with T2DM, whereas the average level for people without T2DM remained above this threshold in this period.

The average level of HbA_1c for people with T2DM was consistently higher than the average level for people without T2DM. There was a decreasing trend in people without T2DM, a pattern that might suggest that an increased number of T2DM patients had been identified among people with SMI through regular physical health checks, as incentivised under QOF.

For blood pressure, the gap between people with and people without T2DM remained constant over time. Systolic blood pressure was higher in people with T2DM throughout this period, a pattern consistent with the baseline measures of these two groups. Although diastolic blood pressure was higher in people with T2DM at baseline, results suggested better control for these patients during follow-up.

Descriptive statistics for the other health outcomes and physical health checks are summarised in Table 12. There was a general pattern that people with T2DM had higher crude risks for poorer physical and mental health outcomes, and received more frequent physical health checks, than people without T2DM.

Regression analysis results for the impact of type 2 diabetes on health outcomes and health-care interventions

The impact of the exposure variable, the status of T2DM, on health outcomes and health-care interventions is summarised in Table 13 using results from the core models. The diagnosis of T2DM was associated with increased risk of poorer physical and mental health outcomes, as well as increased levels of physical health checks. After adjusting for patients’ sociodemographic characteristics, baseline comorbidities and medication use, people with T2DM had an increased risk of macrovascular complications compared with people without T2DM; clinical events recorded in primary care increased by 36% (OR 1.36, 95% CI 1.11 to 1.67) and hospital admissions increased by 62% (IRR 1.62, 95% CI 1.31 to 2.01). The risk of microvascular complications was more than 22 times higher (HR 22.23, 95% CI 17.79 to 27.77) in people with T2DM than in people without T2DM; the increased risk associated with T2DM was 24% for all-cause mortality (HR 1.24, 95% CI 1.12 to 1.38) and 10% for SMI relapses (OR 1.10, 95% CI 1.00 to 1.20).

The results also suggested that levels of physical health checks were associated with T2DM status. Compared with people without T2DM, people with T2DM received increased health checks by 103% for glucose (IRR 2.03, 95% CI 1.91 to 2.16), by 128% for cholesterol (IRR 2.28, 95% CI 2.20 to 2.38), by 62% for blood pressure (IRR 1.62%, 95% CI 1.57 to 1.68) and by 103% for BMI (IRR 2.03, 95% CI 1.93 to 2.15).

For full results of the core models, see Report Supplementary Material 5 and 6, and results of the extended models in Report Supplementary Material 7 and 8. Of the candidate explanatory variables, we found that older age was a significant predictor for poorer physical health outcomes and increased physical health checks, but was associated with a lower risk of SMI relapses. Female patients had reduced risk of poor physical health outcomes and increased risk of SMI relapse. They were also more likely to receive health checks (except for cholesterol tests) than males. People of Asian ethnicity had a reduced mortality rate and increased health checks. There was also a gradient in the impact of socioeconomic deprivation on health outcomes; people who lived in more deprived neighbourhoods had an elevated risk of macrovascular complications and mortality.

The core models also showed that physical comorbidities were generally associated with increased levels of physical health checks and poorer physical health outcomes. For instance, people with cardiovascular disease at baseline had an increased risk of both macro- and microvascular complications, and all-cause mortality. They also received more frequent health checks for diabetes and cardiovascular controls. Cardiovascular disease showed no impact on the risk of mental health outcomes. Dementia was associated with reduced risk of SMI relapse, as well as with receiving fewer physical health checks.

Of the medications we examined, antidepressants were associated with increased risk of macrovascular complications and a lower risk of SMI relapse. Typical antipsychotics were identified as a risk factor for SMI relapse and mortality. Furthermore, there was a general tendency towards physical health checks increasing with baseline medication uptake.

The results of the extended models were consistent with the results of the core models. Using additional information, we found that patients with a higher than normal BMI were more likely to receive physical health checks. Current smokers had a higher risk of macrovascular complications and SMI relapse. There was also evidence suggesting that poor diabetes and cardiovascular controls at baseline were not only risk factors of poorer health outcomes, but also predictors of more frequent physical health checks.

Summary of findings

A key finding is that, after adjustment, having diabetes increases the probability of receiving physical health checks and is a significant risk factor for both poorer physical and mental health outcomes.

The physical and mental health outcomes examined responded differently to variations in patient characteristics. For instance, advancing age was associated with increased risk of poorer physical health outcomes, but with reduced risk of mental health outcomes (SMI relapses); female sex was associated with a lower risk of diabetes complications and mortality, but a higher risk of SMI relapses. Type of SMI was a significant predictor for poorer mental health status, but had less influence on physical health outcomes, and antidepressant uptake reduced the risk of SMI relapses, but increased the risk of macrovascular complications.

The analyses also showed that receipt of the four physical health checks under investigation was affected by a wide range of factors, including older age, female sex, ethnicity, type of SMI, the presence of comorbidities and the use of various medications.

Objective 4: comparing the health outcomes of people with diabetes and severe mental illness with the health outcomes of people with diabetes without severe mental illness

Objective

The objective was to compare health-care interventions and physical and mental health outcomes in people with diabetes and SMI with those of people with diabetes without SMI.

Study population

There was a cohort of 12,999 patients in the CPRD data who had a clinical diagnosis of diabetes. This cohort included 3448 patients in data set B who had been identified as ‘cases’ for having both SMI and T2DM. A total of 9551 patients in data set C (see Figure 2) had been identified as matched ‘controls’ based on age, sex and general practice. For this objective, we used data sets B and C so that patients with both SMI and T2DM (cases) could be compared with non-SMI patients with T2DM (controls) of the same age band, sex and general practice.

Patients with T2DM were included if (1) they satisfied the four common inclusion criteria regarding diagnosis date and age, practice registration and type of diabetes; (2) they had a follow-up length of at least 1 day, in which the start of follow-up was identified as the later date of T2DM diagnosis or the start of UTS data plus 15 months, and the end of follow-up was defined as the earlier of the end of UTS data or the end of study period date; and (3) they were nested within a matched case–control cluster. A total of 9965 patients remained in the final sample for analyses.

Variables and statistical methods

The exposure variable was SMI status. The candidate explanatory variables are listed in Table 3, with detailed descriptions. The outcome variables under examination were as described in Objective 3: comparing health-care interventions and health outcomes in people with severe mental illness and diabetes with those in people with severe mental illness without diabetes, Variables and statistical methods.

The description and analysis methods for physical and mental health outcomes are provided in Table 4. The outcome, mental health status, was identified by the markers of depression and anxiety only, because we were unable to include SMI relapse owing to its perfect correlation with SMI status. The four physical health checks were chosen using the QOF indicators for people with diabetes. We therefore examined the level of health checks for HbA_1c, cholesterol, blood pressure and BMI.

The patient-level observations for this objective were nested within matched case–control clusters, an additional data hierarchy to the data used in previous objectives. We adopted statistical methods that examine variations within case–control clusters, ‘within’ estimators, to compare cases only with controls of the same age, sex and registered with the same practice. For further details on the methodology, see Report Supplementary Material 3.

Results

Descriptive statistics

The comparison of baseline characteristics between cases and controls, and by diagnosis order, are summarised in Table 14. In our sample of 9965 patients, 2192 (22%) cases were matched to 7773 (78%) controls. Just under 90% of cases had been matched to at least three controls. Cases had similar distribution in age and sex to controls. However, they differed for a range of other characteristics, and by diagnosis orders.

In general, people with SMI were more likely to come from the most deprived neighbourhoods than people without SMI. This socioeconomic disadvantage was further increased if SMI was diagnosed before T2DM. We observed a similar pattern in the baseline prevalence of dementia and learning disability, which were more common in people with SMI than in people without SMI; people with SMI diagnosed before T2DM were most likely to be affected.

Recorded physical health status, including cardiovascular disease, hypertension and Charlson Index diseases, showed opposite patterns. People with SMI were less likely to have a recorded diagnosis of these physical comorbidities than people without SMI; prevalence of these diseases was lower in people with SMI diagnosed before T2DM than in people who had T2DM first. As expected, larger proportions of people with SMI had been prescribed antidepressants and antipsychotics than people without SMI. The prescription rates of antihypertensive and lipid-lowering drugs were lower in people with SMI than in controls, a pattern consistent with the lower prevalence of recorded hypertension and cardiovascular disease in people with SMI.

Factors including BMI, smoking status and alcohol use were better recorded for people who had SMI before T2DM than for the rest of the sample. Although results should be interpreted with caution, we found that people with SMI had a higher BMI and were more likely to have substance misuse at baseline. Furthermore, people with SMI tended to have better diabetes and blood pressure controls, but a higher cholesterol level, at baseline than people without SMI.

The health outcomes of diabetes and cardiovascular control were compared between people with and people without SMI (see Figures 7 and 8), and were analysed separately for people diagnosed with SMI first and diagnosed with T2DM first. These analyses showed a common decreasing trend in the average levels of cholesterol and blood pressure during the study period. The mean values of HbA_1c remained stable and around the 7.5% (58 mmol/mol) threshold for both groups. People with SMI had lower HbA_1c and blood pressure levels than people without SMI, a pattern consistent with the comparison of baseline levels. The cholesterol levels of people with SMI, particularly those with SMI diagnosed before T2DM, remained above the cholesterol levels of controls.

FIGURE 7 Objective 4: average cholesterol and HbA1c, 2000/1–2015/16, for people with T2DM and SMI and people with T2DM without SMI. Average cholesterol (mmol/l) and 95% CIs for (a) SMI diagnosed before T2DM; (b) SMI diagnosed after T2DM (including same day); average HbA1c (%) and 95% CIs for (c) SMI diagnosed before T2DM; and (d) SMI diagnosed after T2DM (including same day). (Figure not reproduced here; see the original article.)

FIGURE 8 Objective 4: average blood pressure, 2000/1–2015/16, for people with T2DM and SMI and people with T2DM without SMI. Systolic blood pressure (mmHg) and 95% CIs for (a) SMI diagnosed before T2DM; (b) SMI diagnosed after T2DM (including same day); diastolic blood pressure (mmHg) and 95% CIs for (c) SMI diagnosed before T2DM; and (d) SMI diagnosed after T2DM (including same day). (Figure not reproduced here; see the original article.)

Descriptive statistics for the other outcome variables are provided in Table 15 and show that the pattern of diabetes complications differed between people with SMI and people without SMI, and that complications were also affected by diagnosis order. In general, the crude risks of both recorded macro- and microvascular complications were lower in people with SMI than in people without SMI, if SMI was diagnosed before T2DM. These crude risks were higher in people with SMI than in controls if T2DM was diagnosed first. Depression and anxiety were also more likely to be diagnosed in people with SMI than in people without SMI, and the crude mortality rate was higher in the SMI population. For the four physical health checks, we found that people with SMI tended to have more frequent checks than controls if SMI was diagnosed first, and less frequent checks if T2DM was diagnosed first.

Regression results for the impact of severe mental illness status on physical and mental health outcomes, mortality and health checks

The adjusted impact of SMI status on outcome variables based on our core models is summarised in Table 16. After adjustment, SMI status was not associated with macrovascular complications recorded in primary care data. However, we found a complex association with hospital admissions: having SMI was associated with an increased risk of emergency admissions for macrovascular complications (IRR 1.14, 95% CI 0.96 to 1.36) and with a reduced risk, by 36%, of elective admissions for macrovascular complications (IRR 0.64, 95% CI 0.47 to 0.88). Compared with controls, people with SMI were more likely to be diagnosed with depression and anxiety (OR 1.86%, 95% CI 1.63 to 2.12) and had an almost 90% increased risk of mortality (HR 1.89, 95% CI 1.59 to 2.26).

For physical health checks, people with SMI were more likely to receive tests for cholesterol (IRR 1.05, 95% CI 1.03 to 1.08), blood pressure (IRR 1.03, 95% CI 1.01 to 1.06) and BMI (IRR 1.09, 95% CI 1.06 to 1.11) than people without SMI. There was no difference between cases and controls in the chance of receiving a HbA_1c test.

As suggested by the descriptive statistics, people who had SMI diagnosed before T2DM differed in many aspects from people with T2DM diagnosed first. We therefore included interactions in the regressions to investigate whether or not the impact of SMI could be modified by this diagnosis order. The impact of SMI was not significantly modified by diagnosis order for hospital admissions for macrovascular complications or physical health checks. For macrovascular complications recorded in primary care data, however, the risk of this outcome was reduced for people with SMI if SMI was diagnosed before T2DM, and increased with the presence of SMI if T2DM was diagnosed first. The risk of depression and anxiety was higher in people with SMI than in people without SMI; this risk increased by around 41% for people diagnosed with SMI first and by 245% for people diagnosed with T2DM first (results available on request).

For full results of the core models, see Report Supplementary Materials 9 and 10. These suggest that older age is associated with higher rates of emergency admissions and physical health checks. Socioeconomic disadvantage is associated with higher risk of depression and mortality and reduced chances of receiving regular health checks. The presence of physical comorbidities, particularly cardiovascular disease and high blood pressure, was found to be a risk factor for poorer physical health outcomes. The number of Charlson Index diseases was associated with increased risks of mortality and depression. People who had dementia were less likely to be identified for physical health checks, whereas people on lipid-lowering medications were more likely to receive checks for cholesterol, HbA_1c and BMI. The results of the extended models confirmed these findings (see Report Supplementary Material 11 and 12). Furthermore, there is evidence suggesting that current smoker status and poor diabetes and blood pressure controls might be associated with poorer physical health outcomes. Obesity and abnormal diabetes and cardiovascular controls might increase levels of physical health checks.

Summary of findings

For this objective, we investigated the impact of SMI status on physical and mental health outcomes, as well as the utilisation of health-care interventions. We also explored whether or not health inequalities among people with SMI were associated with the order of diagnosis of SMI and diabetes.

People with SMI had tighter diabetes and blood pressure controls and better chances of receiving physical health checks than people without SMI. Despite this, for health outcomes, we found significant associations between SMI and increased risk for all-cause mortality and depression.

The results also suggested that elective admissions for macrovascular complications were significantly lower for people with SMI, whereas the rate of emergency admissions was elevated for this group (although the difference was not statistically significant). Further investigations indicated that most of these elective admissions had chronic ischaemic heart disease as the main diagnosis, and most emergency patients had been admitted for angina. These admission patterns might suggest a service gap; people with SMI are less likely to be referred by their GPs to specialist care for their cardiovascular disease and, when their symptoms deteriorate, they are more likely to be admitted as emergencies than people without SMI.

The analyses of the primary care data also supported this suggestion. Although the rate for recorded macrovascular complications was similar in people with and people without SMI, people with SMI were less likely to have a primary care record of chronic heart conditions, such as angina and chronic heart disease. This under-reporting/diagnosis pattern for people with SMI, which has been mentioned in the literature,124 could help explain the inequalities in specialist and hospital care experienced by people with SMI.

Objective 6: comparing diabetes care provision and estimating health-care costs for people with and people without severe mental illness

Objective

The objective was to provide a comparison of, and cost estimation for, diabetes health-care provision for people with and people without SMI.

Study population

We used the patient sample identified for objective 4 (see Objective 4: comparing the health outcomes of people with diabetes and severe mental illness with the health outcomes of people with diabetes without severe mental illness, Study population) to compare health-care use and associated costs between people with SMI and T2DM (cases) and their matched controls (non-SMI, T2DM). These patients were included for the analyses of this objective if they (1) had a length of follow-up of at least 1 year and (2) belonged to a case–control cluster. The start and end of follow-up were identified using the same method as for objective 4.

Variables and statistical methods

The exposure variable in this objective was the status of SMI. The candidate explanatory variables are summarised in Table 3.

The outcome variables were health-care use and the associated costs, including:

• primary care service use (consultations, prescriptions and diagnostic tests)
• hospital admissions for physical conditions (number of admissions and length of stay)
• hospital admissions for mental health problems
• associated costs for this service use.

The data construction and analysis methods are provided in Table 4. Mental health-related hospital admissions were identified using Healthcare Resource Group (HRG) codes.125 Primary care service use and hospital admissions were costed using a bottom-up approach. A&E attendance and the use of community mental health services were not included in the analyses, as linkages to these data sources were not available in the CPRD data set. For an overview of all sources of health-care utilisation data and unit costs (both primary and secondary care), see Report Supplementary Material 13. Health-care use and associated costs were aggregated on an annual basis.

For the consultation costs, we followed the approach adopted in Ride et al.126 and multiplied the duration of consultations by the costs of 1 minute of staff time, which were extracted from Unit Costs of Health and Social Care 2018.127 Multiple visits to the same staff on the same day were considered as duplicates, whereas visits to different staff in a day were counted as different visits. For details on the categorising of consultations, see Report Supplementary Material 14.

The costs of prescriptions were calculated as the number of prescriptions multiplied by unit costs in 2018, as extracted from the prescription cost analysis for 2018.128 Prescription records were aggregated at British National Formulary subparagraph level; higher hierarchy levels (paragraph, section or chapter) levels were used if subparagraph codes were unavailable.

For diagnostic tests, we followed Ride et al.126 and grouped the test records into the categories as set out in the NHS reference costs. Costs were then estimated using the levels of tests and the unit costs extracted from the NHS Reference Costs 2017/18.129 For details of this approach, see Report Supplementary Material 15.

The use and cost of hospital care were calculated for admissions to general hospitals only (including non-specialist mental health providers), but not admission to specialist mental health facilities. Hospital activities, such as diagnoses and procedures, were firstly grouped into HRGs using the costing grouper that corresponds to the NHS Reference Costs 2017/18. The generated HRG codes were then used to link the inpatient episodes to the national average costs from NHS Reference Costs 2017/18.129 The calculated costs were therefore expressed in 2018 prices. To compare people with SMI with people without SMI, we separated hospital admissions and associated costs by mental health- and physical health-related admissions using HRG codes.

We conducted generalised linear regression models to explore the impact of SMI status and other explanatory variables on health-care resource use and health-care costs, and to take into account the non-negative, highly skewed and leptokurtic characteristics of cost data. The choice of distributional family and link function were informed by the Park test130 and the Pregibon link test.131

Results

Descriptive statistics

A total of 6383 patients were included in the analysis sample with 1,023,257 primary care and 22,253 hospital admission records. Table 17 shows the distribution of baseline characteristics in the sample. The mean age of the sample was 57.9 years, 48.3% were male, the majority (82.5%) were white, the most common physical comorbidity was hypertension (55%), 26.6% were on antidepressants and 39.8% were current or ex-smokers. Cases had distribution in baseline characteristics that was similar to that of controls, such as age, sex, ethnicity and deprivation. However, people with SMI were more likely to be prescribed medications, such as antidepressants, antipsychotics and antidiabetics. They were also more likely to have lifestyles including smoking and substance use, although this finding is limited by the quality of data recording and number of missing values.

Primary care costs results

Table 18 shows annual resource use and costs for people with diabetes and SMI (cases) and people with diabetes alone (controls). The crude results demonstrate that, on average, people with diabetes and SMI used significantly more primary care services every year than people with diabetes alone. The main driver of this difference was the number of consultations (12.140 and 8.657 contacts per year for cases and controls, respectively); drug prescription and test-related contacts were similar between the two groups.

Hospital costs results

A similar trend was found for hospitalisation. People with diabetes and SMI had more hospital admissions every year and stayed in hospital longer than those with diabetes only (controls). This was regardless of whether or not the admissions were mental health related. As expected, people with SMI had more mental health-related admissions and stayed in hospital longer following those admissions. However, we also found that, for admissions related to physical health conditions, people with SMI and diabetes spent significantly more days in hospital per year (p < 0.0001) than people with diabetes alone. The difference in the number of admissions every year was small (0.679 and 0.556 admissions per year for cases and controls, respectively), but statistically significant (p = 0.018).

We also found that people with SMI incurred more costs than those without SMI. This difference was observed across both primary and secondary care. The main cost driver was secondary care and hospitalisation, which accounted for 83.3% and 72.5% of overall health-care expenditure for people with diabetes and SMI and for people with diabetes alone, respectively.

Regression analysis results for adjusted health-care resource utilisation and costs based on baseline characteristics

To take sample heterogeneity into account, we conducted a series of generalised linear regressions to adjust resource use and cost results by people’s baseline characteristics. These results can be found in the ‘adjusted’ section in Table 18. Baseline characteristics that were controlled for included age at diabetes diagnosis, sex, ethnicity, deprivation, comorbidities, medication use, time since diabetes diagnosis and financial year of diabetes diagnosis. Generalised linear regression with gamma distribution and log-link function was chosen, based on the Park and Pregibon tests, because of the non-normality of the data. After adjustment, the differences in resource use and costs between people with and people without SMI remained significant, except for differences in numbers of prescription-related and test-related consultations.

Figure 9 demonstrates the 10-year trend of annual health-care cost per person for people with diabetes and SMI and people with diabetes but no SMI. As shown, people with SMI incurred significantly higher annual health-care costs across this 10-year horizon. Cases and controls followed a similar trend: higher costs were incurred in the first year after diagnosis, followed by a cost reduction in the second year; thereafter, the annual costs slowly increased year on year. This trend was more prominent in people with SMI and diabetes than in those with diabetes alone.

FIGURE 9 Objective 6: trend for annual health-care cost per person with SMI and T2DM (cases) and matched person with T2DM but no SMI (controls). (Figure not reproduced here; see the original article.)

Table 19 shows the impact of SMI on total costs among people with diabetes. SMI increased the total costs by 82.6% (exp^0.602 – 1), increased the primary care costs by 70.9% and increased secondary care costs by 110.9%.

Summary of findings

People with SMI and diabetes had higher costs than people with diabetes alone. We found that the main cost driver was secondary care, which accounted for 83.3% of overall health-care costs for those with diabetes and SMI and for 72.5% of overall health-care costs for people with diabetes alone. On average, people with diabetes and SMI use more primary care services every year than people with diabetes alone. The same trend applied to hospitalisation: people with diabetes and SMI tended to have more hospital admissions every year and to stay in hospital longer than those with diabetes only. This difference was regardless of whether or not the admissions were mental health related.

Objective 7: identifying health-care interventions associated with better outcomes for people with diabetes and severe mental illness

Objective

The objective was to identify which health-care interventions may be associated with better health outcomes for people with SMI and diabetes.

Study population

For objective 4, we combined data sets B and C (see Figure 2) and used comparisons between people with SMI (cases) and people without SMI (controls) to estimate the association between SMI status and the provision of physical health checks. For objective 7, we conducted further investigations to explore the association between SMI and health checks in the presence of macrovascular complications.

The development of macrovascular complications as a physical health outcome was defined as the presence of myocardial infarction, peripheral vascular disease or stroke during patients’ follow-up periods. Although under constant review, the QOF has been incentivising primary care providers to perform physical health checks as ongoing management for patients on the register for these conditions. For instance, regular checks of blood pressure and cholesterol levels were included as QOF indicators for patients with coronary heart disease (CHD) and stroke before 2011/12. These indicators were also incentivised by the QOF for patients with CHD, stroke and peripheral arterial disease in 2012/13 and 2013/14. Since 2014/15, however, only checks on blood pressure remain incentivised for patients with CHD, stroke or peripheral arterial disease, and regular checks for cholesterol levels are no longer required.111 We hypothesised that physical health checks for blood pressure and cholesterol could influence people’s health outcomes, and would also be affected by the presence of macrovascular complications.

For this objective, we therefore applied restrictions on the patient sample used in objective 4 (see Objective 4: comparing the health outcomes of people with diabetes and severe mental illness with the health outcomes of people with diabetes without severe mental illness, Study population) and compared only cases and controls with the same status of macrovascular complications (as a health outcome), so that the impact of these complications on health checks was equal within case–control clusters. A total of 8724 patients remained in this restricted sample after controls with a macrovascular complication status different from that of cases were excluded.

Variables and statistical methods

The exposure and explanatory variables and statistical methods for this objective were primarily based on the core and extended models, as reported in objective 4 (see Objective 4: comparing the health outcomes of people with diabetes and severe mental illness with the health outcomes of people with diabetes without severe mental illness).

We explored the suitability of the CPRD data set for assessing individual interventions in terms of quality and consistency of recording by practices. Recording was not sufficiently comprehensive to allow detailed measurement of frequent and often complex interventions such as medication history and comprehensive care pathways. For referrals, we were not able to confirm attendance as we did not have linkage to outpatient data. However, we were able to ascertain whether or not lower frequency interventions such as health checks, diabetes education and influenza vaccination had been performed and we used these as the basis of our analysis.

For the impact of SMI on receipt of health checks, we re-estimated the core and extended models of blood pressure, cholesterol, HbA_1c and BMI, as reported in objective 4. An interaction term between the status of SMI and macrovascular complications was added to these models to explore whether or not the impact of SMI could be modified by the status of complications.

For the impact of health checks on health outcomes, we separately estimated the association between the level of each health check, measured as the mean number per year, and markers of depression and anxiety using the core model reported in objective 4 (see Objective 4: comparing the health outcomes of people with diabetes and severe mental illness with the health outcomes of people with diabetes without severe mental illness). We then added an interaction term between the level of health checks and SMI status to explore whether or not the association between health checks and depression could be modified by this.

We did not investigate the impact of health checks on physical health outcomes because of the confounding relationship between the two. We expected that the level of health checks would potentially be affected by the other physical health outcomes for reasons similar to why these health checks would be affected by macrovascular complications status. For instance, patients with poorer diabetes and cardiovascular control could potentially receive more frequent physical checks; death in the follow-up period would reduce health checks received, but deteriorated health status prior to death would increase physical health checks. Owing to this two-way causality, the estimated association between health checks and physical health outcomes would be biased using these patient-level data.

After initial investigation, we were unable to proceed with investigating the impact of retinopathy screening, diabetes education and influenza vaccination on health outcomes because of limitations in the overall data quality and the under-recording issue in primary care data for people with SMI. A descriptive summary of the provision of these health-care interventions can be found in Appendix 4, Table 32.

Results

Descriptive statistics

The baseline characteristics of the restricted sample are summarised in Table 20. Separating patients by macrovascular complication status, we found that patients without complications were diagnosed with SMI and T2DM at a younger age than patients with complications. The mean diagnosis age was 47.0 years for SMI and 56.2 years for T2DM for patients with SMI and without complications. Diagnosis ages were increased by > 10 years for patients with complications. In both groups, controls were closely matched to case patients in terms of age. There were more female patients in the group with complications. Socioeconomic deprivation was more common in people with SMI than in people without SMI.

Physical comorbidities, such as cardiovascular disease, hypertension and Charlson Index comorbidities, were more likely to be diagnosed in the group with macrovascular complications than in the group without complications, consistent with the pattern of prescription rates of antidiabetes, antihypertensive and lipid-lowering drugs. In these groups, patients with SMI had lower prevalence of physical comorbidities and lower prescription rates for antihypertensive and lipid-lowering drugs than people without SMI. This finding, however, may not suggest better baseline health status for people with SMI, but may reflect under-reporting or underdetection for this population (as previously discussed in Objective 4: comparing the health outcomes of people with diabetes and severe mental illness with the health outcomes of people with diabetes without severe mental illness).

Lifestyle and BMI were better recorded in people with SMI and without macrovascular complications. Although there were a high number of missing values, limiting interpretation, results suggest that the group without macrovascular complications had greater BMI at baseline.

Table 21 shows descriptive statistics for the outcome variables. Without adjustment, the number of health checks for HbA_1c and BMI were higher in the group without macrovascular complications and checks on blood pressure were higher in the group with complications. Within groups, people with SMI were more likely to receive checks for cholesterol level and BMI in the group without complications, and less likely to receive these checks in the group with complications, than people without SMI. For blood pressure and HbA_1c, the crude rates of health checks were lower in people with SMI than in people without SMI in both groups. These statistics also showed that people with SMI were more likely to have depression and anxiety recorded in their primary care records.

Regression analysis results for the adjusted impact of severe mental illness on levels of physical health checks

The adjusted impact of SMI on levels of physical health checks is reported in Table 22. These results suggest a significant association between SMI and health checks for blood pressure, cholesterol and BMI. These relationships differ by whether or not macrovascular complications have been diagnosed. Results from the core models show that people with SMI are more likely, by around 5%, to receive checks on blood pressure than people without SMI in the group without complications (IRR 1.05, 95% CI 1.03 to 1.08). This IRR is diminished by around 20% (IRR 0.80, 95% CI 0.70 to 0.91) in the group with complications; people with SMI are less likely, by around 16%, to receive this check [calculation of interaction between diagnosis of SMI and macrovascular complications (IRR 1.05 × 0.80)] than people without SMI in this group. A similar pattern was also found for the health checks for cholesterol and BMI, with HbA_1c showing no significant association with the status of either SMI or macrovascular complications. Results from the extended models confirm these findings. The estimated impact of the other explanatory variables is consistent with the results for objective 4.

The adjusted association between health checks and depression and anxiety is provided in Table 23. These results suggest that health checks for blood pressure and BMI are associated with increased recording of depression and anxiety after adjusting for SMI status and other risk factors. For instance, an additional check on blood pressure per year increases the probability of depression and anxiety being recorded by around 2% (OR 1.02, 95% CI 1.01 to 1.03); and one additional check on BMI increases this probability by around 1% (OR 1.01, 95% CI 1.00 to 1.02). The regressions with interaction terms show weak evidence that, in people with SMI, health checks for blood pressure have a bigger effect on the probability of depression and anxiety being classified than in people without SMI. An additional check per year increases this probability by around 1% for people without SMI (OR 1.01, 95% CI 1.01 to 1.02) and by around 6% [calculation of interaction effect (IRR 1.01 × 1.05)] for people with SMI. The impact of other health checks does not appear to be modified by the SMI status. The estimated impact of the other explanatory variables is consistent with the results for objective 4.

Summary of findings

We found that the association between SMI and physical health checks was influenced by the diagnosis of macrovascular complications. In the group without complications, people with SMI were more likely to receive health checks than people without SMI, whereas SMI was associated with reduced chances for receiving health checks in the group with complications.

The results also suggest that, in general, there is a positive association between the frequency of physical health checks and the likelihood that patients had depression and anxiety recorded. It is important to note that this finding is evidence of an association, rather than establishing causality; patients who saw their GPs more often would be more likely to have their depression and anxiety detected and recorded. This association, however, did not seem to differ between people with and without SMI.