Completion of annual diabetes care processes and mortality: A cohort study using the National Diabetes Audit for England and Wales

To conduct an analysis to assess whether the completion of recommended diabetes care processes (glycated haemoglobin [HbA1c], creatinine, cholesterol, blood pressure, body mass index [BMI], smoking habit, urinary albumin, retinal and foot examinations) at least annually is associated with mortality.


| INTRODUCTION
Optimal management of blood glucose, lipids and blood pressure reduces the microvascular and macrovascular complications of diabetes. [1][2][3] Accordingly, measurement and management of glycated haemoglobin (HbA1c), blood pressure and lipid profile are at the centre of national and international diabetes care guidelines. [4][5][6][7] Regular review of these and other risk factors for complications, including weight and smoking habit, are recommended, as are tests for early detection of kidney, foot and eye disease.
In England, the National Institute for Health and Care Excellence (NICE) recommends that people with type 1 diabetes 4 and type 2 diabetes 5 are offered nine annual processes (measurement of HbA1c, lipids, creatinine, albuminuria, blood pressure and body mass index [BMI], ascertainment of smoking status, and examination of the feet and retinae), and the completion of these has been incentivized in primary care. 8 Most international guidelines also stress the importance of these care processes. However, whilst their regular completion might seem intuitively sensible, the level of evidence to support the guideline-recommended processes, including their effect on clinical outcomes, is usually not known or is rated at the lowest standard of evidence ("expert consensus" or "clinical experience"). 7 In England and Wales, the National Diabetes Audit (NDA) collects patient-level data on people with diagnosed diabetes.
The present study assesses whether recorded care processes completion was associated with mortality over the subsequent decade after adjustment for the risk factors that the care processes uncover, individual demographic characteristics and comorbidities.

| Data sources
The NDA has collated data on people with diagnosed diabetes registered with a primary or specialist healthcare provider in England since 2003. Individuals receiving care from general practice and specialist outpatient services based in acute and community trusts are included if they have a valid code for diabetes mellitus (excluding gestational diabetes) in their electronic health record. 9 The 2009/2010 NDA data collection included data from 6700 (76%) general practices and was estimated to include data on 81.1% people aged 17 years and older with diagnosed diabetes in England and Wales. 10 These data were linked to Hospital Episode Statistics and the Patient Episode Database for Wales, which records all hospital admissions in England and Wales, respectively, and to civil death registrations in both countries collated by the Office for National Statistics.
The legal basis for the NDA data collection and linkage is a "direc-

| Study population and observation period
The study population was people aged between 17 and 99 years on January 1, 2009, diagnosed with type 1 diabetes and type 2 diabetes before January 1, 2009 who were included in the 2009/2010 NDA data collection and still alive on April 1, 2013. Analysis was restricted to individuals who survived 3 years after the exposure period to reduce potential bias from the clinically appropriate suspension of diabetes care processes for people in end-of-life care. Individuals were followed up from April 1, 2013 until death or December 31, 2019.

| Exposures
Data secondarily recorded in general practice systems for retinal examinations for this period are not considered reliable. The primary exposure was, therefore, the number out of a total of eight care processes (blood tests for HbA1c, cholesterol, creatinine, measurement of blood pressure, BMI, albuminuria, smoking habit assessment and the examination of feet) recorded as undertaken between January 1, 2009 and March 31, 2010. As initial exploratory analysis identified that only a minority of people had five or fewer care processes recorded and that people receiving six or seven care processes had similar characteristics and outcomes, these categories were used in the analysis. People who had all eight care processes recorded formed the primary reference group to reflect current national guidelines.
Age and duration of diagnosed diabetes at baseline were calculated using date of birth and date of diagnosis, respectively. Ethnicity was based on self-reported ethnic group as recorded by healthcare providers and classified as White, Mixed, South Asian, Black, other or missing. Type of diabetes was attributed based on the most recent type recorded by a healthcare provider and notified to the NDA. Data from a specialist healthcare provider were assigned precedence over the type of diabetes in the primary care health record. The latest reported risk factor measurements in the period January 1, 2009 to March 31, 2010 for HbA1c, systolic blood pressure, total cholesterol, creatinine, BMI and smoking habit were identified. Estimated glomerular filtration rate (eGFR) was calculated using the Modification of Diet in Renal Disease formula. 12 Hospital admissions for myocardial infarction (ICD-10 codes   I21-22), stroke (ICD-10 codes I61, I63-64, I67.9), heart failure (ICD-10 codes I50), respiratory disease (ICD-10 codes J01-99) and cancer (ICD-10 codes C01-99) between January 1, 2004 and December 31, 2008 were identified.

| Statistical methods
The differences in mean age, duration of diagnosed diabetes, HbA1c and BMI by the number of care processes recorded as undertaken were tested using analysis of variance (ANOVA), with Leveneʼs test to identify differences in variance. Differences in the proportion of people recorded as receiving care processes for categorical variables (sex, social deprivation, ethnicity, smoking habit) were tested using the chi-squared statistic.
Crude mortality rates and mortality rates per 1000 person-years, standardized for age and sex to the European Standard population, were calculated with 95% confidence intervals (CIs) using Byarʼs method. 13 Cox proportional hazard models were created to assess the associations between the number of recorded care processes and mortality for people with type 1 and people with type 2 diabetes. A series of models was created consisting of sequentially more covariates to examine potential confounding factors.
Separate models, adjusting for all risk factors, were created for mortality from cardiovascular disease, cancer, respiratory disease, diabetes-specific causes and renal failure for type 1 diabetes and type 2 diabetes separately.
Models adjusted for all risk factors and stratified by sex, age (less than 65 years old and 65 years and older), ethnic group, quintile of deprivation and whether or not the individual had an acute hospital admission in the year prior to the exposure period were constructed for all-cause mortality in people with type 1 diabetes and in people with type 2 diabetes.
Two models (one for type 1 diabetes and one for type 2 diabetes) adjusted for age, sex, ethnic group, deprivation and whether or not each of the eight care processes had been completed were created to identify if the association with all-cause mortality varied by type of care process. All variables were defined as categorical variables and included a category for missing data. A sensitivity analysis was undertaken in which everyone included in the 2009/2010 NDA and still alive on January 1, 2011 to explore whether the survival bias introduced by excluding deaths shortly after the exposure period altered the findings.
Statistical analysis was undertaken in SAS Enterprise Guide 7.1.

| RESULTS
A total of 179 105 people with type 1 diabetes and 1 397 790 with type

| Characteristics by number of care processes received
Care process completion variation showed little relation to deprivation but was associated with age, ethnicity, HbA1c and smoking status ( A breakdown of the individual care processes received is provided in Tables S1 and S2.  (Table 3).

| Mortality by number of care processes received
After adjustment for all covariates, the gradient of the inverse association of mortality in people with type 2 diabetes with number of recorded care processes was lower for cancer deaths ( for men; Figure 1B). The HRs for death associated with different numbers of recorded care processes were similar in people aged under or over 65 years in both type 1 diabetes and type 2 diabetes ( Figure 1A,B).
In people with type 2 diabetes the HRs for death associated with the number of recorded care processes were similar in White and Black ethnic groups but significantly lower in South Asian ethnic groups ( Figure 1). In people with type 1 diabetes, the CIs were much broader and no differences between ethnic groups were identified. In both type 1 diabetes and type 2 diabetes the HRs associated with numbers of recorded care processes were similar across all deprivation quintiles (Table S3). For people who had one or more acute hospital admission in the year prior to the exposure period the all-cause mortality HR associated with receiving fewer than five care processes was lower than for those who did not have an acute hospital admis-

| Individual care processes
Associations adjusted for age, sex, ethnicity and deprivation were investigated according to individual care process (

| DISCUSSION
This large national population-based cohort of people with type 1 diabetes and type 2 diabetes followed up for means of 7.6 and 6.9 years, respectively, following 15 months of routine care, finds that having five or fewer recorded care processes during that baseline period was associated with subsequent 7-year hazards of all-cause mortality approximately one-third higher compared to having all eight care processes after accounting for demographic characteristics. This higher mortality persists after adjustment for clinical factors known to affect the risk of diabetes-related complications (HbA1c, systolic blood pressure, serum cholesterol, BMI, smoking habit), and cardiovascular and renal comorbidities were taken into account.
The associations were similar between people with type 1 diabetes and type 2 diabetes, at all ages and across socioeconomic groups. In England and Wales most people with type 1 diabetes have specialist-led care while, for type 2 diabetes, most people are managed in a primary care setting. 14 Accordingly, the association between the number of recorded care processes and mortality was independent of the type of care setting. During periods of acute illness or palliative care the medium-to long-term management of diabetes-associated risk may not have clinical priority.
T A B L E 3 Hazard ratios for mortality associated with the number of care processes recorded between January 1, 2009 and March 31, 2010 for people with type 1 diabetes and type 2 diabetes, all-cause mortality with different adjustments and cause-specific mortality  19 Those that did mostly found associations between infrequent attendance and higher levels of glucose, BMI, blood pressure and lipids, a few studies documented higher emergency hospital use and diabetes-related complications, and just one study, using a composite measure of nonattendance and treatment noncompliance, found higher mortality in people with type 1 diabetes. 18,19 As compared to the collective results, analysis of the associations between mortality and noncompletion of individual care processes The present analysis identifies an association between low numbers of annual care processes completed and subsequent 7-year mortality. Therefore, it identifies a group of people who have a higher risk of mortality. But observational analyses cannot establish cause and effect and we cannot exclude residual confounding. One can only speculate on what any mechanism might be. The prominence of respiratory disease among those who died after low rates of care process completion raises one possibility. Respiratory deaths in younger people are predominantly due to pneumonia, for which diabetes is a known risk factor. 20 In our analysis, we tried to account for known pneumonia risks such as smoking, which was more common in the low care process group, and elevated BMI, but we were not able to include other known factors such as high alcohol intake, poor diet and low physical activity. Conceivably, these unmeasured risks triangulate with the likelihood of missing care processes. Alternatively, individuals more engaged with self-care and lower risk lifestyles may attend clinics more often and be keener to complete all the care processes.
Equally, the findings may be due to reverse causality, whereby people with multimorbidities, particularly mental illness, will be less likely to engage with routine follow-up and self-management.
Strengths of the present study include the size of the cohort included in the analysis, covering 76% of practices in England and Wales, the fact that it is drawn from a comprehensive selection of real-world population-based healthcare records, and the length of the follow-up. An important limitation is that neither medication data nor influenza and pneumonia immunizations were available for this analysis, and these could have shed some light on healthcare interactions.
The nature of this analysis means that if people have not received a specific care process the risk factor data arising from that process are missing. In this analysis all variables included in the Cox proportional hazard regression models are treated as categorical variables and have a category for "missing" data. Whilst this does not completely eradicate residual confounding due to missing data, it is much reduced. It is