Data sources
The UK CPRD GOLD database (8) is a longitudinal national primary care database that includes EHR data of over 17 million individuals, currently or previously registered with over 700 general practitioner (GP) practices in the UK. CPRD has also been linked to secondary care data (Hospital Episode Statistics, HES), mortality data from the Office for National Statistics (ONS) and area-based data on measures of social deprivation. The CPRD population has the same profile regarding age, sex and ethnicity as the general population of the UK (8).
Study population
Participants were included if they were aged 50 years or over at any point during their CPRD registration period between 1987 and 2018 and had a diagnosis of diabetes, based on relevant CPRD Medcode or a prescription of anti-diabetes drugs (oral hypoglycaemic agents or insulin) (6). In addition, eligible patients should have been registered in CPRD for at least one year before diabetes onset to allow time for baseline information to be recorded and to ensure that the date of newly diagnosed diabetes was captured. Patients with a diagnosis of type 1 diabetes, or those who had a diagnosis of diabetes or initiation of anti-diabetic treatment before the age of 30 were excluded. Patients who had a diagnosis of dementia before cohort entry were also excluded. A total of 489,205 individuals were included in the analysis.
Exposure assessment
Records of six cardiovascular comorbidities (coronary heart disease, stroke, atrial fibrillation, heart failure, peripheral vascular disease and hypertension) and three major non-cardiovascular comorbidities (chronic kidney disease, chronic obstructive pulmonary disease [COPD] and cancer) were extracted using the corresponding CPRD Medcode and Enttype code. These comorbidities were selected as they are common chronic diseases and leading causes of death and disability in older adults (9). To comprehensively identify hypertensive cases, we additionally used blood pressure recordings (systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg). The date of onset of a specific type of comorbidity was defined according to its first relevant health record. The date of onset of overall cardiovascular comorbidities was defined as the earliest date of developing one of the six comorbidities mentioned above.
In addition, information on the following covariates at cohort entry was extracted: age, sex, calendar year of cohort entry, region in the UK, quintiles of index of multiple deprivation (IMD, a proxy of socio-economic status), body mass index (BMI), smoking status, duration of diabetes (based on the first clinical record of diabetes diagnosis), history of prescription of anti-diabetes medications and recorded diabetic complications.
Outcome ascertainment
The outcome event was dementia incidence. Patients were considered to have dementia if they had: 1) a dementia diagnosis based on Medcode in CPRD; 2) a dementia diagnosis based on ICD codes in linked HES or ONS records; or 3) at least one dementia-specific drug prescription (donepezil, galantamine, rivastigmine or memantine) (6). Among the extracted dementia cases, 96% were based on diagnosis codes and 4% were based on dementia-specific drug record. The outcome event date was defined as the date of the first dementia diagnosis or the first prescription date of dementia-specific drugs, whichever occurred earlier.
Statistical analyses
Distributions of baseline characteristics were summarised and compared between patients with and without cardiovascular disease (except hypertension) at baseline. Time-varying Cox proportional hazards models, with age as the time-scale, were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) of dementia associated with comorbidities. The presence of six cardiovascular comorbidities in aggregate and by type, as well as three non-cardiovascular comorbidities, were treated as time-varying variables and examined in separate Cox models. Patients with newly developed comorbidities during follow-up contributed person-years to the no comorbidity group up until the comorbidity diagnosis date and then contributed person-years to the comorbidity group. For each patient, time of cohort entry was defined as the date of diabetes onset, aged 50 or January 1, 1987, whichever was the latest. The end of follow-up was defined as the date of dementia incidence, death, transfer out date, last data collection date of the GP practice or May 1, 2018, whichever occurred first.
To account for potential confounding biases, three sequential models with increasing levels of covariate adjustment were created for all analyses: Model 1 only adjusted for age, sex, calendar year and region; Model 2 additionally adjusted for IMD (quintiles), smoking status (non-smoker, current smoker, ex-smoker or missing), BMI category (<25, 25 to <30, ≥30 kg/m2 or missing), and history of comorbidities (for mutual adjustment); Model 3 additionally adjusted for diabetes-related factors, including the duration of diabetes, HbA1c level, presence of diabetic complications (including hypoglycaemia) and prescription of anti-diabetes drugs (no drug, only oral hypoglycaemic drug, or insulin). Covariates that could change over time (e.g., HbA1c level and status of diabetic complications) were also modelled as time-varying variables and updated at comorbidity diagnosis date during follow-up. Missing values of BMI category and smoking status during follow-up were imputed with the last observation carried forward.
The statistical analyses were conducted using Stata (version 15, Stata). All statistical tests were two-sided, and the significance level was P < 0.05.