The UK Biobank is a large population-based prospective cohort study, which incorporated data from more than 500,000 participants (aged 37–73 years) across the U.K. between March 2006 and October 2010. The design of the UK Biobank study has been presented elsewhere [30, 31]. Data on socio-demographics, lifestyle, environmental factors, and medical history were collected via touch-screen questionnaires during interview at baseline visit. Standardized physical measurements (e.g. blood pressure and anthropometrics) and biological sample (blood, urine, and saliva) were taken among all the participants after interview.
A total of 19,229 participants with pre-existing T2D (mean age 59.5 ± 7.0, 59.5% men) were included in the current analysis after excluding patients who had pre-existing CAD, MI, HF, and stroke (ischemic and haemorrhagic). The flowchart for the selection of the study population is presented in Supplementary Fig. 1. The prevalent cases of T2D were identified through using the algorithms method  or via electronic health records using the ICD-10 codes (E11).
This study was conducted under the UK Biobank Application Number 68307. The UK Biobank study was approved by the National Information Governance Board for Health and Social Care in England and Wales, and the Community Health Index Advisory Group in Scotland and the North West Multicentre Research Ethics Committee. All participants gave written informed consent.
Assessment of the covariates
Data on age, sex, ethnicity, education levels, smoking history and sleep hours per day were collected through interview at baseline. Townsend deprivation index (TDI) is a composite measure of socio-economic deprivation . Body mass index (BMI, kg/m2) was calculated by body weight in kilogram divided by square of height in meter using data that were examined during a nurse-led interview. Information on habitual diet and alcohol intake was captured by a touchscreen food frequency questionnaire. We generated a healthy diet score to reflect the overall diet based on five components including vegetables, fruits, whole grains, low-fat dairy, and red/processed meat intake. Participants in the highest quintiles of favourable foods (vegetables, fruits, whole grains, and low-fat dairy) intake received 5 points and those in the lowest quintile of red/processed meat were given 5 points. The overall diet score then was categorized into quintiles. Physical activity was assessed using a short form international physical activity questionnaire, and physically active was defined as ≥ 150 min/week moderate or ≥ 75 min/week vigorous or 150 min/week moderate/vigorous activities .
Prevalent hypertension cases were defined by self-report, use of anti-hypertensive medications, essential hypertension cases via linking the electronic health records, or a seated blood pressure ≥ 140/90 mm Hg. Pre-existing cancer was self-report. Comorbidities including GERD, gastric ulcer, duodenal ulcer, peptic ulcer, and gastrointestinal ulcer were identified through self-report and electronic health records. Medication history (e.g. aspirin, clopidogrel, vitamin supplements, and anti-hypertensive, cholesterol-lowering, and anti-diabetic drugs) was self-report. Participants were also asked to provide the medicines in the following visits, if they were not certain about the types of the medications taken. Serum concentrations of glycated hemoglobinA1c (HbA1c) were measured by HPLC analysis on a Bio-Rad VARIANT II Turbo.
The differences in baseline characteristics by PPIs users and non-users were examined using Student’s t-test for continuous variables and chi-squared test for categorical variables. Missing values for covariates were imputed using sex-specific mean values for continuous variables and missing indicator approach for categorical variables. We used multivariable-adjusted Cox proportional hazards regression models to compute the hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations of PPIs use with risks of outcomes of interest.
Three models were fitted. In Model 1, we adjusted for age at recruitment (years) and sex (men, women). In Model 2, we further adjusted for education (college or university degree, other professional qualifications, A/AS levels or equivalent or O levels/General Certificate of Secondary Education [GCSE] or equivalent, none of the above), socio-economic status (TDI, continuous), ethnicity (White, others), BMI (kg/m2, continuous), alcohol intake (never or special occasions, monthly to weekly, daily), smoking status (never, past, current), healthy diet score (in quintiles), sleep duration (≤ 6, 7–8, ≥ 9 hours/day), physical activity status (yes, no), family history of CVD (yes, no), prevalent hypertension (yes, no), prevalent cancer (yes, no), duration of diabetes (years continuous), HbA1c (mmol/mol, continuous), anti-diabetic medications (none, oral drugs, insulin and others), anti-hypertensive medications (yes, no), cholesterol-lowering medications (yes, no), aspirins use (yes, no), and clopidogrel use (yes, no). In Model 3, indications for PPIs use (GERD, gastric ulcer, duodenal ulcer, peptic ulcer, or gastrointestinal ulcer) were additionally adjusted.
We also stratified the analyses by age (≤ 65, > 65 years), sex (men, women), duration of diabetes (≤ 5, > 5 years), smoking status (never, ever), family history of CVD (yes, no), medications for diabetes (none, oral drugs, insulin and others), antiplatelet drugs (aspirins/clopidogrel; yes, no), and indications of PPIs (yes, no). The multiplicative interactions between PPIs use and the stratified factors on the risk of outcomes were tested using the likelihood ratio test by including an interaction term in Model 3. In addition, we assessed the associations between different type of PPIs (omeprazole, lansoprazole, esomeprazole, and other PPIs) and risks of outcomes to clarify whether the observed associations were agent-specific or class-specific.
To explore the robustness of our primary findings, we performed a number of sensitivity analyses. First, to minimize the residual confounding, we assessed the associations in a propensity score-matched cohort of PPIs users (n = 3275) and non-users (n = 3275). Propensity scores were calculated using a logistic regression model including age, sex, TDI, education, ethnicity, BMI, smoking, drinking, physical activity, sleep duration, healthy diet score, family history of CVD, history of hypertension, history of cancer, HbA1c, duration of T2D, aspirins use, clopidogrel use, medications for hypertension, cholesterol and diabetes as covariates. The PPI users and non-users were 1:1 matched using nearest neighbour method without replacement (caliper = 0.1). Second, we performed a two-year lag year analysis to minimize the possibility of reverse causality on the observed associations. Third, we repeated the main analyses using multiple imputation method for covariates by chained equations with 5 imputations. Fourth, to further account for the potential confounding effect of indications of PPIs, we additionally adjusted for use of H2 receptor antagonists in Model 3. Fifth, we investigated the associations of PPIs use with risks of ischemic and haemorrhagic stroke. Further, to increase the statistical power, we combined stroke and transient ischemic attack (TIA) as a composite outcome, and tested the association between PPIs use and risk of stroke/TIA.
All analyses were performed using Stata statistical software, release 15.1 (StataCorp LP, College Station, Texas), and a two-sided p < 0.05 was set as the threshold for statistical significance.