Study design and population
The UK Biobank is a population-based cohort study that recruited more than 500,000 participants who attended one of 22 assessment centers across the United Kingdom between 2006 and 2010, and were followed up to 2020 [
15]. In this cohort study, among 502,507 participants, a total of 469,271 were excluded, including 462,388 without type 2 diabetes, 698 with prevalent diabetic complications, 6129 with missing data on smoking status, alcohol consumption, physical activity, diet, body mass index (BMI), blood cholesterol, and blood pressure factors at baseline, and 56 lost to follow-up evaluations. Finally, 33,236 participants were available for the current study (Additional file
1: Fig. S1).
Data collection
Information on sex, age, ethnicity, education, socioeconomic status, employment status, diabetes duration, and family history of diabetes was collected through a touchscreen questionnaire and interview. Ethnicity was categorized as White, Black or Black British, Asian or Asian British, and other. Education was defined as college or university, upper secondary, lower secondary, vocational, and other. Socioeconomic status was defined based on the Townsend deprivation index [
16] (encompassing information on social class, employment, car availability, housing) and categorized as low (highest quintile), middle (quintiles 2 to 4), and high (lowest quintile) [
17]. Employment status was categorized as working, unemployed, retired, and other. Alcohol consumption was defined based on self-reported intake of red wine, white wine, beer, spirits, and fortified wine and categorized as never/moderate (0 to 14 g/d alcohol for women and 0 to 28 g/d alcohol for men) and non-moderate (> 14 g/d alcohol for women and > 28 g/d alcohol for men) [
18]. Information on sugar sweetened beverages was assessed using a touchscreen questionnaire. Participants who reported never consumes drinks containing sugar was defined as not consuming sugar sweetened beverages [
19]. Information on triglyceride, serum creatinine, CRP levels, white blood cell (leukocyte) counts, and platelet counts were obtained from blood samples collected at study recruitment. The CRP levels was categorized into two groups (low: ≤3 mg/L; high: >3 mg/L) [
20]. Leukocyte counts was coded as low (≤ 10 × 109/L) and high (> 10 × 109/L). Likewise, platelet counts was divided into two groups (low: ≤450 × 109/L; high: >450 × 109/L).
The UK Biobank has full ethical approval from the North West Multi-Center Research Ethics Committee, the National Information Governance Board for Health and Social Care in England and Wales, and the Community Health Index Advisory Group in Scotland. All the UK Biobank participants gave written informed consent before data collection (
http://www.ukbiobank.ac.uk/ethics/).
Diagnosis of type 2 diabetes and its complications
Type 2 diabetes was ascertained based on self-report, medical records, glycated hemoglobin ≥ 6.5%, hospital inpatient registers (International Classification of Diseases [ICD-10] code E11), or use of hypoglycemic agents.
Diabetes complications were assessed using the ICD codes derived from the hospital inpatient registers. The major diabetes complications included renal complications (ICD-10 code E11.2), ophthalmic complications (ICD-10 code E11.3), neurological complications (ICD-10: E11.4), and peripheral circulatory complications (ICD-10 code E11.5).
Assessment of CVH metrics
We defined 6 CVH metrics following the AHA’s recommendations, including smoking status, physical activity, diet, BMI, blood pressure, and cholesterol [
5].
Information on smoking status and physical activity was obtained from the Touchscreen questionnaire, diet was derived from the Food Frequency questionnaire, BMI and blood pressure were obtained from physical measurement, and cholesterol concentration was collected from blood sample measured at recruitment. Smoking status was dichotomized as non-smoker vs. being a smoker. Physical activity was measured as the sum of minutes performing walking, moderate and vigorous activity using the International Physical Activity Questionnaire. Regular physical activity was defined as participants who had moderate activity ≥ 150 min per week or vigorous activity ≥ 75 min per week or ≥ 150 min/week of moderate and vigorous activity [
5]. A healthy diet score was generated based on the seven commonly eaten food groups (fruits, vegetables, whole grains, refined grains, fish, unprocessed meat, and processed meat) following recommendations on dietary priorities for cardiometabolic health [
21]. A healthy diet was based on adequate intake of at least four of seven commonly eaten food groups (1. Fruits: ≥ 3 servings/day; 2. Vegetables: ≥ 3 servings/day; 3. Fish: ≥2 servings/week; 4. Processed meats: ≤ 1 serving/week; 5. Unprocessed red meats: ≤1.5 servings/week; 6. Whole grains: ≥ 3servings/day; 7. Refined grains: ≤1.5 servings/day) [
18]. BMI was calculated as weight (kg) divided by height squared (m
2), and was categorized as non-overweight (< 25 kg/m
2) and overweight (≥ 25 kg/m
2). The mean of two measurements was used, and elevated blood pressure was defined as systolic blood pressure (SBP) ≥ 120 mmHg or diastolic blood pressure (DBP) ≥ 80 mmHg, or use of anti-hypertension agents. Ideal cholesterol was defined as untreated total cholesterol < 200 mg/dL. Table
1 in the Additional file
1 provides additional details regarding the assessment of favorable CVH metrics.
Table 1
Baseline characteristics of participants by cardiovascular health metrics
n (%) | 16,585 (48.87) | 15,549 (45.82) | 1800 (5.30) | |
Age, mean (SD), year | 59.52 (7.13) | 59.3 (7.34) | 57.84 (8.04) | < 0.001 |
Male, n (%) | 11,129 (67.10) | 8034 (51.67) | 468 (26.00) | < 0.001 |
Education level, n (%) | | | | |
College or University | 3562 (21.48) | 4260 (27.4) | 756 (42) | < 0.001 |
Upper secondary | 1497 (9.03) | 1542 (9.92) | 197 (10.94) | |
Lower secondary | 4137 (24.94) | 3922 (25.22) | 419 (23.28) | |
Vocational | 1482 (8.94) | 1252 (8.05) | 82 (4.56) | |
Other | 5907 (35.62) | 4573 (29.41) | 346 (19.22) | |
Socioeconomic status, n (%)a | | | | < 0.001 |
High | 2351 (14.18) | 2673 (17.19) | 413 (22.94) | |
Middle | 8956 (54) | 8969 (57.68) | 1083 (60.17) | |
Low | 5278 (31.82) | 3907 (25.13) | 304 (16.89) | |
Ethnicity, n (%) | | | | < 0.001 |
White | 14,992 (90.39) | 13,424 (86.33) | 1601 (88.94) | |
Asian | 806 (4.86) | 1027 (6.6) | 98 (5.44) | |
Black | 420 (2.53) | 623 (4.01) | 37 (2.06) | |
Other | 367 (2.21) | 475 (3.05) | 64 (3.56) | |
Current employment status, n (%) | | | | 0.249 |
Worked | 9392 (56.63) | 8889 (57.17) | 1021 (56.72) | |
Retired | 5577 (33.63) | 5121 (32.93) | 591 (32.83) | |
Unemployed | 280 (1.69) | 274 (1.76) | 45 (2.5) | |
Other | 1336 (8.06) | 1265 (8.14) | 143 (7.94) | |
Alcohol consumption, n (%) | | | | < 0.001 |
Excessive | 8416 (50.81) | 7303 (47.03) | 788 (43.78) | |
Never/moderate | 8147 (49.19) | 8225 (52.97) | 1012 (56.22) | |
Smoking status, n (%) | | | | < 0.001 |
Smoker | 11,639 (70.48) | 3978 (25.63) | 190 (10.56) | |
Non–smoker | 4874 (29.52) | 11,542 (74.37) | 1610 (89.44) | |
Physical activity, n (%) | | | | |
Inactive | 11,293 (77.89) | 4759 (32.57) | 250 (14.2) | < 0.001 |
Active | 3205 (22.11) | 9853 (67.43) | 1511 (85.8) | |
Diet, n (%) | | | | < 0.001 |
Unhealthy | 13,471 (82.22) | 5678 (36.74) | 262 (14.58) | |
Healthy | 2914 (17.78) | 9778 (63.26) | 1535 (85.42) | |
BMI (kg/m2) | | | | < 0.001 |
≥ 25 (overweight) | 15,742 (97.75) | 12,537 (82.12) | 363 (20.34) | |
< 25 (non–overweight) | 362 (2.25) | 2729 (17.88) | 1422 (79.66) | |
Sugar–sweetened beverages, n (%)b | 906 (5.46) | 843 (5.42) | 77 (4.28) | < 0.001 |
Family history of diabetes, n (%) | 6906 (41.64) | 6537 (42.04) | 581 (32.28) | < 0.001 |
DBP, median (IQR), mm Hg | 83 (76–90) | 82 (75–89) | 76 (69–84) | < 0.001 |
DBP, median (IQR), mm Hg | 143 (132–156) | 143 (131–156) | 133 (117–151) | < 0.001 |
Cholesterol, median (IQR), mmol/L | 4.6 (3.92–5.54) | 4.8 (4.04–5.69) | 4.54 (3.65–5.29) | < 0.001 |
Triglyceride, median (IQR), mmol/L | 2.00 (1.49–2.86) | 1.75 (1.25–2.49) | 1.19 (0.84–1.71) | < 0.001 |
C–reactive protein, median (IQR), mg/L | 2.47 (1.17–4.55) | 1.88 (0.88–3.47) | 0.90 (0.46–1.89) | < 0.001 |
Serum creatinine, median (IQR), umol/L | 72.3 (63.4–83.8) | 71.2 (61.1–80.9) | 65.2 (57.6–73.8) | < 0.001 |
In the current study, favorable CVH metrics included non-smoker, regular physical activity, a healthy diet, non-overweight, untreated resting blood pressure < 120/<80 mm Hg, and untreated total cholesterol < 200 mg/dL. Participants were categorized into three groups according to the number of healthy CVH metrics: (1) unfavorable: participants who had no or only one healthy CVH metric; (2) intermediate: those who had any two healthy CVH metrics; and (3) favorable: those who had 4 or more healthy CVH metrics.
Statistical analyses
Baseline characteristics of the samples were summarized across CVH categories as percentages for categorical variables and means and standard deviations (SDs) for continuous variables with a normally distribution. If continuous variables did not follow a normal distribution, the variables were summarized as median and interquartile range (IQR). A Shapiro–Wilk normality test was used to assess the normality of the distribution. The ANOVA was used to compare the means of continuous variables and normally distributed data; otherwise, the Mann–Whitney U–test was applied. Categorical data were assessed by chi-square test. Person-years were calculated from the date of recruitment to the date of the death or censoring date, whichever event occurred first. Incidence rates (IRs) per 1000 person-years were calculated for each CVH category.
We used multivariate Cox proportional-hazard models to estimate the hazard ratio (HR) and 95% confidence interval (CI). The proportional hazards assumptions for the Cox model were tested using Schoenfeld residuals method, no violation of the assumption was observed. The duration of follow-up was calculated as a timescale between the baseline assessment and the first event of diabetic complications, death, or last study visit (November 31, 2020). To quantify the contribution of CVH metrics to diabetic complications and all-cause mortality incidence, we calculated the population attributable fraction (PAF), which is the estimated proportional reduction in diabetic complications and all-cause mortality that would occur if CVH metrics were favorable. We estimated the PAF for favorable CVH metrics by the punafcc function in STATA, which was specifically developed for time to event studies.
We further calculated the years of life expectancy lost due to unfavorable CVH. The calculation of years of life lost (i.e. difference in life expectancy) involved a two-step process using flexible parametric survival models with age as the time scale [
22]. First, residual life expectancy was estimated as the area under the survival curve up to 100 years old, conditional on surviving at ages 40 to 100 years old (one-year intervals). Second, years of life lost were calculated as the difference between the areas under two survival curves [
23‐
25]. To calculate life expectancy, proportional hazard survival analyses were conducted with the stpm2 command which uses restricted cubic splines to model the baseline cumulative hazard.
To estimate and quantify the mediation effect of inflammation on the association between CVH at baseline and diabetic complications and all-cause mortality, causal mediation analyses was conducted. Causal mediation analyses using Cox models measure the effect of the exposure on the outcome (total effect, TE) decomposed into natural direct effect (NDE) and natural indirect effect (NIE) [
26]. We performed analysis following the previous procedure [
27]: (1) simulated multivariable logistic model for the association between CVH and inflammation marker. (2) performed full adjusted Cox models to measure the effect inflammation marker and CVH on the development of diabetic complications and all-cause mortality; (3) calculate the hazard ratios with the coefficient of previous models. Variances and 95% CI calculated using a resampling method that takes random draws from the multivariate normal distribution of estimates [
28]. C-reactive protein (CRP) is an exquisitely sensitive systemic marker of inflammation [
13,
14]. In the present study, CPR was used as one of the main inflammatory markers to perform mediation analyses. Additionally, considering opportunistic and arbitrary, we also examined other inflammatory markers, such as leukocyte counts and platelet counts to conduct the mediation analysis. The Cox proportional-hazard model, proportional hazard survival analysis, and mediation analysis were adjusted for sex, age, education level, socioeconomic status, ethnic background, alcohol consumption, sugar-sweetened beverages, family history of diabetes, triglyceride, serum creatinine, and CRP. If missing values on covariates, we used multiple imputations based on five replications and a chained-equation method to impute data [
29]. Detailed information on missing data was shown in Additional file
1: Table S2.
Several additional analyses were performed to assess the robustness of our study results. First, we used stratification analysis to examine whether the association between CVH and diabetic complications and all-cause mortality varied by sex, age, socioeconomic status, alcohol consumption, or family history of diabetes. Next, we constructed a weighted CVH score based on the six favorable CVH metrics using the following equation: weighted CVH score = (β1*factor 1 + β2*factor 2 + β3*factor 3 + β4* factor 4 + β5*factor 5 + β6*factor 6) (6/sum of the coefficients). This weighted score ranged from 0 to 6 points and factors in the magnitudes of the adjusted relative risk for each factor. Additionally, to address the role of potential reverse causality, we repeated the main analyses in a sample excluding participants who developed incident diabetic complications or deaths within the first 3-year follow-up period. We repeated the main analyses with additional adjustment for diabetes duration. Finally, we performed the analysis in overall participants (including participants with and without diabetes) to examined the association between CVH and the risk of mortality and life expectancy.
All the analyses were performed using STATA 15 statistical software (Stata Corp, College Station, TX, USA) and R (version 3.6.1, R Foundation for Statistical Computing). All P-values were two-sided, with statistical significance set at 0.05.