Healthy lifestyle behaviors, mediating biomarkers, and risk of microvascular complications among individuals with type 2 diabetes: A cohort study

The UK Biobank is a large community-based prospective cohort study for common diseases of middle and older adults including over 500,000 participants aged 37 to 73 years from 22 sites across England, Scotland, and Wales between March 2006 and October 2010. Extensive data were obtained through touchscreen questionnaires, physical measurements, and biological samples at recruitment. Specific methods of data collection have been described previously [24,25].

Our sample of 15,104 was generated by including patients with T2D identified by using the algorithms method developed by the UK Biobank study [26] and excluding participants with prevalent macro- or microvascular complication cases, had incomplete information on lifestyle behaviors, or withdrawal from the study. The flowchart of patients included in the current study is present in S1 Fig.

The study was approved by the North West Multi-Centre 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 participants provided written informed consent. In the current analysis, we employed the UK Biobank study to test a priori hypothesis; we did not publish an analysis plan before conducting analyses between January 2022 and March 2022. The associations between lifestyle factors and the risk of microvascular complications in participants without excluding those with macrovascular complications and stratified analysis by preexisting cardiovascular disease (CVD) status were performed in response to peer review in July 2022. This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist).

Five lifestyle behaviors, namely, WC, smoking status, physical activity, habitual diet, and alcohol intake, were evaluated in the current analysis. We used WC instead of body mass index (BMI) to avoid the potential obesity paradox [27,28] as evidence found an obesity paradox when obesity was measured by BMI but not when measured by WC in patients with diabetes [29]. WC was measured using the Wessex nonstretchable sprung tape measurement, and low-risk WC was defined as <80 cm for women and <94 cm for men [30,31]. Data on smoking status were self-reported, and noncurrent smoking was defined as low-risk behavior. The frequency of all types of alcohol intake was reported using 6 predefined categories, between never to daily or almost daily. For participants who reported to drink alcohol, data on the average monthly or weekly alcohol intake from 6 types of alcohol beverages were collected. We calculated the average units of alcohol intake using the abovementioned information and defined low-risk drinking as moderate drinking (1 to 14 g/day for women or 1 to 28 g/day for men). Data on the type and duration of physical activity were derived from the questionnaire. Leisure-time physical activity score based on the 5 activities undertaken in the last 4 weeks was computed by multiplying the metabolic equivalent of task [MET] score of each activity by the minutes performed [32,33]. Light DIY (do-it-yourself), walking for pleasure, other exercises (e.g., swimming, cycling, keep fit, bowling), heavy DIY, and strenuous sports were given 1.5, 3.5, 4.0, 5.5, and 8.0 METs, respectively [34]. The midpoints of the frequency and duration of physical activities were used to calculate the time spent on each activity. We then classified the top third of the physical activity score as the low-risk group. In addition, we generated a dietary score to reflect the overall diet quality including 10 components, namely, fruits, vegetables, whole grains, fish, dairy, vegetable oils, refined grains, processed meat, unprocessed meat, and sugar-sweetened beverages. Low-risk diet was defined as meeting 5 or more ideal diet components [35]. Participants with each low-risk behavior were assigned 1 point; otherwise, 0 points. The overall healthy lifestyle score was the sum of individual score of the 5 lifestyle behaviors, ranging from 0 to 5, with higher score indicating healthier lifestyle.

Blood samples were collected from consenting participants at recruitment, separated by components and stored at UK Biobank (−80°C and LN₂) until analysis. Blood biomarkers were externally validated with stringent quality control in the UK Biobank; full details on assay performance have been given elsewhere [36]. We selected the potential biological biomarkers mediating the association between lifestyle factors and microvascular complications based on knowledge of potential pathways, including glycemic control determined by glycated hemoglobin (HbA_1c), lipid profile (total cholesterol [TC], high-density lipoprotein cholesterol [HDL-C], low-density lipoprotein cholesterol [LDL-C], triglycerides, apolipoprotein A, apolipoprotein B, and lipoprotein A), liver function (alanine aminotransferase [ALT], alkaline phosphatase [ALP], aspartate aminotransferase [AST], gamma glutamyltransferase [GGT], total bilirubin, total protein, and albumin), renal function (cystatin C, creatinine, urate, and urea), inflammation (C-reactive protein [CRP], and white blood cell count), and blood pressure indices (systolic blood pressure [SBP] and diastolic blood pressure [DBP]).

Comparisons of baseline characteristics across the categories of the overall healthy lifestyle score were made using ANOVA or chi-squared test. We also compared the differences between patients included in the current analysis and those who were excluded due to missing values. Person-years were calculated from the date of recruitment to the date of death, first endpoint, lost to follow-up, or the end of follow-up, whichever came first. The lost to follow-up variable in the UK Biobank has been created by amalgamating data from 5 possible sources: (1) Death reported to UK Biobank by a relative; (2) NHS records indicate they are lost to follow-up; (3) NHS records indicate they have left the UK; (4) UK Biobank sources report they have left the UK; (5) Participant has withdrawn consent for future linkage. The end of follow-up dates were 1 April 2017, 17 September 2016, and 1 November 2016, for centers in England, Wales, and Scotland, respectively. Cox proportional hazards regression models were used to calculated hazard ratios (HRs) and 95{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} confidence intervals (CIs) for the associations of individual lifestyle behaviors and overall healthy lifestyle score with risks of total and individual microvascular complications in patients with T2D. We imputed the missing values of covariates (≤7{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) using multiple imputations by chained equations with 5 imputations (SAS PROC MI with a fully conditional specification method and PROC MIANALYZE). Linear regression model and logistic regression model with all the covariates in the fully adjusted model were used to impute continuous variables and categorical variables, respectively. The percentage of missing values are present in S1 Table.

Three models were built. In Model 1, we adjusted for age (continuous, years), sex (male, female), Townsend Deprivation Index (continuous), and race/ethnicity (White, others). In Model 2, we further adjusted for education attainment (college or university degree, A/AS levels or equivalent or O levels/GCSEs, NVQ or HND or HNC or equivalent or other professional qualifications, none of the above), sleep duration (<6, 6 to 8, or ≥9 hours/day), family history of CVD (yes, no), family history of hypertension (yes, no), and prevalence of hypertension (yes, no). Finally, in Model 3, diabetes duration (continuous, years), HbA_1c (continuous, mmol/mol), use of diabetes medication (none, only oral medicine, insulin, and others), use of antihypertensive medication (yes, no), use of lipid-lowing medication (yes, no), and use of aspirin (yes, no) were additionally adjusted. Further, restricted cubic spline analysis was applied to test dose–response relationships between the healthy lifestyle score and risks of outcomes. We also calculated the population-attributable fractions (PAFs) using the {e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}par SAS Macro (https://www.hsph.harvard.edu/donna-spiegelman/software/par/) to estimate the proportion of microvascular complications that could theoretically be avoided if all participants adhered to 4 or more low-risk lifestyle behaviors.

Mediation effects of biomarkers on the associations of overall lifestyle score with risks of total and individual microvascular complications were evaluated using mediation package in R. Indirect, direct, and total effects for each mediator were computed via combining the mediator and outcome models with the adjustment of all the covariates in Model 3. Nonparametric bootstrap resampling was used to compute the CIs of the proportions of mediations. We selected the available biomarkers from the UK Biobank for the mediation analyses based on knowledge of potential causal pathways to predisposing to microvascular complications or mortality [19,37–40]. The selected biomarkers were considered as potential mediators following two-step analysis. First, we assessed the associations of all biomarkers with the overall lifestyle score using the multivariable-adjusted linear regression models. Second, we evaluated the associations of biomarkers that were significantly associated with the overall lifestyle score, with risks of all the outcomes using the multivariable-adjusted Cox regression model. We then chose the biomarkers significantly associated with each outcome for the mediation analysis accordingly.

Among 15,104 participants with T2D (60.3{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} male; mean age, 59.3 years), there were 3,406 (22.6{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}), 6,080 (40.3{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}), 4,062 (26.9{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}), 1,556 (10.3{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) having 0 or 1, 2, 3, and 4 or 5 low-risk lifestyle behaviors, respectively. The baseline characteristics are shown in Table 1. Participants with more low-risk lifestyle behaviors were more likely to be men, White, less deprived, highly educated, sleep recommended hours, have a lower level of HbA_1c, and have a lower prevalence of hypertension. They were less likely to use aspirins and medications for diabetes, dyslipidemia, and hypertension. In addition, compared the participants who were excluded due to missing values, those included in the current analysis were more likely to be men, White, less deprived, highly educated, noncurrent smokers, physically active, moderate alcohol drinkers, and eat healthier (S2 Table).

During 117,445 person-years of follow-up (median 8.1 years; interquartile range 7.3 to 8.8 years; maximum 11.9 years), there occurred 1,639 (10.9{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) deaths and 1,296 (8.6{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) composite microvascular complications cases, including 558 (3.7{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) diabetic retinopathy, 625 (4.1{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) diabetic kidney disease, and 315 (2.1{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) diabetic neuropathy. Among all the cases, one case of diabetic kidney disease was uniquely identified from death records. S3 Table shows the associations between individual lifestyle behaviors and all the outcomes. Being physically active, with lower WC, and moderate alcohol intake were associated with a lower risk of microvascular complications, while noncurrent smoking and healthy diet were not. The overall healthy lifestyle score was associated with lower risks of all the outcomes in a dose–response manner (all Ps for linear trend ≤0.01; Table 2 and Figs 1 and S2). Compared with participants with 0 to 1 low-risk lifestyle behavior, participants with 4 to 5 low-risk lifestyle behaviors had HRs (95{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} CIs) of 0.65 (0.46, 0.91) for diabetic retinopathy, 0.43 (0.30, 0.61) for diabetic kidney disease, 0.46 (0.29, 0.74) for diabetic neuropathy, and 0.54 (0.43, 0.68) for the composite microvascular complications, respectively. For each number increment in low-risk lifestyle behavior, there was a 13{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} lower risk of diabetic retinopathy (HR, 0.87; 95{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} CI: 0.80, 0.95), 22{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} lower risk of diabetic kidney disease (HR, 0.78; 95{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} CI: 0.72, 0.85), 27{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} lower risk of diabetic neuropathy (HR, 0.73; 95{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} CI: 0.65, 0.83), and a 18{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} lower risk of the composite microvascular complications (HR, 0.82; 95{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} CI: 0.77, 0.87).

Fig 1. Dose–response relationship of the healthy lifestyle score with risk of microvascular complications among individuals with T2D.

X-axis showed the numbers of low-risk lifestyle behaviors, and y-axis showed the HRs of the composite microvascular complications (A), diabetic retinopathy (B), diabetic kidney disease (C), and diabetic neuropathy (D). Black curves were HRs, and grey zones were 95{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} CIs. Multivariable-adjusted models were adjusted for age (continuous, years), sex (male, female), ethnicity (White, others), education attainment (college or university degree, A/AS levels or equivalent or O levels/GCSEs or equivalent or other professional qualifications, or none of the above), Townsend Deprivation Index (continuous), sleep duration (<6, 6–8, or ≥9 hours/day), family history of CVD (yes, no), family history of hypertension (yes, no), prevalence of hypertension (yes, no), diabetes duration (continuous, years), HbA_1c (continuous, mmol/mol), use of diabetes medication (none, only oral medication pills, or insulin or others), use of antihypertensive medication (yes, no), use of lipid-lowing medication (yes, no), and use of aspirin (yes, no). All P_{-nonlinearity} were ≥0.09 and all P for overall association were <0.001 (except for diabetic retinopathy: P for overall association = 0.008). CI, confidence interval; CVD, cardiovascular disease; HR, hazard ratio; T2D, type 2 diabetes.

https://doi.org/10.1371/journal.pmed.1004135.g001

In addition, the estimated PAFs of nonadherence to 4 or more low-risk lifestyle factors were 39.0{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} (17.7{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}, 56.8{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) for diabetic kidney disease and 25.3{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} (10.0{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}, 39.4{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}) for the composite microvascular complications (Table 2).

All the biomarkers were significantly associated with the overall lifestyle score except for total protein, lipoprotein A, and SBP (S4 Table). The associations between the selected biomarkers and all outcomes are shown in S5 Table. Six significant mediators were detected on the associations of lifestyle score with risk of the composite microvascular complications and diabetic kidney disease, namely, albumin, HDL-C, triglycerides, apolipoprotein A, CRP, and HbA_1c. The relationship between the lifestyle behaviors and risk of diabetic neuropathy was mediated by cystatin C, GGT, total bilirubin, albumin, HDL-C, triglycerides, apolipoprotein A, CRP, and HbA_1c with the proportion of mediation effect ranging from 3.22{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} to 11.35{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}. Collectively, the mediators explained 23.20{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}, 24.40{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf}, and 31.90{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} of the associations of overall lifestyle behaviors with composite microvascular complications, diabetic kidney disease, and diabetic neuropathy, respectively. In addition, our data showed that among all the potential biomarkers, only HbA_1c was a significant mediator that explained 15.26{e4f787673fbda589a16c4acddca5ba6fa1cbf0bc0eb53f36e5f8309f6ee846cf} of the relationship between the overall lifestyle score and risk of diabetic retinopathy (Table 3).

Consistent results were observed when analyses were stratified by age, sex, education, diabetes duration, use of hypoglycemic medication, and HbA_1c level. No significant interaction was observed between the healthy lifestyle score and the stratified factors on the outcomes considering multiple comparisons (S3 Fig). Further, the results of different combinations of low-risk lifestyle factors showed that the increased numbers of low-risk lifestyle factors were associated with graded lower risks of diabetic retinopathy, diabetic kidney disease, diabetic neuropathy, and the composite microvascular complications (S6 Table).

In the sensitivity analyses, the results were generally robust when excluding patients with events that occurred within the first 2 years of follow-up, defining low-risk alcohol intake as moderate drinking and nondrinking, generating the lifestyle score using BMI or waist-to-hip ratio instead of WC, or generating the overall lifestyle score as a weighted score (S7–S10 Tables). The association between overall lifestyle behaviors and risk of diabetic kidney disease was slightly attenuated when estimated glomerular filtration rate (eGFR) was additionally adjusted, and the results of mediation analysis for diabetic kidney disease were largely unchanged with the additional adjustment of eGFR (S11 and S12 Tables). Further, we observed similar results when patients with preexisting CVD were included and in patients with preexisting CVD, although diabetic retinopathy did not reach statistical significance in patients with preexisting CVD probably due to the insufficient power (S13 and S14 Tables). Finally, consistent results were demonstrated when we used 2 competing risk models accounting for the death (S15 Table).