Early-life risk factors
In this study, we selected three early-life modifiable factors (birth weight, maternal smoking around birth, and breastfeeding) that have each been reported in prior UK Biobank studies to be associated with T2D development [6,7,8, 19]. At the enrollment visit, participants reported information on the early-life factors using touchscreen questionnaires. Participants reported information on maternal smoking status during pregnancy and breastfeeding status with the following questions: “Did your mother smoke regularly around the time when you were born?” and “Were you breastfed when you were a baby?”. We defined the risk factors as “yes” for maternal smoking and “no” for breastfeeding. Participants were asked to enter their birth weight in kilograms. We defined low birth weight as a risk factor (<2.5 kg), with birth weight ≥2.5 kg serving as the reference group. We combined normal and high birth weight as the reference group because a previous meta-analysis study had illustrated an “L-shape” relation between birth weight and T2D risk [5], and in UK Biobank participants, high birth weight was associated with a lower risk of T2D [19]. We developed an early-life risk factor score (ERS) by calculating the cumulative number of the risk factors for each individual. The ERS ranges from 0 to 3, with a higher score representing more disadvantaged experiences. In addition, we further constructed weighted early-life risk factors, using the equation: (β1 × factor 1 + β2 × factor 2 + β3 × factor 3) × (3/sum of the β coefficients). Here, β1, β2, and β3 denoted the estimates of β coefficients for breastfeeding, maternal smoking around birth, and birth weight, respectively.
Modifiable Healthy Lifestyle Factors during Adulthood
In the current study, four modifiable behavioral lifestyle factors were selected according to the AHA’s LE8 construct to compute a modifiable healthy lifestyle score (MHS) during adulthood for each participant. The four lifestyle factors include diet, physical activity, nicotine exposure, and sleep health. The lifestyle information was obtained from the UK Biobank questionnaires at the enrollment visit. The detailed scoring process of the four components was provided in Additional file 1: Table S1-S2. In brief, each lifestyle metric was given a score between 0 and 100, with a higher score indicating a healthier lifestyle. We divided the healthy lifestyle scores into 3 categories based on AHA’s guidelines: healthy (80–100), moderate (50–79), and unhealthy (0–49) [13].
Assessment of T2D and T2D Polygenic risk scores
We identified T2D individuals with both prevalent T2D at baseline and new incident T2D during the follow up. Participants’ prevalent T2D at baseline was determined through the hospitalization records, death registration, self-reported medical history and medication use, blood glucose (random glucose ≥11.1 mmol/L), and glycated hemoglobin (≥6.5%). Incident T2D was determined by hospital inpatient records from the Hospital Episode Statistics for England, Scottish Morbidity Record data for Scotland, and the Patient Episode Database for Wales and was identified based on International Classification of Diseases, Tenth Revision (ICD-10) codes (E11) [20]. And all participants enrolled in the UK Biobank study were followed up. We tracked the incidence of newly diagnosed T2D among participants who were free of T2D at baseline during the follow-up period. Additionally, the UK Biobank provided detailed records on participants’ use of diabetes medications, mortality data, and the endpoint of follow-up (September 30, 2021, for centers in England; February 28, 2018, for centers in Wales; and July 31, 2021, for centers in Scotland). We did not exclude individuals with T2D at baseline because T2D is most likely to happen after early-life experiences; thus, excluding these participants may underestimate the association between ERS and the risk of T2D.
A standard polygenic risk score (PRS) for type 2 diabetes (T2D-PRS) (UK Biobank Data-Field 26285) has been utilized as a metric to quantify genetic predisposition for T2D. This standard PRS was derived by conducting a meta-analysis across multiple external Genome-Wide Association Study (GWAS) datasets and applying it uniformly to all participants within the UK Biobank. Detailed information regarding access to these PRS data and GWAS resources can be found in an earlier study [21]. Individualized PRS values were calculated by summing the posterior effect size of each variant weighted by its allele dosage across the entire genome. The standard T2D-PRS has been utilized in previous UK Biobank studies [22, 23], and has demonstrated a significantly improved predictive performance than other PRS models [24]. For analytical purposes, individuals were categorized based on their T2D-PRS scores into three PRS groups: low risk (tertile 1), medium risk (tertile 2), and high risk (tertile 3). Higher tertiles correspond to greater genetic risk of developing type 2 diabetes.
Covariates
We considered following sociodemographic, biological and behavioral characteristics as covariates based on directed acyclic graphs, including age (continuous in years), sex (male; female), ethnicity (White; Asian; Black; Mixed/other), educational attainment, household income (<18,000; 18,000-30,999; 31,000-51,999; >52,000 £/year), Townsend deprivation index (TDI), a measure of neighborhood-level deprivation (continuous), region (England; Scotland; Wales), alcohol drinking status (never; former; current), year of birth (1934-1940; 1941-1950; 1951-1960; 1961-1970), family history of diabetes (yes; no), and body mass index (<25; 25 to <30; ≥30 kg/m2), the amount of smoking (never; former; current), the amount of drinking (three or four times a week and more; once a month to twice a week; special occasions only; never). Education attainment was classified into three categories: high (College or University degree), medium (A levels/AS levels or equivalent, NVQ, HND, HNC, or equivalent, and other professional qualifications), and low (CSEs or equivalent, O levels/GCSEs or equivalent, or none of the above). Genetic ancestry, indicating genetic similarities indicating the geographic origins of common ancestors, was classified into two categories: European and non-European.
Statistical analysis
Baseline characteristics of participants were described as numbers (proportions) for categorical variables and mean (± standard deviation [SD]) for continuous variables, according to ERS categories, T2D-PRS tertiles, MHS categories, and T2D status. Differences in characteristics were compared using Chi-square test for categorical variables and ANOVA for continuous variables.
Cox proportional hazards regression models were fitted to examine the associations of ERS with the risk of T2D, accounting for left, right, and interval censoring, with ERS as a categorical variable and using individuals with 0 early-life risk factors as the reference group. Hazard ratios (HRs) and 95% confidence intervals (CIs) were computed. Linear trends were tested by treating the ERS as a continuous variable. Participants’ age was selected as the time scale, with follow-up time calculated from birth until time of T2D event, or age at enrollment (those with prevalent T2D at baseline but without diagnostic age), or right censored. We conducted following models: Model 1: an unadjusted model; Model 2: adjusted for sex; Model 3: further adjusted for age, TDI, birth year, ethnicity, household income, educational attainment, and region based on Model 2; Model 4: further adjusted for family history of diabetes based on Model 3. To examine whether the association were modified by the genetic risk or adulthood lifestyles, we conducted stratified analyses, according to participants’ T2D-PRS level (tertile 1; tertile 2; tertile 3) and MHS level (unhealthy; moderate; healthy), respectively. To test the interactions between early-life risk factors and T2D-PRS, and MHS, cross-product of the two terms (e.g., early-life risk factors and PRS) were included in the fully adjusted Cox model. In joint analyses, we used individuals with 0 early-life risk factor combined with low tertile of T2D-PRS or healthy adulthood lifestyle as the referent, respectively.
Seven sensitivity analyses were performed to test the robustness of the association between ERS and T2D risk: 1) we assessed the association between weighted ERS and the risk of T2D; 2) we explored the association between ERS, PRS, MHS and T2D, the association between ERS and PRS on T2D, and the association between ERS and MHS on T2D stratified by sex; 3)we further adjusted for BMI, the amount of drinking, the amount of smoking, and genetic ancestry; 4) we explored the associations between ERS and T2D risk without considering left-censoring by excluding participants with prevalent T2D at baseline; 5) when examining the effect of low birth weight, we used individuals with normal birth weight (2.5-4 kg) as the reference instead of the combination of normal and high birth weight (≥2.5 kg), to exclude the potential effect of high birthweight; 6) we excluded participants who died without T2D in follow-up; 7) we examined the 8 combinations of the three early-life factors with T2D risk.
All analyses were performed using SAS version 9.4. All values were two-sided and a P < 0.05 was considered statistically significant.