Type 2 diabetes risks and determinants in second-generation migrants and mixed ethnicity people of South Asian and African Caribbean descent in the UK

Aims/hypothesis Excess risks of type 2 diabetes in UK South Asians (SA) and African Caribbeans (AC) compared with Europeans remain unexplained. We studied risks and determinants of type 2 diabetes in first- and second-generation (born in the UK) migrants, and in those of mixed ethnicity. Methods Data from the UK Biobank, a population-based cohort of ~500,000 participants aged 40–69 at recruitment, were used. Type 2 diabetes was assigned using self-report and HbA1c. Ethnicity was both self-reported and genetically assigned using admixture level scores. European, mixed European/South Asian (MixESA), mixed European/African Caribbean (MixEAC), SA and AC groups were analysed, matched for age and sex to enable comparison. In the frames of this cross-sectional study, we compared type 2 diabetes in second- vs first-generation migrants, and mixed ethnicity vs non-mixed groups. Risks and explanations were analysed using logistic regression and mediation analysis, respectively. Results Type 2 diabetes prevalence was markedly elevated in SA (599/3317 = 18%) and AC (534/4180 = 13%) compared with Europeans (140/3324 = 4%). Prevalence was lower in second- vs first-generation SA (124/1115 = 11% vs 155/1115 = 14%) and AC (163/2200 = 7% vs 227/2200 = 10%). Favourable adiposity (i.e. lower waist/hip ratio or BMI) contributed to lower risk in second-generation migrants. Type 2 diabetes in mixed populations (MixESA: 52/831 = 6%, MixEAC: 70/1045 = 7%) was lower than in comparator ethnic groups (SA: 18%, AC: 13%) and higher than in Europeans (4%). Greater socioeconomic deprivation accounted for 17% and 42% of the excess type 2 diabetes risk in MixESA and MixEAC compared with Europeans, respectively. Replacing self-reported with genetically assigned ethnicity corroborated the mixed ethnicity analysis. Conclusions/interpretation Type 2 diabetes risks in second-generation SA and AC migrants are a fifth lower than in first-generation migrants. Mixed ethnicity risks were markedly lower than SA and AC groups, though remaining higher than in Europeans. Distribution of environmental risk factors, largely obesity and socioeconomic status, appears to play a key role in accounting for ethnic differences in type 2 diabetes risk. Graphical abstract Supplementary Information The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-021-05580-7.


ESM
: Additional baseline characteristics of UK Biobank participants by ethnicity; European, South Asian an African Caribbean origin groups. Data are n (%) and mean (standard deviation). First/second-generation assigned by year of migration. European and South Asian/ African Caribbean first-and second-generation groups are age and sex-matched (2:1:1). European, Mixed and South Asian/ African Caribbean groups are age and sex-matched (4:1:4).

ESM Figure 7: GENESIS principal components (PCs) using five k-means on the non-EUR sample.
We identified and removed individuals who were clustered in terms of Chinese ancestry alongside their self-reported ethnicity.

15
ESM Figure 11: Diagrams of mediational model between South Asian ethnic groups (A, B), generations (C) and level of admixture (D) and the proportion of the observed difference in type 2 diabetes prevalence, which is explained by five mediators (smoking, deprivation, WHR, height, and years of education) and their interrelationships. The dashed arrows indicate non-significant association and the numbers are standardised estimates, age and sex adjusted. The mediated percentages shown are rounded to the nearest integer and for this reason they might not be added up to the total (*).

ESM Figure 12: Diagram of mediational model between Mixed Europeans/South Asian versus South
Asians in type 2 diabetes prevalence, using five mediators (smoking, deprivation, WHR, height, and years of education) and their interrelationships, with focus on the pathway ethnicity-deprivation-WHR. The dashed arrows indicate non-significant association and the numbers are standardised estimates, age and sex adjusted. When breaking down the component parts (red arrows), we observed that ethnicity was strongly associated with deprivation (βstd=0.154, 95% CI 0.122 to 0.186), which in turn had a marked impact on WHR (βstd=0.056, 95% CI 0.033 to 0.080), attenuating any effect of ethnicity itself (βstd=0.013, 95% CI -0.010 to 0.036). Thus, a large part (37%) of the effect of ethnicity on the WHR in ESA versus Europeans was mediated via deprivation.

Regression models and sensitivity analysis
Model 1: Ethnicity/generations+ age +sex +WHR (for SA)/BMI (for AC) +deprivation +smoking+height+years of education Model 2: Model 1 +physical activity + scores for healthy diet pattern+scores for unhealthy diet pattern ESM Figure 18: Forest plots of multivariate regression models for diabetes adding diet and physical activitysensitivity analysis. Sensitivity analysis was conducted on full case data for diet and physical activity.

28
ESM Figure 23: Diagrams of mediational model between South Asian ethnic groups (A, B), generations (C) and level of admixture (D) and the proportion of the observed difference in type 2 diabetes prevalence, which is explained by five mediators (smoking, deprivation, BMI, height, and years of education) and their interrelationships. The dashed arrows indicate non-significant association and the numbers are standardised estimates, age and sex adjusted. The mediated percentages shown are rounded to the nearest integer and for this reason they might not be added up to the total (*).

30
ESM Figure 24: Diagrams of mediational model between African Caribbean ethnic groups (A, B), generations (C) and level of admixture (D) and the proportion of the observed difference in type 2 diabetes prevalence, which is explained by five mediators (smoking, deprivation, WHR, height and years of education) and their interrelationships. The dashed arrows indicate non-significant association and the numbers are standardised estimates, age and sex adjusted. The mediated percentages shown are rounded to the nearest integer and for this reason they might not be added up to the total (*).