Of the 8815 participants who took part in the phase 3 examination (that is, the baseline of the analyses reported here), we excluded 1578 for one or more of the following reasons: prevalent type 2 diabetes at phase 3 (n=162), missing follow-up on type 2 diabetes status (n=588), and missing data on risk factors/markers at phase 1 (n=1392). The final sample consisted of 7237 participants (2196 women). Excluded participants were more likely to belong to the low socioeconomic status group and were slightly older than those included in the analyses (P<0.001).
Table 1 shows baseline characteristics of the participants included in the analysis. During the mean 14.2 years follow-up, 818 incident cases type 2 diabetes were identified on the basis of a 75 g oral glucose tolerance test (n=425; 52%), use of antidiabetic drugs (n=188; 23%), or diagnosis by a physician (n=205; 25%). Participants in the lowest socioeconomic category had an almost twofold higher incidence of type 2 diabetes than those in the highest category (10.9 v 5.6 per 1000 person years). The prevalences of smoking, unhealthy diet, physical inactivity, obesity, high triglyceride concentration, and low high density lipoprotein cholesterol were higher in the lowest compared with the highest socioeconomic group (P<0.001), whereas heavy drinking was more common in the highest socioeconomic group (P<0.001). Socioeconomic status was unrelated to high blood pressure (P=0.72).
Table 1 Study participants’ characteristics at phase 3 (baseline) and incidence of diabetes over 17.7 years of follow-up by socioeconomic status. Values are numbers (percentages) unless stated otherwise
Table 2 shows results for the association between risk factors/markers for diabetes and incidence of type 2 diabetes. Unhealthy behaviours were related to an increased risk of developing type 2 diabetes over the follow-up. As anticipated, the associations of overweight (hazard ratio 1.92, 95% confidence interval 1.64 to 2.25) and obesity (4.79, 3.96 to 5.80) with incidence of type 2 diabetes were particularly strong. Hypertension and adverse lipid categories were also associated with type 2 diabetes, as was family history of diabetes.
Table 2 Association of health behaviours and other risk factors assessed at baseline (phase 3) with type 2 diabetes incidence (n=7237)
Table 3 shows results for the association between socioeconomic status and incidence of type 2 diabetes, as well as the contribution of health behaviours, body mass index, and biological risk markers assessed at phase 3 to this association. The hazard ratio for the lowest versus the highest socioeconomic status was 1.86 (1.48 to 2.32). This was attenuated by 17% when we controlled for health behaviours at baseline and by 18% when we controlled for body mass index. Overall, health behaviours and body mass index attenuated the association between socioeconomic status and type 2 diabetes by 33% (−1% to 78%). Adjustment for baseline biological risk markers (entered as continuous variables in the model) lowered the association by an additional 12%. In total, 45% (17% to 105%) of the socioeconomic gradient in type 2 diabetes was explained.
Table 3 Contribution of baseline risk factors/markers (phase 3) in explaining social inequalities in type 2 diabetes incidence (n=7237)
In table 4 shows the results when the risk factors/markers were assessed repeatedly over the follow-up. The first column shows that with longitudinal assessment over the follow-up, health behaviours and body mass index in combination explained 36% (22% to 64%) of the association between socioeconomic status and type 2 diabetes. This proportion increased to 45% (28% to 77%) when we additionally took blood lipids and systolic blood pressure into account. The right side of table 4 shows results of simultaneous adjustment for long term exposure to the risk factors/markers and their changes over time. In this model, health behaviours attenuated the socioeconomic status coefficient by 24% and body mass index alone by 23%; the total percentage attenuation due to health behaviours and body mass index was 45% (28% to 75%). Blood lipids and blood pressure contributed to an additional 8%, and in the final fully adjusted model 53% (29% to 88%) of the association between socioeconomic status and type 2 diabetes was explained.
Table 4 Contribution of repeatedly measured risk factors/markers in explaining social inequalities in type 2 diabetes incidence (n=7237)
We repeated all analyses in subgroups including only participants with complete data, only those who were free from coronary heart disease at baseline, and only white participants, as well as including adjustment for waist circumference and body mass index assessed at age 25. These yielded similar results to those reported in the main analysis (results available on request).
We repeated all analyses for men and women separately (supplementary tables A to D). Although results did not materially differ between the sexes, modifiable risk factors tended to explain a larger proportion of the socioeconomic gradient in type 2 diabetes in women than in men (attenuation for the model accounting for long term exposure to the risk factors: 42% in men and 81% in women, P for difference between the two estimations=0.43).
To examine whether our results are robust across different indicators of socioeconomic status, we repeated all analyses using education and income as alternative measures. Our findings of a major contribution of modifiable risk factors to socioeconomic differences in type 2 diabetes were consistent across the indicators of socioeconomic status (supplementary tables E and F). We also repeated the analysis using employment grade in six categories (supplementary table G). Again, the results were very similar to those in the main analysis based on employment grade in three categories.
Finally, we used three standard approaches to examine whether missing data on risk factors at baseline affected the findings. Firstly, we repeated the analysis using only three of the risk factors examined: smoking, alcohol consumption, and body mass index. These risk factors were available on a larger sample of the population (n=7750), allowing us to assess whether their role in explaining socioeconomic differences in type 2 diabetes in this larger sample was similar to that found in the smaller sample included in the main analysis. Regarding the model with baseline assessment of these risk factors, the contribution of smoking was 10% in the larger sample compared with 9% in the smaller sample. The contribution of alcohol consumption was 3% versus 3%, and that of body mass index was 20% versus 18%. Secondly, we used “inverse probability weighting” to correct the estimates for non-response.36
These analyses yielded similar results to those reported in the main analysis (supplementary table H). Thirdly, we used multiple multivariate imputation to replace missing values for risk factors at the study baseline (Stata ice/micombine procedures). Analyses on the imputed dataset (n=8232; 927 incident diabetes cases) yielded results virtually identical to those reported in the main analysis (results available on request).