Abstract

Background Type 2 diabetes mellitus (T2DM) is one of the most common chronic diseases worldwide. In high-income countries, low socioeconomic status seems to be related to a high incidence of T2DM, but very little is known about the intermediate factors of this relationship.

Method We performed a case-cohort study in eight Western European countries nested in the EPIC study (n = 340 234, 3.99 million person-years of follow-up). A random sub-cohort of 16 835 individuals and a total of 12 403 incident cases of T2DM were identified. Crude and multivariate-adjusted hazard ratios (HR) were estimated for each country and pooled across countries using meta-analytical methods. Age-, gender- and country-specific relative indices of inequality (RII) were used as the measure of educational level and RII tertiles were analysed.

Results Compared with participants with a high educational level (RII tertile 1), participants with a low educational level (RII tertile 3) had a higher risk of T2DM [HR: 1.77, 95% confidence interval (CI): 1.69–1.85; P-trend < 0.01]. The HRs adjusted for physical activity, smoking status and propensity score according to macronutrient intake were very similar to the crude HR (adjusted HR: 1.67, 95% CI: 1.52–1.83 in men; HR: 1.88, 95% CI: 1.73–2.05 in women). The HRs were attenuated only when they were further adjusted for BMI (BMI-adjusted HR: 1.36, 95% CI: 1.23–1.51 in men; HR: 1.32, 95% CI: 1.20–1.45 in women).

Conclusion This study demonstrates the inequalities in the risk of T2DM in Western European countries, with an inverse relationship between educational level and risk of T2DM that is only partially explained by variations in BMI.

Introduction

In high-income countries, low socio-economic status (SES) has been linked to higher incidence of mortality from different chronic diseases, such as some types of cancer,1 respiratory diseases2 and cerebrovascular and cardiovascular diseases.3,4 Cardiovascular diseases, the group of chronic diseases in which the relationship with SES is the most evident,5 share a number of aetiological factors with type 2 diabetes mellitus (T2DM).6,7 Several studies showed a strong inverse relationship, particularly in women, between T2DM prevalence and SES as determined by educational level,8,-12 income10,13 and deprivation score.9,14,-17 The most recent research, focused mainly on the relationship between SES and T2DM incidence, has been carried out in North America and Europe, including a number of large-scale studies18,-26 and a recent meta-analysis.27

SES refers to an individual’s social position relative to other members of a society, and cannot be measured directly. Instead, different proxies for SES are used, and often include educational level, income or occupation.28 These three indicators are strongly related and complementary, but are not interchangeable.29 The major advantage of using educational level as a proxy for SES is its simplicity and universality.30 Nevertheless, the relationship between educational level and health is complex, as it is mediated by anthropometric factors, lifestyle, behaviour, access to health services and knowledge of health promotion.31 Furthermore, improving national educational levels could be a feasible public health goal for European countries, and would make the study of the association between educational level and risk of disease a basis of observation that could readily be translated into public health measures.

The aim of this study was to assess the association between risk of T2DM and educational level (as a proxy for SES) for the first time in a large European nested case-cohort study with a validated case ascertainment procedure and detailed information on other risk factors that were measured in a standardized manner across eight Western European countries.

Subjects and methods

Study population

The participants, methods, study design and measurements have been described previously.32 Briefly, the InterAct Project is funded by the Sixth European Framework Programme. It was initiated to investigate how genetic and lifestyle behavioural factors, particularly diet and physical activity, interact for the risk of developing diabetes and how knowledge about such interactions may be translated into preventive action. As part of the wider InterAct project, consortium partners have established a case-cohort study of incident type 2 diabetes (InterAct study) based on cases occurring in European Prospective Investigation into Cancer and Nutrition (EPIC) cohorts between 1991 and 2007 in 8 of 10 EPIC countries participating in InterAct. All EPIC participants signed an informed consent form; approval for study was obtained from the ethical review boards of the International Agency for Research on Cancer and from all local institutions where participants were recruited.33

A total of 340 234 EPIC participants were followed up for a mean time of 11.7 years during which 12 403 verified incident cases of T2DM were identified and included in the InterAct project. A random sub-cohort of 16 835 individuals was selected from participants with available stored blood and buffy coat, stratified by centre. Because of the random nature of sub-cohort selection, which is a feature of the case-cohort design applied in the present study, the sub-cohort also included 778 individuals who developed T2DM during follow-up. Information on educational level was present for 12 108 incident cases of T2DM and 15 850 sub-cohort individuals, who were included in the final analyses [295 incidence cases of T2DM (∼2%) and 985 sub-cohort individuals (∼6%) had missing information and were excluded].

Ascertainment and verification of incident cases of T2DM

Incident cases of T2DM occurring between 1991 and 2007 were ascertained and verified by EPIC centres participating in the InterAct project. A high-sensitivity approach was adopted for the ascertainment of incident cases of T2DM, with the aim of identifying all incident cases and excluding all individuals with prevalent diabetes. Evidence used to ascertain incident T2DM included self-reported T2DM in follow-up questionnaires (T2DM diagnosed by a medical doctor or anti-diabetic drug use), linkage to primary and secondary care registers, medication use (prescription registers), hospital admission and mortality data.32

To increase the specificity of the case definition, and to avoid inclusion in the study of cases based on self-report of T2DM alone, further evidence was sought for all incident cases of T2DM. T2DM cases were included in the study only if confirmation of the diagnosis was secured from no less than two independent sources, including individual medical-record review.

Relative indices of inequality as the measure of educational level

Educational level was used as a proxy for SES, and was categorized as primary school or none (low educational level), vocational or other secondary school (middle educational level), and university or vocational postsecondary school (high educational level). Relative indices of inequality (RII) were then calculated to avoid the problem of differences in the distribution of educational levels between different countries, genders and birth cohorts. The RII was assigned by ranking the earlier-mentioned educational levels according to the proportion of participants within relevant strata for each country, by 10-year age group and gender.34,35 The midpoint of the cumulative proportional distributions of each educational level was used to calculate the RII score. For example, if within a given category (country, age group, gender) 60% of participants attended primary school only (low educational level), 30% attended secondary school (middle educational level) and 10% attended high school (high educational level), in that category each participant that attended primary school only would be assigned a score of 0.30 (midpoint of 60%), each participant that belonged to the intermediate stratum would be assigned a score of 0.75 (0.60 of the first level, plus 0.30/2, the midpoint of 30%). Accordingly, the remaining 10% of subjects in the stratum of high educational level would receive a score of 0.95 (0.60 of the first level + 0.30 of the second level + 0.10/2, the midpoint of 10%). This calculation was performed for each specific category.

Other variables

Self-administered questionnaires were used to collect information on lifestyle factors and dietary habits upon recruitment into the EPIC study. For the present study we used the information on body mass index (BMI, continuous variable), marital status (single, married, separated and widowed), smoking status (never, former and current smoker), alcohol consumption (never, former and current drinker), physical activity (covering occupation and recreational activity according to a validated physical activity index; inactive, moderately inactive, moderately active and active)36 and macronutrient intake (total energy, total proteins, total carbohydrates and total fats, continuous variable). Weight and height measurements were also obtained upon recruitment into the EPIC study using standardized protocols. Information on marital status was available in six of the countries participating in the InterAct project.

Statistical analysis

The distribution of exposures and putative confounders were summarized in the sub-cohort using means and standard deviations for continuous variables, and percentages and frequencies for categorical variables. We used a causal diagram37 to represent our qualitative a priori assumptions about the underlying mechanistic or indirect pathways by which educational level may affect incident T2DM. In Figure 1 we present the directed acyclic graph (DAG)44 that was used to determine the confounders that were introduced into the statistical model.

Figure 1

Directed acyclic graph (DAG) for the causal relationship between education level and incidence for type 2 diabetes mellitus. An arrow from a factor to another means possible association. Bold arrows mean known consistent causal associations. Factors in the boxes are confounders to adjust for, following DAG rules37

Figure 1

Directed acyclic graph (DAG) for the causal relationship between education level and incidence for type 2 diabetes mellitus. An arrow from a factor to another means possible association. Bold arrows mean known consistent causal associations. Factors in the boxes are confounders to adjust for, following DAG rules37

In the present study participants were assigned an RII score of 1 (low inequality = high educational level), 2 (middle inequality = middle educational level) or 3 (high inequality = low educational level) based on tertiles of these RII values.

In our causal model we observed that individual dietary variables have a small effect, so we decided to summarize this information using a propensity score,45 which yielded a more efficient regression model. In the analysis presented in this report, the propensity score represents the probability that an individual belongs to a certain RII tertile given his or her macronutrient intake (total energy, total proteins, total carbohydrates and total fats).

We fitted two models: model 1 included physical activity, smoking status and propensity score according to macronutrient intake; model 2 also included BMI (in quartiles). We did not adjust analyses for age or gender, as the investigated principal determinant (RII) was already standardized by those variables. Neither did we adjust the analyses for alcohol consumption, following the results of the DAG analysis.

Hazard ratios (HR) were calculated using a weighted Cox proportional hazards regression analysis, modified to be appropriate for this case-cohort design.46 The proportional hazard assumptions were tested in the sub-cohort using the Grambsch and Therneau test. Age was used as an underlying time variable. We used Prentice weight as suggested in the paper by Onland-Moret et al.:47 all sub-cohort individuals (whether or not they later became T2DM cases) were equally weighted in the likelihood function. Incident cases of T2DM (not including sub-cohort individuals who developed T2DM) were not weighted until they were diagnosed with T2DM. Once diagnosed they were considered incident cases of T2DM and given the same weight as the incident cases of T2DM in the random sub-cohort.

The HRs for T2DM were estimated separately for each country, and then combined across countries using meta-analytical methods. As resulting heterogeneity was not significant, we used a fixed effects model to obtain pooled HRs.48 Separate sensitivity analyses were performed, one further adjusting for marital status (in the six centres that had this information available), and the other after excluding incident cases of T2DM diagnosed within the first 2 years of follow-up. All statistical tests were two-sided. Analyses were performed using STATA v10.0 and Comprehensive Meta-Analysis v2.2.

Results

Educational levels by country and gender in incident cases of T2DM and sub-cohort individuals are displayed in Table 1; the differences in educational levels across countries are to some extent because of the different sampling strategies utilized during the recruitment of the EPIC cohort, and to different age and gender distributions in different countries. Table 2 shows the characteristics of sub-cohort individuals by RII tertiles. Mean BMI increased from RII tertile 1 (more educated individuals) to RII tertile 3 (less educated individuals). The percentage of current smokers, never and former alcohol drinkers and physically active men and women decreased with increasing educational level.

Table 1

Education level by country and sex in incident cases of type 2 diabetes mellitus and participants in the random sub-cohort in the InterAct project; data shown are numbers of individuals (percentage)

Country Men
 
Women
 
 University or vocational postsecondary school Other secondary school Vocational secondary school Primary school or none University or vocational postsecondary school Other secondary school Vocational secondary school Primary school or none 
France         
    Cases – – – – 96 (34.8) 132 (47.8) – 48 (17.4) 
    Sub-cohort – – – – 225 (40.0) 273 (48.5) – 65 (11.5) 
Italy         
    Cases 57 (10.6) 130 (24.2) 69 (12.9) 281 (52.3) 68 (8.0) 145 (17.2) 76 (9.0) 555 (65.8) 
    Sub-cohort 97 (14.8) 181 (27.7) 95 (14.5) 281 (43.0) 184 (13.8) 320 (24.1) 150 (11.3) 674 (50.8) 
Spain         
    Cases 84 (6.4) 93 (7.0) 151 (11.4) 995 (75.2) 27 (2.2) 34 (2.8) 30 (2.5) 1128 (92.5) 
    Sub-cohort 202 (14.8) 119 (8.8) 166 (12.2) 873 (64.2) 201 (9.1) 135 (6.2) 115 (5.2) 1751 (79.5) 
UK         
    Cases 86 (17.9) 46 (9.6) 165 (34.4) 183 (38.1) 42 (11.5) 25 (6.8) 118 (32.2) 181 (49.5) 
    Sub-cohort 105 (23.8) 42 (9.5) 166 (37.6) 128 (29.1) 147 (21.2) 81 (11.7) 232 (33.5) 233 (33.6) 
The Netherlands         
    Cases 23 (13.5) 31 (18.1) 88 (51.5) 29 (16.9) 50 (7.8) 166 (25.7) 215 (33.3) 214 (33.2) 
    Sub-cohort 85 (30.8) 55 (19.9) 100 (36.2) 36 (13.1) 238 (19.6) 384 (31.6) 358 (29.4) 236 (19.4) 
Germany         
    Cases 299 (31.4) 52 (5.5) 290 (30.4) 313 (32.8) 94 (14.9) 26 (4.1) 253 (40.2) 257 (40.8) 
    Sub-cohort 383 (44.7) 47 (5.5) 246 (28.7) 181 (21.1) 336 (28.1) 89 (7.4) 528 (44.1) 244 (20.4) 
Sweden         
    Cases 208 (15.0) 209 (15.1) 314 (22.7) 653 (47.2) 173 (14.1) 103 (8.4) 330 (26.9) 621 (50.6) 
    Sub-cohort 251 (19.7) 255 (20.0) 271 (21.3) 497 (39.0) 371 (22.3) 238 (14.3) 465 (27.9) 590 (35.5) 
Denmark         
    Cases 247 (20.7) 90 (7.5) 362 (30.3) 496 (41.5) 57 (6.7) 71 (8.3) 396 (46.2) 333 (38.8) 
    Sub-cohort 332 (29.1) 96 (8.4) 326 (28.6) 386 (33.9) 102 (10.3) 112 (11.4) 460 (46.7) 311 (31.6) 
Country Men
 
Women
 
 University or vocational postsecondary school Other secondary school Vocational secondary school Primary school or none University or vocational postsecondary school Other secondary school Vocational secondary school Primary school or none 
France         
    Cases – – – – 96 (34.8) 132 (47.8) – 48 (17.4) 
    Sub-cohort – – – – 225 (40.0) 273 (48.5) – 65 (11.5) 
Italy         
    Cases 57 (10.6) 130 (24.2) 69 (12.9) 281 (52.3) 68 (8.0) 145 (17.2) 76 (9.0) 555 (65.8) 
    Sub-cohort 97 (14.8) 181 (27.7) 95 (14.5) 281 (43.0) 184 (13.8) 320 (24.1) 150 (11.3) 674 (50.8) 
Spain         
    Cases 84 (6.4) 93 (7.0) 151 (11.4) 995 (75.2) 27 (2.2) 34 (2.8) 30 (2.5) 1128 (92.5) 
    Sub-cohort 202 (14.8) 119 (8.8) 166 (12.2) 873 (64.2) 201 (9.1) 135 (6.2) 115 (5.2) 1751 (79.5) 
UK         
    Cases 86 (17.9) 46 (9.6) 165 (34.4) 183 (38.1) 42 (11.5) 25 (6.8) 118 (32.2) 181 (49.5) 
    Sub-cohort 105 (23.8) 42 (9.5) 166 (37.6) 128 (29.1) 147 (21.2) 81 (11.7) 232 (33.5) 233 (33.6) 
The Netherlands         
    Cases 23 (13.5) 31 (18.1) 88 (51.5) 29 (16.9) 50 (7.8) 166 (25.7) 215 (33.3) 214 (33.2) 
    Sub-cohort 85 (30.8) 55 (19.9) 100 (36.2) 36 (13.1) 238 (19.6) 384 (31.6) 358 (29.4) 236 (19.4) 
Germany         
    Cases 299 (31.4) 52 (5.5) 290 (30.4) 313 (32.8) 94 (14.9) 26 (4.1) 253 (40.2) 257 (40.8) 
    Sub-cohort 383 (44.7) 47 (5.5) 246 (28.7) 181 (21.1) 336 (28.1) 89 (7.4) 528 (44.1) 244 (20.4) 
Sweden         
    Cases 208 (15.0) 209 (15.1) 314 (22.7) 653 (47.2) 173 (14.1) 103 (8.4) 330 (26.9) 621 (50.6) 
    Sub-cohort 251 (19.7) 255 (20.0) 271 (21.3) 497 (39.0) 371 (22.3) 238 (14.3) 465 (27.9) 590 (35.5) 
Denmark         
    Cases 247 (20.7) 90 (7.5) 362 (30.3) 496 (41.5) 57 (6.7) 71 (8.3) 396 (46.2) 333 (38.8) 
    Sub-cohort 332 (29.1) 96 (8.4) 326 (28.6) 386 (33.9) 102 (10.3) 112 (11.4) 460 (46.7) 311 (31.6) 
Table 2

Characteristics of sub-cohort individuals by relative index of inequality (RII) tertiles [means and standard deviations (SD) for continuous, and percentages and frequencies for categorical variables] in the InterAct project

 Sub-cohort
 
 RII tertile 1
 
RII tertile 2
 
RII tertile 3
 
 Mean/% SD/N Mean/% SD/N Mean/% SD/N 
Age (years) 52.1 9.2 51.2 8.8 54.3 9.0 
Sex (% men) 40.5 2419 33.6 1943 41.5 2019 
Body mass index [kg/m2] 25.5 4.0 26.2 4.34 27.1 4.5 
Marital status       
    Single (%) 11.3 440 8.9 315 7.5 222 
    Married (%) 75.3 2943 78.6 2779 76.2 2265 
    Separated (%) 9.1 354 8.5 300 9.9 295 
    Widowed (%) 4.4 172 4.0 142 6.4 189 
Smoking status       
    Never (%) 45.7 2729 47.5 2752 45.2 2197 
    Former (%) 29.8 1776 25.8 1493 25.5 1239 
    Current (%) 24.1 1436 26.1 1513 28.8 1400 
    Unknown (%) 0.4 26 0.5 32 0.5 23 
Alcohol       
    Never (%) 5.4 324 7.7 444 9.1 440 
    Former (%) 4.7 283 5.6 325 5.9 286 
    Current (%) 65.9 3931 67.5 3910 54.6 2654 
    Unknown (%) 23.9 1429 19.2 1111 30.4 1479 
Physical activity       
    Inactive (%) 22.3 1323 22.8 1316 27.5 1323 
    Mod. inactive (%) 36.7 2176 33.7 1939 29.4 1413 
    Mod. active (%) 23.5 1392 23.1 1328 20.9 1004 
    Active (%) 17.5 1039 20.4 1177 22.2 1068 
Nutrient intake       
    Total energy (kcal/d) 2165.6 615.4 2123.2 635.4 2125.1 662.5 
    Total proteins (g/d) 91.3 28.0 90.0 29.5 89.5 29.2 
    Total carbohydrates (g/d) 233.1 73.3 232.4 75.3 236.6 30.9 
    Total fats (g/d) 84.6 29.6 82.3 29.9 82.6 28.9 
 Sub-cohort
 
 RII tertile 1
 
RII tertile 2
 
RII tertile 3
 
 Mean/% SD/N Mean/% SD/N Mean/% SD/N 
Age (years) 52.1 9.2 51.2 8.8 54.3 9.0 
Sex (% men) 40.5 2419 33.6 1943 41.5 2019 
Body mass index [kg/m2] 25.5 4.0 26.2 4.34 27.1 4.5 
Marital status       
    Single (%) 11.3 440 8.9 315 7.5 222 
    Married (%) 75.3 2943 78.6 2779 76.2 2265 
    Separated (%) 9.1 354 8.5 300 9.9 295 
    Widowed (%) 4.4 172 4.0 142 6.4 189 
Smoking status       
    Never (%) 45.7 2729 47.5 2752 45.2 2197 
    Former (%) 29.8 1776 25.8 1493 25.5 1239 
    Current (%) 24.1 1436 26.1 1513 28.8 1400 
    Unknown (%) 0.4 26 0.5 32 0.5 23 
Alcohol       
    Never (%) 5.4 324 7.7 444 9.1 440 
    Former (%) 4.7 283 5.6 325 5.9 286 
    Current (%) 65.9 3931 67.5 3910 54.6 2654 
    Unknown (%) 23.9 1429 19.2 1111 30.4 1479 
Physical activity       
    Inactive (%) 22.3 1323 22.8 1316 27.5 1323 
    Mod. inactive (%) 36.7 2176 33.7 1939 29.4 1413 
    Mod. active (%) 23.5 1392 23.1 1328 20.9 1004 
    Active (%) 17.5 1039 20.4 1177 22.2 1068 
Nutrient intake       
    Total energy (kcal/d) 2165.6 615.4 2123.2 635.4 2125.1 662.5 
    Total proteins (g/d) 91.3 28.0 90.0 29.5 89.5 29.2 
    Total carbohydrates (g/d) 233.1 73.3 232.4 75.3 236.6 30.9 
    Total fats (g/d) 84.6 29.6 82.3 29.9 82.6 28.9 

In Figure 1 we show the a priori causal model (using the DAG) that we applied in the analyses, which reflects the current knowledge about the relationship between diabetes and SES.

In the present study, compared with the highest relative educational level, being in the lowest educational position was associated with a 64% increased HR for T2DM in men and 90% increased HR in women (adjusted HR: 1.64, 95% CI: 1.51–1.80 and 1.90, 95% CI: 1.75–2.07, respectively). The HRs adjusted for physical activity, smoking status and propensity score according to macronutrient intake were very similar to the crude HR (adjusted HR: 1.67, 95% CI: 1.52–1.83 in men; adjusted HR: 1.88, 95% CI: 1.73–2.05 in women). The HRs were attenuated only when they were further adjusted for BMI (BMI adjusted HR: 1.36, 95% CI: 1.23–1.51 in men; BMI adjusted HR: 1.32, 95% CI: 1.20–1.45 in women) (Table 3).

Table 3

Crude and adjusted hazard ratios (HR) and 95% Confidence Intervals comparing relative index of inequality (RII) tertiles and T2DM, stratified by sex

 Crude HR HR adjusted for physical activity, smoking, and macronutrients* HR adjusted for BMI, physical activity, smoking and macronutrients* 
MEN    
RII tertile 1 Ref. Ref. Ref. 
RII tertile 2 1.30 (1.19-1.43) 1.33 (1.21-1.46) 1.15 (1.03-1.28) 
RII tertile 3 1.64 (1.51-1.80) 1.67 (1.52-1.83) 1.36 (1.23-1.51) 
WOMEN    
RII tertile 1 Ref. Ref. Ref. 
RII tertile 2 1.34 (1.24-1.45) 1.35 (1.25-1.47) 1.16 (1.06-1.28) 
RII tertile 3 1.90 (1.75-2.07) 1.88 (1.73-2.05) 1.32 (1.20-1.45) 
OVERALL    
RII tertile 1 Ref. Ref. Ref. 
RII tertile 2 1.27 (1.22-1.33) 1.17 (1.04-1.45) 1.11 (1.03-1.19) 
RII tertile 3 1.77 (1.69-1.85) 1.76 (1.40-2.22) 1.30 (1.21-1.39) 
 Crude HR HR adjusted for physical activity, smoking, and macronutrients* HR adjusted for BMI, physical activity, smoking and macronutrients* 
MEN    
RII tertile 1 Ref. Ref. Ref. 
RII tertile 2 1.30 (1.19-1.43) 1.33 (1.21-1.46) 1.15 (1.03-1.28) 
RII tertile 3 1.64 (1.51-1.80) 1.67 (1.52-1.83) 1.36 (1.23-1.51) 
WOMEN    
RII tertile 1 Ref. Ref. Ref. 
RII tertile 2 1.34 (1.24-1.45) 1.35 (1.25-1.47) 1.16 (1.06-1.28) 
RII tertile 3 1.90 (1.75-2.07) 1.88 (1.73-2.05) 1.32 (1.20-1.45) 
OVERALL    
RII tertile 1 Ref. Ref. Ref. 
RII tertile 2 1.27 (1.22-1.33) 1.17 (1.04-1.45) 1.11 (1.03-1.19) 
RII tertile 3 1.77 (1.69-1.85) 1.76 (1.40-2.22) 1.30 (1.21-1.39) 

*Propensity score for macronutrient intake (total energy, total fats, total proteins and total carbohydrates). BMI: body mass index.

The combined results from the meta-analysis approach by country (Figure 2a and b) show that the association between educational level and the risk of T2DM was consistent across countries (P-value for heterogeneity = 0.44, I2 = 0.00, RII tertile 3 vs RII tertile 1). In the forest plot comparing RII tertile 3 and RII tertile 1, the only countries in which the effect of educational level on the risk of T2DM was less evident were France, Italy and the UK.

Figure 2

Country-specific hazard ratios of type 2 diabetes mellitus (a) comparing RII tertile 2 vs RII tertile 1 and (b) comparing RII tertile 3 vs RII tertile 1 adjusted for physical activity, smoking status, BMI and propensity score for macronutrients (total energy, total fats, total proteins and total carbohydrates)

Figure 2

Country-specific hazard ratios of type 2 diabetes mellitus (a) comparing RII tertile 2 vs RII tertile 1 and (b) comparing RII tertile 3 vs RII tertile 1 adjusted for physical activity, smoking status, BMI and propensity score for macronutrients (total energy, total fats, total proteins and total carbohydrates)

A sensitivity analysis was used to assess the potential influence of marital status (excluding from the analyses the two countries that did not record this information at baseline, n excluded cases = 4594) and early diagnosis of T2DM (excluding from analyses the 1044 incident cases of T2DM diagnosed within the first 2 years of follow-up). The results remained unchanged (data not shown).

Discussion

The results of this study provide evidence that lower educational level is associated with a higher risk of T2DM in men and women in Western European countries. Educational level (and SES in general) does not have a direct biological effect on disease; instead its effects are mediated by other risk factors that can be biologically related to disease (i.e. smoking status, BMI, physical activity). In the InterAct project, the association between lower educational level and higher risk of T2DM was not demonstrably mediated by any of the variables that we considered are usually regarded as modifiable behavioural risk factors. BMI was the only variable that partially explained the inequalities in the risk of T2DM due to differences in educational level.

There was a trend in the association between educational level and risk of T2DM in which lower educational levels were related to higher risks of T2DM incidence. This result supports the conclusion of a recent meta-analysis of T2DM and SES based on 20 cohort and case-control studies in high-income countries.27

In addition, in one of the more recent studies included in that meta-analysis, Robbins et al.21 studied the association between educational level, occupation, income and T2DM risk in 11 069 respondents to the First National Health and Nutrition Examination Survey (NHANES1). They found an association between all three indicators of SES and T2DM incidence in women, although it was attenuated when adjustment for potential confounders was performed, and a less consistent association in men. Similar results have been reported by Ross et al.26 on the association between educational level and T2DM incidence, which was more evident in women than in men in the adjusted models.

In the Whitehall II study on 10 308 civil servants, Kumari et al.22 found that employment grade was related to T2DM incidence only in men in a fully adjusted model. Finally, in a study of 6147 Alameda County residents in the USA, Maty et al.23 found that the excess risk associated with income and educational level was largely accounted for by obesity.

The main limitation of these previous studies and the subsequent meta-analysis was that the incidence of diabetes and the information on type of diabetes were mainly self-reported in the large cohort studies, and when diabetes was not self-reported, the sample sizes were often small. Furthermore, in a meta-analysis it is not always possible to adjust for potential confounders. In the InterAct project we overcame these problems by analysing data from a large case-cohort study in which the T2DM diagnosis was always validated, and in which, using detailed information about potential risk factors, we were able to build different multivariate models and identify the mediator variables that better explained inequalities.

In the present study we used relative educational level as a proxy for SES because it is easily and accurately reported and is unaffected by poor health in adulthood. Educational level reflects childhood and adolescent SES based on the SES of parents, as well as partially reflecting adult occupation, income and social prestige.49,50

Disparities in the SES of parents can be observed very early in childhood, through lower birthweights and an earlier adiposity rebound in the low SES group.15 Some,51,-54 but not all,25,55,61 studies have found an independent association between lower SES in childhood and increased risk of T2DM. Furthermore, a recent meta-analysis concluded that lower childhood SES is associated with T2DM and obesity later in life.50

In contrast, high educational level may be associated with health in different ways: individuals with a high educational level may be more receptive to prevention messages, have a higher ability to change their health behaviour and be prone to better use health care systems.31,62

Differences in educational systems in the distribution of women and older people across educational levels and in the legislation regarding the age at which students can be expelled from school do exist between the European countries included in the InterAct project. To overcome the difficulties in comparing educational levels, due to these differences, we computed the RII, which has been used frequently in the study of social inequalities in health.63,64 The purpose of this index in the present study was to quantify the effect of the RII score on T2DM risk. The RII expresses inequality within the whole socioeconomic continuum and can be interpreted as the ratio of the risk of T2DM between the high educational level (RII tertile 1) and the low educational level (RII tertile 3).

Socio-economic inequalities in health have been attributed to a number of different mechanisms, including unhealthy behaviours,65 fetal/infant malnutrition56,64 or adult nutritional inadequacies,50,57,66 depression or psychological stress,24,51,52,58 inadequate access to preventive health care and other inequalities in material circumstances.59,60

Adult BMI was examined as a mediating factor in the association between SES and T2DM, as were measures of physical activity, dietary caloric and macronutrient intakes and tobacco use.14 Obesity is a strong and well-documented risk factor for T2DM and has been shown to be related to SES.12,38,39 Previous research showed that the association of T2DM with SES was partially explained by BMI, particularly in women.26 In the present study, an association between BMI and RII was observed: men and women with higher BMI had low educational levels. Furthermore, as expected, the risk of incident T2DM for RII tertile 3 vs RII tertile 1 was attenuated after adjustment for BMI.

Physical inactivity12,38 is another important lifestyle factor that is related to SES and is a strong risk factor for T2DM. Ford et al.67 found that women with higher SES were physically more active than those with lower SES, though this difference was not apparent in men. In the present study there was no clear association between physical activity index and SES, and in the multivariate model the risk of incident T2DM for RII tertile 3 vs RII tertile 1 was not explained by physical inactivity.

The causal relationship between T2DM and smoking status is unclear,40 but smoking is known to be associated with lower SES.68,69 In our study there was evidence of a clear trend in smoking status in the RII tertiles. As a result, the multivariate model was adjusted for smoking status in both genders.

It is also unclear whether low SES results in unhealthy nutrition. Differences in dietary patterns, including fat intake and alcohol consumption, have been documented among socio-economic groups defined by income and educational level. Studies conducted in Canada showed that lower household income was inversely associated with fat intake.70 We did not see a clear relationship between educational level and nutrient intake in the present study.

Several investigators have observed differences by gender in the effects of SES on T2DM.9,-11,17,21,27 We did not find any evidence of a difference by gender in the association of SES with risk of T2DM in an adjusted model.

Finally, this study shows on-going social inequalities in the risk of T2DM, which are consistent across different European countries, even in those with a tradition of high social equality such as Sweden and Denmark.

We already listed some of the strengths of the present study, such as the possibility to study the relationship between diabetes and SES in more than one country and the availability of detailed information on many risk factors measured in a standardized manner across countries.

However, the limitations of the study also warrant discussion. First, the study population consisted predominantly of White men and women from Western Europe. Other ethnic groups, such as African or Hispanic groups, which in previous studies have been reported to have a much higher prevalence of diabetes, were not represented in our cohort. Furthermore, the EPIC study tends to recruit people belonging either to the lower strata of higher social classes or the highest strata of lower social classes, which is also the case in other large cohort studies. This phenomenon could imply a lower range of exposures and that observed contrasts may be less marked than the real ones. Finally, residual confounding caused by unmeasured variables such as psychological factors and stress is possible. Caution must be exerted in interpreting these results, as using only one indicator of SES may provide less information than using multiple indicators, such as occupation and income.29

In conclusion, based on the results of this study, it is possible to imply that there are substantial inequalities in the risk of T2DM in Western Europe, which are demonstrable in men and women alike. BMI is the only potential mediator in the relationship, explaining at least a part of the risk of T2DM due to differences in educational levels. The benefits of greater social equality tend to be largest among the least educated individuals in each country, but could extend to more than two-thirds of their respective populations. These results suggest that social programmes to improve the educational level of the general population could possibly reduce the risk of T2DM in Western Europe.

Funding

This work was supported by: J.W.J.B., IS: NL Agency [grant IGE05012] and Incentive Grant from the Board of the UMC Utrecht (The Netherlands); H.B.B.dM., A.M.W.S. and D.L.vd.A: Dutch Ministry of Public Health, Welfare and Sports (VWS), Netherlands Cancer Registry (NKR); LK Research Funds; Dutch Prevention Funds; Dutch ZON (Zorg Onderzoek Nederland); World Cancer Research Fund (WCRF); Statistics Netherlands (The Netherlands); F.L.C.: Cancer Research UK; P.W.F.: Swedish Research Council; Novo Nordisk; Swedish Heart Lung Foundation; Swedish Diabetes Association; J.H., K.O. and A.T.: Danish Cancer Society; R.K.: Deutsche Krebshilfe; SP, Ministero della Salute-regione Toscana Progetto Integrato Oncologia-PIO; Compagnia di San Paolo; J.R.Q.: Asturias Regional Government; M.T.: Health Research Fund (FIS) of the Spanish Ministry of Health; the CIBER en Epidemiología y Salud Pública (CIBERESP), Spain; Murcia Regional Government [No 6236]; R.T.: AIRE-ONLUS Ragusa; AVIS-Ragusa, Sicilian Regional Government and AIRC Italy. Funding for the InterAct project was provided by the EU FP6 programme [grant number Integrated Project LSHM_CT_2006_037197].

Acknowledgements

We thank all EPIC participants and staff for their contribution to the study. We thank Michael Joffe for his help in drawing and interpreting the causal model (DAG).

Conflict of interest: None declared.

KEY MESSAGES

  • Lower educational level is inversely associated with the risk of type 2 diabetes mellitus across different Western European countries, even in those with a history of social equality.

  • Only adjustment for body mass index, and not for other risk factors, partially reduced inequalities in the risk of type 2 diabetes mellitus due to differences in educational levels in men and women.

References

1
Brawley
OW
Berger
MZ
Cancer and disparities in health: perspectives on health statistics and research questions
Cancer
 , 
2008
, vol. 
113
 
Suppl. 7
(pg. 
1744
-
54
)
2
Frampton
GK
Shepherd
J
Dorne
JL
Demographic data in asthma clinical trials: a systematic review with implications for generalizing trial findings and tackling health disparities
Soc Sci Med
 , 
2009
, vol. 
69
 (pg. 
1147
-
54
)
3
Cox
AM
McKevitt
C
Rudd
AG
Wolfe
CD
Socioeconomic status and stroke
Lancet Neurol
 , 
2006
, vol. 
5
 (pg. 
181
-
88
)
4
Clark
AM
DesMeules
M
Luo
W
Duncan
AS
Wielgosz
A
Socioeconomic status and cardiovascular disease: risks and implications for care
Nat Rev Cardiol
 , 
2009
, vol. 
6
 (pg. 
712
-
22
)
5
Marmot
MG
Smith
GD
Stansfeld
S
, et al.  . 
Health inequalities among British civil servants: the Whitehall II study
Lancet
 , 
1991
, vol. 
337
 (pg. 
1387
-
93
)
6
Stern
MP
Haffner
SM
Body fat distribution and hyperinsulinemia as risk factors for diabetes and cardiovascular disease
Arteriosclerosis
 , 
1986
, vol. 
6
 (pg. 
123
-
30
)
7
Perry
IJ
Wannamethee
SG
Walker
MK
Thomson
AG
Whincup
PH
Shaper
AG
Prospective study of risk factors for development of non-insulin dependent diabetes in middle aged British men
BMJ
 , 
1995
, vol. 
310
 (pg. 
560
-
64
)
8
Millar
WJ
Wigle
DT
Socioeconomic disparities in risk factors for cardiovascular disease
CMAJ
 , 
1986
, vol. 
134
 (pg. 
127
-
32
)
9
Robbins
JM
Vaccarino
V
Zhang
H
Kasl
SV
Socioeconomic status and type 2 diabetes in African American and non-Hispanic white women and men: evidence from the Third National Health and Nutrition Examination Survey
Am J Public Health
 , 
2001
, vol. 
91
 (pg. 
76
-
83
)
10
Tang
M
Chen
Y
Krewski
D
Gender-related differences in the association between socioeconomic status and self-reported diabetes
Int J Epidemiol
 , 
2003
, vol. 
32
 (pg. 
381
-
85
)