Longitudinal data were used from the GLOBE study conducted in the Netherlands (these data are available from the first author upon request). Detailed information about the study design and sampling methods are provided elsewhere [20, 21]. In 1991, a random sample of 27,027 non-institutionalised Dutch persons aged 15–75 years living in the city of Eindhoven and its surrounding area was drawn from the municipal population register. This sample was sent a postal questionnaire (response rate 70.1%, N=18,793). All measures described below were self-reported in the baseline questionnaire.
Adulthood and childhood socioeconomic position
In this study, all ‘adulthood’ measures (i.e. adulthood socioeconomic position as well as adulthood risk factors) pertain to people age 40 years or older (since the analyses were restricted to respondents in this age range at baseline, see below), and ‘childhood’ refers to the age of 12 years. Adulthood socioeconomic position (SEP) was determined by the respondents’ highest attained education level, with four categories: 1-low (primary education), 2- (lower professional and intermediate general education), 3- (intermediate professional and higher general education), 4-high (higher professional education and university) .
Participants were asked to retrospectively recall the occupational title of their father when they were twelve years of age, or, if their father was unemployed, the title of his last occupation (childhood socioeconomic conditions). These data were classified according to the Erikson, Goldthorpe, and Portocarero scheme,  and three categories were created: professionals (top-level management, advanced academic competencies, high level of independence), white-collar (middle management, routine non-manual work), and blue-collar occupations (skilled and unskilled manual work).
Smoking status was categorised as never, former and current smoker based on the response to the question “Do you smoke?” [12, 13].
Physical activity was based on three questions, asking for time spent per week (never, <1 hour, 1–2 hour, 2≥ hour; analysed as 0 hour, 0.5 hour, 1.5 hour, and 2.5 hour) on transport-related activity (walking, cycling), leisure time physical activity (gardening, walking, cycling), and sports activity . Time spent on transport and leisure activity was summed as ‘moderate physical activity’. Participants were classified as inactive (no sports and 0–1 hours of moderate physical activity), little active (either no sports and 1–2 hours of moderate physical activity, or <1 hours of sports and 0–1 hours of moderate physical activity), moderately active (2.5-3.5 hours of moderate physical activity and sports combined), or active (at least 3,5 hours of sports or moderate physical activity combined) .
Alcohol consumption was calculated from two questions, one asking for the number of days per week drinking any alcoholic drinks, and the second asking for the number of alcohol drinks (units) consumed on such a day [12, 13]. Participants were categorised as abstainers (0 units/week), light drinkers (1–7 units for women, 1–10 units for men), moderate drinkers (8–14 units for women, 11–21 units for men), and heavy drinkers (>14 units for women, >21 units for men) [8, 24].
Body mass index (BMI) was calculated from self-reported weight in kilograms/self-reported height in meters2, and respondents were classified as being underweight (BMI ≤20), average weight (BMI 20–25), overweight (BMI 25–30), or obese (BMI 30≥) .
Four items that are indicators of the financial situation of the household were measured and have been applied in several studies among the GLOBE cohort: [12, 13]type of health insurance (private, public), car ownership (yes, no), housing tenure (rented house, house owner), and financial problems with paying bills for food, rent, electricity etc. over the preceding year (no, some, big problems) [12, 13]. Adverse neighbourhood conditions were measured by four questions about noise from neighbours, noise from traffic, smells, and vandalism in the neighbourhood (no, 1≥ adverse conditions) [12, 13]. Adverse housing conditions were measured by three questions on cold, mould, and dampness in the house (no, 1≥ adverse conditions) [12, 13].
Data describing psychosocial factors included indicators of marital status (married, single, divorced, widowed) and negative life events. Respondents were asked if they experienced each of nine negative life events in the preceding year, such as a decline in financial position, severe disease of partner, and divorce (no event, 1 event, 2≥ events) [13, 26]. Furthermore, use of medicine for anxiety (yes, no) and whether respondents had experienced depression, severe nervousness or burn-out over the last five years (yes, no) were applied as psychosocial indicators .
CVD mortality and data linkage
Cause-specific mortality data were obtained from Statistics Netherlands. Causes of death were coded in accordance with the 9th and 10th version of the International Classification of Diseases (ICD), with codes 390–459 (ICD 9) or I00-I99 (ICD 10) for cardiovascular diseases. For each GLOBE respondent, the mortality follow-up extended from the baseline survey (April 1st, 1991) until October 15, 2007. If respondents died in the follow-up period, their death date was used to calculate survival time. If they did not die, survival time was calculated with October 15, 2007 as the final date. If respondents moved out of the Netherlands between baseline and October 15, 2007, survival time was calculated from the baseline until the date they emigrated.
Baseline respondents were excluded from the current analyses if they were 1) younger than 40 years of age at baseline (2 946 men, 2 902 women), 2) reported in the baseline questionnaire that they had experienced severe heart problems or a heart attack (599 men, 314 women) or a stroke (100 men, 48 women) in the preceding five years, or 3) had a missing value for adulthood SEP (199 men and 234 women) (these categories partly overlapped). The current analyses were based on the remaining 11 701 participants (5 395 men and 6 306 women).
Analyses were performed in SPSS , the significance level used was .05, and all analyses were adjusted for age. Analyses were undertaken for men and women separately, since the socioeconomic distribution of risk factors and their relative importance for explaining CVD inequalities may differ for men and women . For each of the risk factors and for childhood socioeconomic conditions, respondents with a missing value remained in the analyses as a separate category (see Additional file 1 for prevalence rates of missing values).
Using Cox proportional hazard models, we assessed associations between adulthood SEP and CVD mortality. We assessed the distribution of childhood socioeconomic conditions and adulthood risk factors by adulthood SEP using Chi-square tests, and associations of childhood socioeconomic conditions with CVD mortality (adjusted for age), and adulthood risk factors with CVD mortality (adjusted for age, childhood socioeconomic conditions and adulthood SEP) by Cox proportional hazard models.
Factors that were significantly related to cardiovascular mortality and that varied by adulthood SEP were included in the following models: 1) adulthood SEP; 2) adulthood SEP + childhood socioeconomic conditions; 3) adulthood SEP + material factors; 4) adulthood SEP + behavioural factors; 5) adulthood SEP + psychosocial factors; 6) adulthood SEP + all adulthood risk factors; 7) adulthood SEP + childhood socioeconomic conditions + all adulthood risk factors. For each model, the percent change in relative hazards for SEP-groups compared to model 1 was evaluated. A 95% CI was calculated around the percentage attenuation using a bias-corrected accelerated bootstrap method with 1000 re-samplings in the statistical program R .