The participants were 20
040 men and women aged 40-79 at baseline, drawn from the Norfolk component of the European Prospective Investigation of Cancer (EPIC-Norfolk). This prospective population study first surveyed participants in 1993-7, 99.5% of whom were white. The detailed recruitment strategy and sample distribution of the whole cohort have been previously described.14
Briefly, our participants were recruited from age-sex registers of general practices. As nearly all people in the United Kingdom are registered with general practitioners through the National Health Service, the age-sex registers form a population based sampling frame. From the inception of the EPIC-Norfolk cohort, data collection was broadened to enable the examination of a wider range of determinants of chronic diseases, including stroke. The Norfolk cohort was comparable with national population samples with respect to characteristics including anthropometry, blood pressure, and lipids but with a lower prevalence of current smokers.14
At the 1993-7 baseline survey, participants completed a detailed health and lifestyle questionnaire. They were asked about medical history with the question “Has a doctor ever told you that you have any of the following?” followed by a list of conditions that included heart attack, stroke, diabetes, and cancer. Smoking history was derived from yes/no responses to the questions “Have you ever smoked as much as one cigarette a day for as long as a year?” and “Do you smoke cigarettes now?” Alcohol consumption was derived from the question “How many alcoholic drinks do you have each week?” with four separate categories of drinks. A unit of alcohol (about 8 g) was defined as a half pint (about 0.2 l) of beer, cider, or lager; a glass of wine; a single unit of spirits (whisky, gin, brandy, or vodka); or a glass of sherry, port, vermouth, or liqueurs. Total alcohol consumption was estimated as the total units of drinks consumed in a week. For these analyses, a moderate drinker was defined as someone who drank one or more units a week (that is, not a non-drinker) but not more than 14 units a week.
Aspirin use was ascertained by a question “Have you taken aspirin continuously for three months or more?” Habitual physical activity was assessed with two questions referring to activity during the past year. The detailed description of this physical activity questionnaire has been previously reported.12 13
This index was validated against heart rate monitoring with individual calibration in two independent studies.15 16
For the purposes of the current study, we dichotomised the population into physically inactive (sedentary job and no recreational activity) and physically not inactive (any category with activity levels above the latter).
We classified social class according to the registrar general’s occupation based classification scheme into five main categories, with social class I representing professionals, social class II managerial and technical occupations, social class III subdivided into non-manual and manual skilled workers, social class IV partly skilled workers, and social class V unskilled manual workers. For men, social class was coded with their current occupation at the time of survey, except when they were unemployed in which case their partner’s social class was used. For women it was based on their partner’s occupation unless the partner’s social class was unclassified or missing. If they had no partner, it was based on their own occupation. Unemployed men and women without a partner were coded as unclassified and excluded in the current study.
We also re-categorised social class into manual (III manual, IV and V) and non-manual (I, II and III non-manual) social classes.17
Trained nurses carried out a health examination at a clinic. They measured height and weight with participants in light clothing without shoes and calculated the body mass index (weight (kg)/(height (m)2). Blood pressure was measured using an Accutorr sphygmomanometer after each participant had been seated for five minutes. We used the mean of two measurements of blood pressure in analysis. At the baseline clinic visit nurses also took non-fasting blood samples in plain and citrate bottles. After overnight storage in a dark box in a refrigerator at 4-7°C, they were spun at 2100 g for 15 minutes at 4°C, and plasma and serum samples obtained. Serum concentrations of total cholesterol, high density lipoprotein cholesterol, and triglycerides were measured on fresh samples with the RA 1000 (Bayer Diagnostics, Basingstoke).
Six months after the start of the study, when funding became available, we also collected samples to measure vitamin C concentration. Plasma vitamin C was measured from blood drawn into citrate bottles and stabilised in a standardised volume of metaphosphoric acid stored at −70°C. Plasma vitamin C concentration was estimated with a fluorometric assay within a week of sampling.18
The coefficient of variation was 5.6% at the lower end of the range (mean 33.2 μmol/l) and 4.6% at the upper end (mean 102.3 μmol/l). We have previously reported that high plasma vitamin C concentration is inversely associated with mortality from all causes. Because humans do not manufacture vitamin C and have to rely on exogenous sources, plasma vitamin C is a good biomarker of plant food intake; previous studies have reported that a blood value of 50 mmol/l or more indicates an intake of at least five servings of fruit and vegetables daily.19 20
We therefore used plasma vitamin C concentrations as an objective biomarker of fruit and vegetable intake.
We constructed a simple pragmatic health behaviour score (box).13
Participants scored one point for each of the following health behaviours: current non-smoking, physically not inactive, moderate alcohol intake (1-14 units a week), and plasma vitamin C concentration ≥50 µmol/l, indicating an intake of fruit and vegetables of at least five servings a day. Participants could therefore have a total health behaviour score ranging from 0 to 4. We chose these particular health behaviours and their categorisation on the basis of extensive previous evidence on the relation between these lifestyle factors and health end points.
Health behaviour score
- Not inactive=1—that is, if a person has a sedentary occupation, at least half an hour of leisure time activity a day, such as cycling or swimming; or else a non-sedentary occupation with or without leisure time activity
- One or more but <14 units/week=1; 1 unit=about 8 g alcohol—that is, one glass of wine, one small glass of sherry, one single shot of spirits, or one half pint (about 0.2 l) of beer
Fruit and vegetable intake
- Five servings or more as indicated by blood concentration of vitamin C ≥50 µmol/l=1
We ascertained incident cases of stroke using death certificate data and hospital record linkage. All participants were flagged for death at the UK Office of National Statistics (ONS), and trained nosologists coded death certificates using the international classification of disease (ICD), revisions 9 and 10. Participants are also linked to NHS hospital information systems so that admission anywhere in the UK are notified to EPIC-Norfolk through routine annual record linkage. Stroke death was defined as ICD-9 codes 430-438 or ICD-10 codes 60-69 anywhere on the death certificate. Incident stroke was defined as death from stroke or hospital discharge code ICD-9 codes 430-438 or ICD-10 codes 60-69 for the first ever stroke. The current study is based on follow-up to end March 2007. A separate validation study showed this method for stroke ascertainment had high positive predictive value of 94% (unpublished data). The follow-up period was defined as time interval between the date of the health examination at enrolment to the date of death for those who died, the date of first stroke for those who had a stroke, and the end of follow-up (31 March 2007) for the remaining participants.
We used SPSS for Windows version 14.0 (SPSS, IL, USA) for statistical analyses. We excluded participants with a history of stroke and myocardial infarction at baseline (n=913) and those who had any missing values for the variables included in the study (n=9492). The missing values are mainly because out of 30
445 participants who provided the baseline data, only 25
633 attended the health check where we measured cardiovascular disease risk factors and obtained blood samples for measurement of vitamin C concentration. We included only participants with all available data for all the covariates in the models.
We used Cox proportional hazards models to determine the associations between health behaviours, either individually or as their combined score, and the risk of incident stroke during the follow-up. Multivariate Cox regression models were constructed for health behaviour scores (0-4) with the highest score category (4) as the reference category.
We made multivariate adjustments to examine how far the effect of health behaviours might be explained by known cardiovascular risk factors. We adjusted for age (and sex in the combined model) in model A; age (sex), body mass index, systolic blood pressure, cholesterol concentration, aspirin use, and history of diabetes mellitus in model B; and as for model B with the addition of social class in model C.
To address the issue of reverse causality—that is, when people with subclinical chronic disease might be likely to change their lifestyle, such as reducing their physical activity—we excluded all those who had stroke within the first two years of follow-up and constructed model D controlling for all of the above mentioned variables. We also performed stratified analyses by sex, age category (<65 and ≥65), body mass index (<25, 25-30, ≥30), and social class (non-manual and manual).