The Etude Epidémiologique de femmes de la Mutuelle Générale de l´Education (E3N) (32) is a French prospective cohort started in 1990 comprising 98,995 women aged 40-65 years at baseline and insured by the MGEN (Mutuelle Générale de l´Education Nationale), a health insurance plan for workers in the National Education System and their families. The objective of E3N was to study the main risk factors of cancer and chronic diseases. The E3N is the French component of the European Prospective Investigation into Cancer and Nutrition. The cohort received ethical approval from the French National Commission for Computerized Data and Individual Freedom (Commission Nationale Informatique et Libertés), and all participants in the study signed an informed consent.
Participants returned mailed questionnaires on lifestyle information and disease occurrence every 2 to 3 years. The average response rate at each questionnaire cycle was 83 %, and the total loss to follow-up was 3%.
From the 74,522 women who responded to a dietary questionnaire in 1993 we excluded those women with prevalent hypertension, coronary disease or stroke (n = 26,974) before or at the 1993 questionnaire, and those with unrealistic energy consumption (the 1st and 99th percentiles of the distribution in the population, n = 896). The final study population included 46,652 women.
Assessment of the Dietary Inflammatory Index
In 1993 dietary data was collected using a two-part questionnaire detailing consumption of 208 food items during the year prior to the questionnaire, which has been shown to be valid and reproducible (33). Women were asked to answer questions about quantities and frequencies of consumption of food groups. Eleven possible responses were available, never or less than once a month; 1 to 3 times a month, and 1 to 7 times a week. A photo booklet was added to help estimate portion sizes (34). From this questionnaire and using a detailed food composition table, mean daily intakes of energy (excluding energy from alcohol), alcohol, and nutrients were estimated.
The adapted DII was estimated as previously described (21). Briefly, the adapted dietary inflammatory index proposed by Woudenberg et al (19) was used in combination with the updated dietary components weights by Shivappa et al (35) instead of the weights proposed by Cavicchia et al (36). This DII has been proposed on the basis of nutritional rationale. First, the inflammatory weights of dietary components are multiplied by the standardised energy adjusted intake, which acts to reduce between-person variation. Second, the intake of all components are standardised by subtracting the mean intake of the population (in this case E3N, n = 74,522) from the individual’s intake, and then divided by the standard deviation of the intake from the population. Finally, the DII calculation did not include alcohol, total fat, and energy, in order to avoid estimating of the inflammatory effects of ethanol, fat and energy twice. A total of 30 of the 37 possible dietary components were used for DII calculation (see the supplementary data) based on the food frequency questionnaire. A positive DII score is representative of a pro-inflammatory diet, and negative values of an anti-inflammatory diet.
Participants were asked to report whether they had hypertension at baseline (1993) and in each follow-up questionnaire (1994, 1997, 2000, 2002, 2005, 2008, 2011, and 2014), the date of diagnosis, and the use of antihypertensive treatments. The month and year of diagnosis were provided for most cases (69%). For individuals who were missing the month of diagnosis (14% of cases), it was imputed to June of the year of diagnosis. The median time between the date of diagnosis and the date of response to the first questionnaire after diagnosis was 12 months. Thus, for the cases with no year of diagnosis (n=17%), we assigned it to be 12 months before they reported hypertension in a questionnaire. In 2004, a drug reimbursement database became available for 97.6 % of participants. We used the self-reported date of diagnosis or the first date of drug reimbursement for antihypertensive medications (Anatomical Therapeutic Chemical Classification System codes C02, C03, C07, C08, and C09) whatever happened first, as the date of diagnosis for cases identified after 2004.
In addition, using the information of the MGEN health insurance plan drug claim database, we assessed the validity of self-reported hypertension within the E3N cohort. We compared hypertension self-report to antihypertensive drug reimbursement (any of the above specified codes). A positive predictive value of 82% was observed (37).
Assessment of covariates
Family history of hypertension, education (no high school diploma, high school diploma), and smoking (ever smoker, current smoker, or never smoker) were based on self-reports, and for diabetes and treated dyslipidaemia we used cases which had been validated through the use of a drug reimbursement database (38). A Mediterranean/prudent diet score was determined from dietary data using principal component analysis, as previously described (39).
We assessed usual physical activity with a questionnaire in 1993 that included questions on weekly hours spent walking, cycling and performing light and heavy household chores, and questions on recreational activities and sports (e.g., swimming and tennis) considering the winter and summer seasons. It included questions on the time spent walking (to work, shopping, and leisure time), cycling (to work, shopping, and leisure time), housework, and sports activities (such as racket sports, swimming). Metabolic equivalents (METs) per week were estimated by multiplying the hourly average METs for each item based on values from the Compendium of Physical Activities (40) by the reported activity duration.
Self-reported height and weight at baseline were used to calculate body mass index (BMI), defined as weight (kg) divided by squared height (m2). In the cohort, self-reported anthropometry is considered reliable from a validation study (41).
Participants were split into quintiles depending on DII score depending on the population distribution. Characteristics between participants were tabulated (see table 1) depending on their quintile of DII. Correlations between variables were assessed using Pearson’s R.
Hazards ratios and 95% confidence intervals were estimated from Cox regression models with age as the time scale. Time at entry was the age at the beginning of follow-up (1993), exit time was the age when participants were diagnosed with hypertension, died (dates of death were obtained from the participants’ medical insurance records), were lost to follow-up, or were censored at the end of the follow-up period (June 15, 2014), whichever occurred first. P-values for trends were calculated using the median category value as a semi-continuous variable in the models.
Four models assessing DII as the exposure were assessed; the first was controlled for age as the timeline (Model 0). Multivariable models were first adjusted for various known risk-factors for hypertension: total physical activity (MET-hours/week, continuous), family history of cardiovascular disease (yes/no), smoking (never, former, and current at baseline), education (no high school diploma, high school diploma) (Model 1), diabetes and dyslipidaemia status (Model 2), and finally BMI (kg/m2, continuous) (Model 3). As dyslipidaemia and diabetes may affect both exposure and outcome, models with and without these variables were assessed, but no difference in estimates was obtained, thus the variables were retained.
Spline regression with 5 degrees of freedom was used to assess the dose-response relationship between DII and the risk of incident hypertension. Tests for interactions (ANOVA) were performed for BMI as a continuous variable to determine if it was an effect modifier. As a hypothesis generating exercise, this was also done for smoking status (never, former, and current at baseline), and Mediterranean diet score (continuous). If tests were indicative of effect modification, models were stratified on this variable.
Missing values (less than 5 % of participants) were imputed using the mean for continuous, or median for categorical variables. All statistical analyses used R version 3.5.1 (www.r-project.org) and the survival package (www.github.com/therneau/survival), with an alpha of statistical significance equal to 0.05. Results from Cox-models were interpreted as hazard ratios (HR) (95% confidence interval (CI)). The proportional hazards assumption was assessed by plotting the Schoenfeld residuals using the cox.zph package in R.
Several sensitivity analyses were performed. A model with DII categorised dichotomously as negative (anti-inflammatory) or positive (pro-inflammatory) was assessed, in the same method as previously described. We also calculated the DII using only the women in the final study cohort as reference (n = 46,652) when the mean intake of the population was subtracted, as opposed to the 74,522 women of the considered DII score. Similarly, an alternative DII score which takes into account the inflammatory potential from total energy intake was assessed in a sensitivity analysis (“non-adapted” DII, method of Shivappa et al (35)). Results for these scores are presented in the supplementary data. In order to account for reverse causation, we excluded cases diagnosed within 5 and then within 10 years.