Study population
This study included subjects from the Swedish epidemiological investigation of RA (EIRA), a population-based case-control study, from defined geographical areas of central Sweden. The EIRA study was initiated in 1996 and is still ongoing. The general study design has been described in detail elsewhere [15]. Briefly, the study includes patients with incident RA (cases) (mean time between symptom onset and diagnosis 9.7 months) diagnosed by a rheumatologist according to the either the American College of Rheumatology (ACR) 1987 criteria or the 2010 ACR/European League Against Rheumatism (EULAR) classification criteria for RA. To each patient two controls were randomly selected from the register of the general population, in connection to inclusion of the patient, matched on age, sex and residential area.
A baseline questionnaire was distributed to patients and controls with comprehensive questions on lifestyle factors, educational level and comorbidities. From November 2005 the EIRA questionnaire also included a 124-item food frequency questionnaire (FFQ). The current study therefore included participants from November 2005 to September 2014.Participants with incomplete an FFQ where excluded (n = 21).
Participants provided written informed consent, and ethical approval was obtained from the Regional Ethical Review Board at Karolinska Institutet, Stockholm, Sweden.
Exposure definition
The self-reported FFQ was used to evaluate the participants’ food intake during the last year before inclusion. In the FFQ participants were asked to indicate how often on average they had consumed various foods by using eight predefined frequency categories ranging from “never or less than once per month” to “three or more times per day”. Fat and energy intake was calculated by multiplying the average frequency of consumption of each food by the fat and energy content, respectively, of age-specific portion sizes, according to the Swedish National Food Administration Database [16].
The Mediterranean diet was defined according to the Mediterranean diet score based on the work of Trichopoulou et al. [17]. The score range between 0 and 9 and each person was assigned 1 point if she/he consumed more than the sex-specific median of the population of the beneficial components, such as vegetables, legumes, fruits and nuts, cereal and fish, or if she/he consumed less than the median of red meat and dairy products, which were considered detrimental components. Moreover, 1 point was assigned to men who consumed between 10 g and 50 g of alcohol per day and to women who consumed between 5 g and 25 g per day, and 1 point if the ratio of monounsaturated lipids to saturated lipids was above the median. Hence, a higher score corresponded to higher adherence to the Mediterranean diet. We did not include butter among the dairy products since it was reported in the FFQ only as “ever consumption”. In sensitivity analysis, we considered ever consumption of butter as consumption once per day and added it to the dairy products group.
Modifications to the original Mediterranean score have been proposed throughout the years [18, 19]. In sensitivity analyses, we defined the Mediterranean diet according to the alternate Mediterranean diet score (aMed), which not include dairy products in the score [13]. Moreover, we used a lower upper limit for alcohol intake among men (30 g instead of 50 g per day), according to the modified Mediterranean diet (mMED) score [19].
Covariate definition
All covariate data were collected at baseline. Smoking was categorised as never, current, former or irregular smoker. Body mass index (BMI) was categorised as underweight (< 18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25–29.9 kg/m2) or obese (≥ 30 kg/m2). Education was categorised as <10 years, 10–12 years, and > 12 years of education. Physical activity was reported as sedentary, moderate, moderate/regular, exercise/regular. Nutritional supplements were defined as intake in the last 5 years of at least one of the following supplements: multivitamins (with or without minerals), vitamin C, vitamin E, vitamin B, vitamin A, vitamin D, calcium, zinc, iron, magnesium, folic acid, β-caroten or omega 3. Additional covariates were considered when analysing women only: parity (quartiles), age at menarche (quartiles), use of oral contraceptive (yes/no) and use of hormone replacement therapy (yes/no). Genotyping of the HLA-DRB1 gene was conducted as previously described [20, 21]. Among HLA-DRB1 genes, DRB1*01, DRB1*04 and DRB1*10 genes were defined as SE alleles. Any genotype containing one or two of these genes was considered as having “any SE allele” versus those not having any of the genes (“no SE alleles”).
Statistical analyses
The Mediterranean diet score was analysed both as a continuous and as a categorical variable, the latter according to the quartiles of the distribution. Odds ratios (ORs) and their 95% confidence intervals (CIs) were calculated using conditional logistic regression, and multivariable models were adjusted for smoking, BMI, educational level, physical activity, use of dietary supplements, and energy intake. Analyses were stratified by gender, rheumatoid factor (RF) and anti-citrullinated protein antibody (ACPA) status. We additionally stratified by smoking status and presence of at least one copy of the HLA-DRB1 shared epitope allele. In sensitivity analyses, we excluded persons with extreme energy intake (i.e. 3 standard errors from the mean value on the log-transformed scale, n = 49). Moreover, to evaluate the influence of each single item, we analysed data excluding one component of the score at a time.
We estimated the dose–response trend association between the Mediterranean diet score and risk of RA using restricted cubic splines with knots at 10, 50 and 90 centiles of the Mediterranean score (e.g. at 2, 4, and 7) [22].
To analyse the association in a group of patients presumed to have more homogeneous disease aetiology, we restricted the analyses to patients and controls with at least one copy of the HLA-DRB1 shared epitope allele (SE) and who were also current smokers. For this analysis, since patients and controls were no longer matched, we performed logistic regression adjusted for age, gender and residential area, and then additionally adjusted for the confounding variables listed above. Statistical analyses were implemented using SAS (V.9.4) and Stata (V.13.1), and p values ≤0.05 were considered significant.
Patient involvement
There was no patient involvement in the current study.