The study was conducted as a frequency-matched case-control study. Patients with RA diagnosed within the previous 5 years were identified in rheumatology and internal medicine departments throughout Denmark, which has a predominantly Caucasian population of approximately 5.2 million inhabitants. To be included, patients had to be diagnosed with RA between ages 18 and 65 years and fulfill the American College of Rheumatology (ACR) 1987 classification criteria for RA 26 between August 1998 and July 2003. Information about date of diagnosis, defined as the date when the RA diagnosis was clinically confirmed by a rheumatologist, and cumulative fulfillment of the ACR 1987 classification criteria for RA was obtained from medical records by a rheumatologist at each department or by the project coordinator (MP) and a rheumatologist (MK) from the study team.

Controls who were frequency-matched by gender and birth year were randomly selected from the Danish population by means of the Civil Registration System, a national database that keeps track of all demographic changes in Denmark 27. Using identical invitation letters to cases and controls, we aimed at a 1:1 case-control ratio for women and a 1:2 case-control ratio for men, but all invited subjects who agreed to participate were included. A higher number of controls per case among men was chosen in order to enhance statistical power in analyses of RA in men.

The questionnaire was tested in a pilot experiment comprising 50 patients with RA and 50 controls whose data are not included in this report. Three trained female medical students carried out all interviews between September 2002 and February 2004. Bimonthly meetings were held to ensure that all interviews were conducted in a uniform manner. Interviews were conducted as computer-assisted telephone interviews, and answers were entered directly into a database. Logical tests were built into the program to keep data entry errors at a minimum. Each telephone interview took approximately half an hour and included questions about a wide range of exposure and confounder variables, including level of education, age at menarche, parity, spontaneous abortions, breastfeeding, age at menopause, use of oral contraceptive pills and hormone replacement therapy, marital status, lifetime number of sexual partners of opposite sex, age at first sexual intercourse, lifetime number of anal-intercourse partners of opposite sex, lifetime number of same-sex sexual partners (male study participants only), lifetime number of prostitute visits (male study participants only), histories of venereal diseases (including chlamydia, genital herpes, acuminate condylomas, gonorrhea, and syphilis), smoker status, pack-years smoked (one pack-year equivalent to 20 cigarettes per day for 1 year with one cigarette equivalent to 1 g, one cheroot to 3 g, and one cigar to 4 g of tobacco), coffee, alcohol, and wine consumption 10 years before interview, frequency of fish intake as a hot meal or on bread (at least once a week, 1–3 times a month, less than once a month) 10 years before interview, intake of fish oil (ever/never), vegetarian diet (ever/never), body mass index at age 20 years and 10 years before interview, level of physical activity at work and during leisure time 10 years before interview, pets in childhood and in adulthood, histories of mononucleosis, hay fever, atopic dermatitis, asthma before age 45 years, stomach or duodenal ulcer, heavy diarrhea of at least 4 days' duration, type I diabetes, thyroid disease, periodontal disease, urinary tract infection, cancer, blood transfusion, tonsillectomy, adenoidectomy, appendectomy, splenectomy, and schizophrenia among first-degree relatives.

Blood samples were collected at rheumatology departments (patients) or by general practitioners (patients and controls), and serum was stored at -20°C. Anti-CCP immunoglobulin (Ig) G antibodies were determined by a second-generation enzyme-linked immunosorbent assay (ELISA) using the Immunoscan RA kit (Euro-Diagnostica AB, Malmö, Sweden), and parvovirus B19 IgG antibodies were determined by ELISA using the Biotrin Parvovirus B19 IgG kit (Dako Denmark A/S, Glostrup, Denmark) according to instructions provided by the manufacturers. Levels of IgG antibodies to Epstein-Barr viral capsid antigen (VCA) were determined in arbitrary units by ELISA using the Biotest anti-EBV VCA IgG kit (Meda A/S, Allerød, Denmark). A standard pool of serum was given the value of 100 arbitrary units. All samples were diluted 1:40 and tested together with dilutions (1:20, 1:40, 1:80, and 1:160) of the standard. Samples that were negative at 1:40 were retested undiluted together with dilutions of the standard. This enabled us to convert optical density values for the study samples into arbitrary units.

To make exposure information comparable for patients and controls, a pseudo-year of diagnosis was attributed to controls according to the frequency distribution of year of RA diagnosis in patients of the same sex. Throughout, we disregarded information about exposures after the year (patients) or pseudo-year (controls) of diagnosis. We performed logistic regression analyses for men and women separately to study associations of exposure variables with overall RA risk regardless of RA subtype. For each exposure variable, we also combined information for women and men when there was no significant interaction with gender (likelihood ratio test, p > 0.05). In these analyses, we adjusted only for birth year, for year or pseudo-year of diagnosis, and (in combined analyses) for gender.

Risk factors for anti-CCP-positive and anti-CCP-negative RA were identified in a polytomous logistic regression model using these two groups of patients and the control group as the dependent variable. This model included gender, birth year, year or pseudo-year of diagnosis, place of residence (five categories), educational status (four categories), and exposure variables identified in a three-step selection process. (a) In the first step, exposure variables with a p value less than 0.05 in the gender-combined analyses described above were included in the final polytomous logistic regression model. (b) In the second step, we performed polytomous logistic regression to study the associations with risk of anti-CCP-positive and anti-CCP-negative RA for exposure variables with a p value less than 0.20 in the gender-combined analyses from the first step. The rather broad screening criterion of p less than 0.20 was used to ensure inclusion of potential RA subtype-specific risk factors for which associations might be blurred in subtype-unrestricted analyses. Adjustment was made for gender, birth year, and year or pseudo-year of RA diagnosis. Exposure variables with a p value less than 0.05 for at least one of the RA subtypes were included in the final multifactorial model. (c) In the third step, we considered a polytomous logistic regression model that included gender, birth year, year or pseudo-year of diagnosis, place of residence, educational status, and the exposure variables identified in steps one and two. Using the method of forward inclusion, we introduced in this model, one by one, all variables with a p value between 0.05 and 0.20 in at least one of the initial subtype-specific analyses in step two or the gender-combined analyses in step one. Introduced variables with a p value less than 0.05 for at least one RA subtype were included in the final model.

Tests for difference between exposure categories were evaluated by either trend tests or tests for homogeneity, as relevant. Trend tests were performed by treating categorical variables as continuous variables in the regression analyses. In trend tests for ordered categorical variables, representing an underlying continuous variable (for example, coffee consumption), categories were assigned the category median of the underlying variable. In trend tests for ordered categorical variables with no well-defined underlying continuous variable (for example, physical activity), the categories were assigned consecutive numbers starting with one. Tests for difference between risk factor associations with anti-CCP-positive and anti-CCP-negative RA were evaluated by polytomous logistic regression by comparing either subtype-specific trends or subtype-specific categorical variables, as relevant. All tests were likelihood-ratio-based. All analyses were carried out using SAS software (PROC LOGISTIC procedure in SAS 9.1; SAS Institute Inc., Cary, NC, USA). Throughout, we considered two-sided p values < 0.05 as statistically significant.

The study was approved by the Scientific Ethical Committees for Copenhagen and Frederiksberg (J. no. KF 01-039/01) and the Danish Data Protection Agency (2001-41-0658).