Design and study population

The present report was based on data from two case–control studies of environmental and genetic risk factors for MS. The first study is EIMS (Epidemiological Investigation of Multiple Sclerosis) with a study base comprising the Swedish population aged 16–70 years. Potential incident cases of MS, including those with clinically isolated syndrome, were recruited via 40 study centres, including all university hospitals in Sweden. All cases were examined and diagnosed by a neurologist located at the unit where the case was entered. The treating neurologist provided information regarding the patient’s onset of disease, and whether the patient fulfilled the McDonald criteria.14 For each potential case, two controls were randomly selected from the national population register, matched by age in 5-year strata, sex and residential area. Only cases that fulfilled the McDonald criteria were included in the analyses. Cases that did not fulfil the criteria at the time of this report were excluded, but not their corresponding controls. All aspects of the study were approved by the Regional Ethical Review Board at Karolinska Institutet.

The other case–control study recruited prevalent cases identified among members of Kaiser Permanente Medical Care Plan, Northern California Region (KPNC) using electronic health records. KPNC is an integrated health services delivery system with a membership of 3.3 million that comprises about 25–30% of the population of a 22-county service area in northern California. Cases of MS, aged 18 years through 69 years, were required to be current KPNC members who had received the MS diagnosis by a neurologist. Diagnoses were validated by chart review and radiology and pharmacy records, according to McDonald criteria.15 The treating neurologist of each potential MS case was contacted to gain approval regarding contacting the case; potential cases who did not have MS, were deemed unable to participate due to being too severely ill or impaired, or who were no longer KPNC members, were thus excluded by their neurologists. Controls were randomly selected from KPNC members; they did not have a diagnosis of MS or related conditions and were individually matched to cases on sex, birth date, race/ethnicity and zip code of the case residence. The study protocol was approved by the Institutional Review Boards of the KP Division of Research and the University of California, Berkeley. Details of the research resource have been described elsewhere.15

Data collection

In EIMS, information on lifestyle factors and different exposures was collected using a standardised questionnaire given to the cases shortly after they had received their diagnosis, and was sent by mail to the controls. During the study period (April 2005 to March 2013), completed questionnaires were obtained from 2055 cases who fulfilled the McDonald criteria,13 and from 4518 matched controls, which is equivalent to 93% of invited cases and 73% of invited controls. In November 2013, complementary questions were sent to all participants who had answered the standardised questionnaire during the aforementioned period. Among other questions, participants were asked to report their coffee consumption during different age periods by asking ‘How many cups of coffee did you usually drink per day when you were 15–19, 20–29, 30–39 and 40+ years old?’. For each age period, the answer options were 0, 1–3, 3–4, 5–6, 7–8 and 8 or more cups of coffee daily. Those who reported 7–8 or 8 or more cups of coffee daily were combined into one category since few participants consumed more than 8 cups of coffee daily. The complementary questions were answered by 82% of the cases and 66% of the controls. Information regarding coffee consumption was missing for 57 cases and 166 controls, and these individuals were excluded. Cases younger than 15 years of age at disease onset and their corresponding controls were also excluded. The present study is thus comprised of 1620 cases and 2788 controls.

In the KPNC study, participants completed a computer-assisted telephone interview at study entry administered by trained staff regarding lifestyle factors and various exposures; recruitment began in mid-2007. As of a recent data freeze in August 2014, the study included a total of 1479 cases and 1185 controls. Within this data set, there were 1163 cases and 1178 matched controls. The study participation proportions were approximately 80% for cases and 66% for controls. Information on consumption habits of caffeinated coffee was obtained by asking participants ‘the highest number of servings per day they have ever been consumed for 6 months or more’ (answer options were 0, 1, 2–3 and 4 or more cups of caffeinated coffee daily), and those who reported consuming one or more servings of any caffeinated beverage were asked ‘to estimate at what age they started drinking caffeine on a regular basis’. Information regarding coffee consumption was missing for four cases and six controls, and these individuals were excluded. The results based on the KPNC study are thus comprised of 1159 cases and 1172 controls.

For each case in both studies, the year of the initial appearance of symptoms indicative of MS was defined as the index year. Coffee consumption was considered prior to the index year in the cases and during the same period of time in the corresponding controls. Participants from both data sets were categorised into groups based on their daily consumption of coffee (number of cups of coffee). The size of one cup of coffee varies between the countries. In Sweden, one cup of coffee was defined as 150 mL, whereas a cup of coffee was defined as 237 mL (8 oz) in the KPNC study.

Confounding factors and other covariates

All analyses in EIMS were adjusted for relevant demographic factors as well as for factors previously associated with MS risk. These factors included the design variables age, sex, residential area, as well as ancestry, smoking habits (number of pack years),16 exposure to passive smoking,17 sun exposure habits,18 and body mass index at age 20 years.19 ,20 Age was categorised into the same eight intervals that were used in the matching procedure: 16–19, 20–24, 25–29, 30–34, 35–39, 40–45, 45–49 and 50–70 years of age. Assessment of ancestry was based on whether the participant was born in Sweden or not, and whether either of the participant's parents had immigrated to Sweden. A participant who was born in Sweden and whose parents had not immigrated, was classified as Swedish. Smoking was adjusted for by using a continuous variable (pack years of smoking). Exposure to passive smoking was dichotomised into ever-exposed or never-exposed before the index year. On the basis of three questions regarding ultraviolet radiation (UVR) exposure where each answer option was given a number ranging from 1 (the lowest exposure) to 4 (the highest exposure), we constructed an index by adding the numbers together, and thus, acquired a value between 3 and 12. UVR exposure was adjusted for as a continuous variable. Body mass index at age 20 years was calculated by dividing self-reported weight in kilograms by self-reported height in metres squared, and adjusted for as a continuous variable.

Additional covariates included educational level (compulsory school, vocational upper secondary school, theoretical upper secondary school, other education and university), socioeconomic status (skilled and unskilled workers, assistant, intermediate, and higher non-manual employees), alcohol consumption at inclusion in the study (yes or no), a history of infectious mononucleosis (yes or no) and HLA-DRB1*15 status (positive or negative), but these factors did not significantly influence the association between coffee consumption and MS and were not retained in the final models.

In the KPNC study, sex, birth date, race/ethnicity and zip code of the case residence were taken into consideration by matched analysis. Adjustment was made for similar potential confounders as in the EIMS study, including smoking habits, passive smoking, sun exposure habits and body mass index in one's 20s21 (average of self-reported highest and lowest non-pregnancy weight during one’s 20s, divided by self-reported height in metres squared at the time of interview). Smoking was dichotomised into ever-smokers and never-smokers before the index year. Exposure to passive smoking was dichotomised into those who had and had not been exposed to passive smoking after the age of 18 years.

Sun exposure was dichotomised into high exposure (always or almost always sunbathed in the summer at the age of 10 years), and low exposure (sometimes, rarely or never sunbathed in the summer at the age of 10 years). Body mass index in one's 20s was entered as a continuous variable. Adjustments were also made for educational level (college or no college education, or unknown), a history of infectious mononucleosis (yes, no or unknown), and HLA-DRB1*15 status (positive, negative, or unknown), but these factors had only minor influence on the results. Similar to analyses for EIMS, these factors were not retained in the final models as there was no significant change in the association of coffee consumption and the odds of MS.

Statistical analysis

Using logistic regression, the occurrence of MS among participants with different coffee consumption habits was compared with those who never drank coffee, by calculating ORs with 95% CIs. In EIMS, we investigated the influence of coffee consumption at the index year, and 5 and 10 years prior to the index year. We performed both matched and unmatched analyses of EIMS. In the unmatched analysis we were able to include a higher number of controls, and therefore had increased statistical power. Only the results from the unmatched analyses are presented in this report since these were in close agreement with those from the matched analyses but had tighter CIs. A trend test for a dose–response relationship regarding coffee consumption and risk of MS was performed by using a numerical integer variable for coffee consumption, ranging from 0 to 4 in EIMS, and 0 to 2 in KPNC.

In KPNC, we investigated the influence of coffee consumption at the index year, and 5 years prior to the index year, using conditional logistic regression. We assumed that consumption of coffee began at the age the subjects first began consuming caffeinated beverages. For each individual, the reported highest number of servings per day that coffee had ever been consumed for 6 months or more was considered to represent typical consumption during the whole period of regular coffee consumption. A trend test for a dose–response relationship regarding coffee consumption and risk of MS was performed by using a numerical variable for coffee consumption ranging from 0 to 2. To differentiate if results were related to an overall effect of caffeine or were likely attributable to coffee intake specifically, we also conducted analyses based on servings of tea and soda separately, using the same assumptions as we did for coffee for ‘consumption before onset’. This analysis was performed on KPNC data only. Data on non-caffeinated beverages were neither collected for the KPNC study participants nor for the EIMS participants. In Sweden, consumption of non-caffeinated beverages is rare.

We also conducted a meta-analysis in which the results of the two case–control studies were combined by calculating a weighted average of the two study-specific adjusted ORs (weights were proportional to the inverse of the variance for each OR). All analyses were conducted using Statistical Analysis System (SAS) V.9.2.