Clinical Rheumatology

https://doi.org/10.1007/s10067-018-4261-5

BRIEF REPORT

Inflammatory dietary pattern and risk of developing rheumatoid

arthritis in women
Jeffrey A. Sparks 1,2 & Medha Barbhaiya 3 & Sara K. Tedeschi 1,2 & Cianna L. Leatherwood 1,2 & Fred K. Tabung 4,5 &
Cameron B. Speyer 1 & Susan Malspeis 1 & Karen H. Costenbader 1,2 & Elizabeth W. Karlson 1,2 & Bing Lu 1,2

Received: 19 May 2018 / Revised: 1 August 2018 / Accepted: 8 August 2018

# International League of Associations for Rheumatology (ILAR) 2018

Abstract
Our objective was to investigate whether a dietary pattern derived using inflammatory biomarkers is associated with rheumatoid
arthritis (RA) risk. We prospectively followed 79,988 women in the Nurses’ Health Study (NHS, 1984–2014) and 93,572 women
in the NHSII (1991–2013); incident RA was confirmed by medical records. Food frequency questionnaires (FFQ) were com-
pleted at baseline and approximately every 4 years. Inflammatory dietary pattern was assessed from FFQ data using the Empirical
Dietary Inflammatory Pattern (EDIP), including 18 anti-/pro-inflammatory food/beverage groups weighted by correlations with
plasma inflammatory biomarkers (interleukin-6, C-reactive protein, and tumor necrosis factor-α receptor 2). We investigated
associations between EDIP and RA using Cox regression. We identified 1185 incident RA cases over 4,425,434 person-years.
EDIP was not associated with overall RA risk (p trend = 0.21 across EDIP quartiles). Among women ≤ 55 years, increasing EDIP
was associated with increased overall RA risk; HRs (95% CIs) across EDIP quartiles were 1.00 (reference), 1.14 (0.86–1.51),
1.35 (1.03–1.77), and 1.38 (1.05–1.83; p for trend = 0.01). Adjusting for BMI attenuated this association. Increasing EDIP was
associated with increased seropositive RA risk among women ≤ 55 years (p for trend = 0.04). There was no association between
EDIP and RA among women > 55 years (EDIP-age interaction, p = 0.03). An inflammatory dietary pattern was associated with
increased seropositive RA risk with onset ≤ 55 years old, and this association may be partially mediated through BMI.

Keywords Diet . Epidemiology . Inflammation . Rheumatoid arthritis

Introduction risk factors may help to elucidate pathogenesis and could

identify lifestyle changes to lower RA risk. Individual foods
Systemic inflammation is important in the development of and beverages are associated with RA, perhaps by contribut-
rheumatoid arthritis (RA)-related autoimmunity such as auto- ing to systemic inflammation; fish and alcohol intake may
antibodies prior to clinical presentation [1]. Identifying RA lower RA risk, while red meat and sugar-sweetened soda con-
sumption may increase RA risk [2–5].
Dietary pattern analyses combine multiple foods/bever-
* Jeffrey A. Sparks ages, a preferable method for investigation since these are
jsparks@bwh.harvard.edu not consumed in isolation. We previously reported that a
healthier dietary pattern was associated with lower RA risk,
1
Department of Medicine, Division of Rheumatology, Immunology,
particularly seropositive RA risk for women aged ≤ 55 years
and Allergy, Brigham and Women’s Hospital, 60 Fenwood Road, [6]. However, the dietary quality measure investigated was
#6016U, Boston, MA 02115, USA developed for other chronic diseases [6].
2
Harvard Medical School, Boston, MA, USA Therefore, we investigated the association between di-
3
Hospital for Special Surgery, New York City, NY, USA
etary inflammatory potential assessed using the empirical
4
dietary inflammatory pattern (EDIP) score and RA risk.
Department of Nutrition, Harvard T.H. Chan School of Public
Health, Boston, MA, USA
We used two Nurses’ Health Study (NHS) cohorts, char-
5
acterized by rich data on lifestyle factors including repeat-
Department of Internal Medicine, Division of Medical Oncology,
The Ohio State University College of Medicine,
ed dietary assessments. We hypothesized that higher EDIP
Columbus, OH, USA scores (indicating pro-inflammatory diets) would be
 Clin Rheumatol

associated with increased RA risk, particularly seroposi- beverage group related to these biomarkers was then weighted
tive RA diagnosed at a younger age. by the beta coefficient in the multivariable linear regression
model. The summary EDIP score was the weighted sum for all
intake of food groups. The raw EDIP scores were rescaled by
Methods dividing by 1000 to reduce the magnitude of scores.
The 9 pro-inflammatory groups (positively associated with
Study population IL-6/CRP/TNFαR2) previously identified were processed
meat, red meat, organ meat, non-dark meat fish, other vegeta-
The NHS and NHSII are prospective cohorts of US registered bles (other than green leafy and dark-yellow vegetables), re-
nurses. The NHS enrolled 121,700 women aged 30–55 years fined grains, high-energy beverages, low-energy beverages,
in 1976; the NHSII enrolled 116,670 women aged 25–42 years and tomatoes [8]. The 9 anti-inflammatory groups (inversely
in 1989. Participants answered questionnaires at each study’s associated with IL-6/CRP/TNFαR2) previously identified
inception and every 2 years during follow-up. Each time point were beer, wine, tea, coffee, dark-yellow vegetables, leafy
where questionnaires are collected is referred to as a cycle of green vegetables, snacks, fruit juice, and pizza [8].
follow-up. The baseline for this analysis is 1984 for NHS and We pooled both cohorts for statistical efficiency due to
1991 for NHSII when a comprehensive Food Frequency planned subgroup analyses and in anticipation of a modest
Questionnaire (FFQ) was introduced into each cohort. effect size for a dietary exposure. The EDIP for each cycle,
Excluding participants with prevalent RA and FFQ non- as a time-varying exposure, was calculated using cumulative
responders at baseline in each cohort, 79,988 NHS partici- average dietary intake to reflect long-term diet, by averaging
pants, followed from 1984 to 2014 and 93,572 NHSII partic- the dietary intake from baseline until that follow-up cycle. The
ipants followed from 1991 to 2013, were analyzed. All aspects EDIP values in our analysis ranged between − 3.6 (most anti-
of the study, including obtaining informed consent from par- inflammatory) and + 2.5 (most pro-inflammatory). Since there
ticipants, content/mailing/data management of questionnaires, are no known clinically meaningful EDIP cutpoints, individ-
and identification of incident RA cases, were approved by the uals were ranked from lowest to highest EDIP score and then
Partners HealthCare Institutional Review Board. placed in quartiles at the baseline of each cohort. The lowest
(first) quartile represents the group with the least inflammato-
Dietary assessments ry dietary intake (the reference group) and the highest (fourth)
quartile represents the group with the most inflammatory die-
FFQs assessed dietary intake over the previous year, ranking tary intake. This type of analysis accounts for all previous and
frequency of each food/beverage on a scale from never or < 1/ current dietary measures for each cycle predicting RA risk in
month to ≥ 6 servings/day [6, 7]. FFQs were administered in the subsequent period. For example, we used the averaged
1984, 1986, and every 4 years until 2010 in the NHS and 1991 dietary intake at 1984, 1986, and 1990 to calculate EDIP,
and every 4 years until 2011 in the NHSII. and predict the subsequent RA incidence from 1990 to
1994, similarly, using the averaged diet up to 1994 to predict
Empirical dietary inflammatory pattern score RA incidence from 1994 to 1998, and so on.

The empirical dietary inflammatory pattern (EDIP) was devel- Identification of incident RA
oped as an agnostic method to classify food/beverage groups
as anti- or pro-inflammatory, as described in more detail else- Women who self-reported incident RA were mailed a ques-
where [8]. In the previous validation study, the NHS was used tionnaire [9]. Medical records were obtained to verify RA
as the discovery dataset, while the NHSII and the Health diagnosis and determine diagnosis date/serologic status. All
Professional Follow-up Study (a similar large prospective co- cases were reviewed by two rheumatologists and confirmed
hort study, but performed among only men) were used as the RA according to either 1987 American College of
validation datasets [8]. Briefly, intake of each food/beverage Rheumatology or 2010 ACR/European League Against
group was tested for correlations with plasma inflammatory Rheumatism criteria [10, 11]. Seropositive RA was defined
biomarker levels (interleukin-6 [IL-6], C-reactive protein as positive rheumatoid factor (RF) or anti-cyclic citrullinated
[CRP], and tumor necrosis factor-α receptor 2 [TNFαR2]) peptide (anti-CCP).
using reduced rank regression [8]. Reduced rank regression
is a statistical method used to derive a set of dietary patterns Covariates
associated with biomarkers on the causal pathway to an out-
come. The advantages of this method are that the predictive Time-varying covariate data were obtained through biennial
potential for the outcome is maximized while simultaneously questionnaires. Age and household income, derived from US
accounting for the collinearity of predictors. Each food/ Census-tract median income by zip code, were continuous
Clin Rheumatol

variables. Smoking was categorized by status (never/past/cur- examining whether EDIP effects on RA varied by calendar
rent) and pack-years. Body mass index (BMI) was categorized time. We tested whether the association of EDIP with RA risk
according to the World Health Organization recommendations differed before or after the year 2000 by including an interac-
into underweight/normal, overweight, and obese or consid- tion between EDIP and an indicator variable for calendar time
ered as a continuous variable. Self-reported physical activity (before 2000 vs. 2000 or later) in a separate model and
was a continuous variable measured as weekly metabolic reporting the p value.
equivalents. We included reproductive/hormonal factors such
as oral contraceptive use, parity, breastfeeding duration, and
menopausal status/postmenopausal hormone (PMH) use. Results

Statistical analysis Baseline characteristics of both cohorts (total n = 173,560)

according to EDIP quartiles are shown in Table 1. Those with
We reported baseline characteristics in each cohort according lower inflammatory dietary pattern were older, smoked more,
to EDIP quartile. We investigated EDIP and RA risk, overall had higher income, had higher physical activity, and had low-
and stratified by serologic status and by pre-specified age er BMI than women with higher inflammatory dietary pattern.
groups. We previously found differences in metabolic/ We identified 1185 incident RA cases (62.6% seropositive)
dietary RA risk factors related to age of onset ≤ 55 or > over 4,425,434 person-years of follow-up (mean follow-up
55 years old [5, 6, 12]. We used that age cutpoint as an a priori per participant of 25.5 years).
hypothesis due to different RA clinical phenotypes based on Table 2 shows the association of EDIP with RA risk. There
age of onset and as an approximation of completion of men- was no association of EDIP with all RA (p for trend = 0.21),
opausal transition. seropositive RA (p for trend = 0.38), or seronegative RA (p for
We used a prospective cohort design wherein dietary intake trend = 0.37).
at each cycle of follow-up was used to predict incident RA Table 3 shows the association of EDIP with RA by age (≤
occurring in the subsequent period prior to the next cycle of 55 years and > 55 years). Among women ≤ 55 years, increas-
dietary data. In all analyses, the sample at each cycle that we ing EDIP was associated with increased overall RA risk. HRs
analyzed was free of RA or other connective tissue diseases. (95% CIs) across EDIP quartiles were 1.00 (reference), 1.14
We hypothesized that a long-term measure of dietary intake (0.86–1.51), 1.35 (1.03–1.77), and 1.38 (1.05–1.83) with a
was most important for risk of a chronic disease such as RA. significant linear trend (p for trend = 0.01). These results were
Therefore, we used cumulative average updated EDIP as a attenuated when additionally adjusting for BMI (HRQ4vs.Q1
measure of long-term dietary intake. This type of analysis 1.25, 95% CI 0.94–1.65; p for trend = 0.09). In mediation
accounts for all previous and current dietary measures for each analyses, the estimated proportion of EDIP effect mediated
cycle predicting RA risk in the subsequent 2-year cycle. by BMI was 41.8% (95% CI 10.3–81.8%). Among women
Time-varying Cox regression models estimated hazard ratios ≤ 55 years, intake of pro-inflammatory diets was associated
(HR) and 95% confidence intervals (CI) for the association of with increased seropositive RA risk (p for trend = 0.04), but
EDIP quartiles with RA (reference, first quartile). We calculated EDIP was not associated with seronegative RA (p for trend =
p values for trend using the median EDIP score for each quartile 0.13). Additionally, we did not observe the association of
as a continuous variable in the model. Person-years com- EDIP with RA risk differs between before or after the year
menced from the return date of the baseline questionnaire to 2000 (EDIP-calendar time interaction p = 0.27). Among
date of last follow-up/death/censor, whichever came first. women > 55 years old, there was no association of EDIP with
In base models, we adjusted for age, questionnaire period, all RA (p for trend = 0.55; EDIP-age interaction p = 0.03),
cohort, and total energy intake. We built multivariable models seropositive RA (p for trend = 0.47), or seronegative RA (p
based on the primary analysis and included covariates that for trend = 0.96).
were associated with EDIP and RA. The final multivariable
model included the base factors and census-tract household
income, smoking pack-years, and menopausal status/hormone Discussion
use (premenopausal, postmenopausal/never PMH use,
postmenopausal/current PMH use, and postmenopausal/past In this study of 173,560 women with up to 30 years of follow-
PMH use). We further adjusted for BMI, and performed for- up, an inflammatory dietary pattern was associated with in-
mal mediation analysis to examine and quantify the mediating creased seropositive RA risk among women aged ≤ 55 years.
effect of BMI on the association of EDIP with RA [13]. Since However, we found no association of inflammatory dietary
the imaging and laboratory tests used to diagnose RA, as well pattern for overall RA risk or among women age > 55 years.
as treatment strategies for those newly diagnosed, have In addition, the association between inflammatory diet and
changed over time, we performed a sensitivity analysis RA may be partially mediated through BMI.
 Clin Rheumatol

Table 1 Age-standardized baseline characteristics of Nurses’ Health Study (1984) and Nurses’ Health Study II (1991) participants by Empirical
Dietary Inflammatory Pattern (EDIP) quartiles (n = 173,560)

Nurses’ Health Study (n = 79,988) Nurses’ Health Study II (n = 93,572)

EDIP quartiles* Q1 (least Q2 Q3 Q4 (most Q1 (least Q2 Q3 Q4 (most

inflammatory) (n = 20,884) (n = 19,234) inflammatory) inflammatory) (n = 22,508) (n = 24,153) inflammatory)
(n = 23,174) (n = 16,696) (n = 20,217) (n = 26,694)

Mean age, 51.2 (6.9) 51.3 (7.2) 51.0 (7.2) 50.1 (7.3) 37.6 (4.4) 36.9 (4.6) 36.3 (4.7) 35.8 (4.7)
years (SD)**
Mean US 68.3 (27.6) 65.4 (26.0) 63.0 (24.5) 60.6 (23.6) 64.6 (24.4) 63.1 (23.4) 60.9 (21.9) 57.8 (20.3)
Census-tract
household income
($USD × 1000)
Mean BMI, 24.0 (3.8) 24.6 (4.3) 25.4 (4.9) 26.6 (5.8) 23.7 (4.4) 24.0 (4.7) 24.6 (5.2) 25.9 (6.3)
kg/m2 (SD)
Mean total METs/ 15.9 (23.9) 14.4 (20.5) 13.3 (18.7) 12.1 (17.8) 24.6 (31.4) 21.6 (27.2) 19.9 (25.6) 18.3 (25.1)
week (SD)
Current smoking, % 29.1 23.7 20.8 21.7 15.7 11.7 10.0 12.3
Menarche at < 12 23.1 22.8 23.0 24.3 24.9 23.9 23.8 25.2
years of age, %
Oral contraceptive 49.6 48.5 47.5 47.5 84.9 84.6 84.6 84.0
use, %
Parous, % 92.4 92.7 92.5 92.3 69.9 71.8 73.4 72.8
Breastfed 17.7 17.3 16.8 15.7 30.6 30.5 29.3 24.5
≥ 12 months, %
Postmenopausal, % 59.1 60.2 60.7 62.0 3.1 3.0 3.3 3.8
Current 22.7 22.6 22.9 21.0 2.5 2.4 2.7 3.0
postmenopausal
hormone use, %
Mean energy intake, 1705 (509) 1662 (501) 1705 (514) 1938 (564) 1787 (541) 1689 (508) 1715 (518) 1944 (577)
kcal (SD)

*The cutpoints for EDIP quartiles were based on baseline data pooled from both cohorts
**Not age-adjusted
Missing values are not shown

These results add to the literature suggesting that diet plays NHSII collected data on date of menopause, there is typically
an important role in RA pathogenesis, perhaps by altering sys- a period of perimenopause lasting for months or even years,
temic inflammation and autoimmunity [2–6]. Fish intake may such that a clear transition point is typically not obvious. We
exert a protective effect on RA by the anti-inflammatory effects previously showed that menopausal status has a strong impact
of omega-3 polyunsaturated fatty acids [2]. Among at-risk in- on RA risk [17]. Therefore, we relied on the age of 55 years to
dividuals, erythrocyte membrane-bound omega-3 PUFA levels clearly classify women as younger or older in this analysis.
were inversely associated with anti-CCP/RF positivity [14]. Younger- and older-onset RA also have clinical phenotypic
Previous studies investigating EDIP found robust associations differences related to presentation and response to treatment.
between EDIP scores and colorectal cancer risk but no associ- Patients with older-onset RA may be more likely to be sero-
ation with ovarian cancer risk [15, 16]. Similar to previous negative and have a less severe onset compared to younger-
studies investigating dietary/metabolic factors in RA, the in- onset RA. Older-onset RA may also be more likely to be
flammatory dietary pattern was important specifically in responsive to glucocorticoids and may not be as likely to
younger-onset seropositive RA, defined as ≤ 55 years [5, 6, require combinations of drugs compared to younger-onset
12]. Thus, metabolic/dietary lifestyle factors may affect RA risk RA. While there is no standard age cutpoint to classify
differently based on age of onset as well as serologic status. younger- vs. older-onset RA, we chose 55 years given the
We found different results based on an age cutpoint of ≤ 55 biologic plausibility related to menopause and clinical RA
or > 55 years. Menopause has known effects on both metab- differences based on age at onset. It is possible other age
olism and adipose tissue distribution such that women may cutpoints might have yielded different results.
both gain adiposity while also losing bone mass. Therefore, We previously reported a protective effect on RA for a health-
dietary intake behaviors and metabolism change after meno- ier dietary quality based on the 2010 Alternative Healthy Eating
pause as well as the reliability of anthropometric measures Index (2010-AHEI). The 2010-AHEI was derived based on
such as body mass index. The median age for menopause consensus from experts related to risk of developing diabetes,
completion is about 55 years of age. While the NHS and cancer, and cardiovascular disease [18]. Therefore, the 2010-
Clin Rheumatol

Table 2 Hazard ratios (95% CI) for rheumatoid arthritis according to EDIP quartiles in the Nurses’ Health Study and Nurses’ Health Study II

Empirical Dietary Inflammatory Pattern (EDIP) quartiles

Q1 (least inflammatory) Q2 Q3 Q4 (most inflammatory) p trend‡

All RA (n = 1185 cases)

Cases/person-years 252/960,107 367/1,316,534 327/1,218,122 239/930,671
Age-adjusted* 1.00 (Ref) 1.08 (0.92 to 1.27) 1.07 (0.91 to 1.26) 1.10 (0.92 to 1.32) 0.31
Multivariable model 1 (main)** 1.00 (Ref) 1.11 (0.95 to 1.31) 1.12 (0.95 to 1.32) 1.12 (0.94 to 1.35) 0.21
Multivariable model 2† 1.00 (Ref) 1.09 (0.93 to 1.28) 1.08 (0.91 to 1.28) 1.07 (0.89 to 1.28) 0.54
Seropositive RA (n = 741 cases)
Cases/person-years 151/958,050 231/1,314,010 222/1,215,646 137/928,344
Age-adjusted* 1.00 (Ref) 1.13 (0.92 to 1.39) 1.20 (0.98 to 1.48) 1.03 (0.82 to 1.31) 0.61
Multivariable model 1 (main)** 1.00 (Ref) 1.18 (0.96 to 1.45) 1.28 (1.04 to 1.58)0 1.07 (0.85 to 1.36) 0.38
Multivariable model 2† 1.00 (Ref) 1.16 (0.94 to 1.42) 1.24 (1.00 to 1.52) 1.01 (0.79 to 1.29) 0.72
Seronegative RA (n = 444 cases)
Cases/person-years 101/957,570 136/1,313,598 105/1,215,052 102/928,321
Age-adjusted* 1.00 (Ref) 1.00 (0.78 to 1.30) 0.87 (0.66 to 1.15) 1.22 (0.92 to 1.61) 0.32
Multivariable model 1 (main)** 1.00 (Ref) 1.01 (0.78 to 1.31) 0.88 (0.67 to 1.16) 1.20 (0.91 to 1.60) 0.37
Multivariable model 2† 1.00 (Ref) 0.99 (0.77 to 1.29) 0.85 (0.64 to 1.12) 1.15 (0.86 to 1.53) 0.59

The cutpoints for EDIP quartiles were based on baseline data pooled from both cohorts
*Adjusted for age, questionnaire cycle, and cohort
**Adjusted for age, questionnaire cycle, cohort, total energy intake, census-tract household income, smoking pack-years, hormone use (premenopausal,
postmenopausal/never postmenopausal hormone use, postmenopausal/current postmenopausal hormone use, and postmenopausal/past postmenopausal
hormone use)
†
Additionally adjusted for BMI categories (< 25, 25–29.9, ≥ 30 kg/m2 )
‡
p trend was derived from tests of linear trend across categories of EDIP using the median value of each category as a continuous variable

AHEI was not derived related to RA or other autoimmune dis- confounders could have affected our results. While the relation-
eases which may have different pathogenesis. Further, the ship between diet and BMI is complex, we considered BMI as a
methods for selecting components of the 2010-AHEI do not mediator since diet affects BMI but not vice versa. In our study,
consider the correlation structure of food and nutrient intake. over 40% of the EDIP association with RA may be mediated
The EDIP was derived using reduced rank regression, which through BMI. However, another interpretation of these results
offers some advantages over the 2010-AHEI [6]. First, this would be that BMI confounds the association between the EDIP
method uses objective inflammatory plasma biomarkers that and RA risk such that BMI explains most of the observed excess
are known to be important in the causal pathway for develop- risk. Overall, our results provide further support for dietary mod-
ment of RA [1]. Therefore, the associations are more likely be ifications to maintain healthy weight to possibly decrease RA
directly applicable to RA development compared to the 2010- risk [12]. EDIP was developed using the population studied by
AHEI. Second, the statistical methods for deriving the EDIP correlating food/beverage groups with levels of plasma inflam-
take into account the correlation structure of food/nutrient matory markers [8]. While many of the food/beverage groups
groups lowering the possibility that the association is confound- were classified as expected (e.g., beer/wine as anti-inflammato-
ed by intake from other measured or unmeasured components of ry), there were some unexpected classifications (e.g., pizza as
the diet [6]. However, it is possible that both of these methods anti-inflammatory, which may be due to the higher bioavailable
classifying patterns of dietary intake are measuring a similarly lycopene in pizza than in fresh tomatoes), perhaps related to the
healthy or unhealthy diet for RA risk. Future comparative stud- dietary habits of a population of older health professionals [8].
ies may be needed to directly compare the predictive ability for EDIP scores may be interpreted as reflecting the potential of diet
RA for these and other dietary measures in order to inform to contribute to systemic inflammation. EDIP can be considered
public health recommendations. as a surrogate long-term measure contributing to IL-6/CRP/
Our study was limited by including only women who were TNFαR2 levels, but the external generalizability to non-health
well-educated and working at cohort entry so may not be gen- professionals of EDIP is currently unclear. The biomarkers used
eralizable. While detailed time-varying data on covariates such to develop EDIP are important in RA pathogenesis, so this die-
as smoking and BMI were available, other unmeasured tary pattern may be more specific for RA than others [6].
 Clin Rheumatol

Table 3 Hazard ratios (95% CI) for rheumatoid arthritis according to EDIP quartiles in the Nurses’ Health Study and Nurses’ Health Study II stratified
by age ≤ 55 or > 55 years

Empirical Dietary Inflammatory Pattern (EDIP) quartiles

Q1 (least inflammatory) Q2 Q3 Q4 (most inflammatory) p trend‡

Age ≤ 55 years*** All RA (n = 490 cases)

Cases/person-years 85/455,964 123/610,841 144/630,515 138/598,330
Age-adjusted* 1.00 (Ref) 1.11 (0.84 to 1.46) 1.31 (1.00 to 1.71) 1.40 (1.07 to 1.84) 0.01
Multivariable model 1 (main)** 1.00 (Ref) 1.14 (0.86 to 1.51) 1.35 (1.03 to 1.77) 1.38 (1.05 to 1.83) 0.01
Multivariable model 2† 1.00 (Ref) 1.11 (0.84 to 1.46) 1.27 (0.97 to 1.67) 1.25 (0.94 to 1.65) 0.09
Seropositive RA (n = 319 cases)
Cases/person-years 53/454,800 78/609,357 104/628,972 84/596,561
Age-adjusted* 1.00 (Ref) 1.12 (0.79 to 1.59) 1.51 (1.08 to 2.10) 1.36 (0.96 to 1.92) 0.03
Multivariable model 1 (main)** 1.00 (Ref) 1.16 (0.82 to 1.65) 1.58 (1.13 to 2.21) 1.37 (0.96 to 1.94) 0.04
Multivariable model 2† 1.00 (Ref) 1.12 (0.79 to 1.59) 1.46 (1.04 to 2.05) 1.19 (0.83 to 1.71) 0.21
Seronegative RA (n = 171 cases)
Cases/person-years 32/454,512 45/609,227 40/628,581 54/596,511
Age-adjusted* 1.00 (Ref) 1.13 (0.72 to 1.78) 1.01 (0.63 to 1.61) 1.53 (0.98 to 2.38) 0.07
Multivariable model 1 (main)** 1.00 (Ref) 1.14 (0.72 to 1.80) 1.01 (0.63 to 1.62) 1.45 (0.92 to 2.28) 0.13
Multivariable model 2† 1.00 (Ref) 1.12 (0.71 to 1.77) 0.98 (0.61 to 1.57) 1.38 (0.87 to 2.18) 0.20
Age ≥> 55 years*** All RA (n = 695 cases)
Cases/person-years 167/508,826 244/713,996 183/595,759 101/338,264
Age-adjusted* 1.00 (Ref) 1.06 (0.87 to 1.29) 0.95 (0.77 to 1.17) 0.90 (0.70 to 1.15) 0.29
Multivariable model 1 (main)** 1.00 (Ref) 1.10 (0.90 to 1.34) 1.00 (0.81 to 1.23) 0.94 (0.73 to 1.21) 0.55
Multivariable model 2† 1.00 (Ref) 1.09 (0.89 to 1.33) 0.99 (0.80 to 1.22) 0.93 (0.72 to 1.20) 0.49
Seropositive RA (n = 422 cases)
Cases/person-years 98/507,922 153/712,941 118/594,817 53/337,689
Age-adjusted* 1.00 (Ref) 1.13 (0.88 to 1.46) 1.03 (0.79 to 1.35) 0.79 (0.57 to 1.11) 0.20
Multivariable model 1 (main)** 1.00 (Ref) 1.19 (0.92 to 1.54) 1.12 (0.85 to 1.47) 0.85 (0.60 to 1.19) 0.47
Multivariable model 2† 1.00 (Ref) 1.19 (0.92 to 1.54) 1.12 (0.85 to 1.47) 0.86 (0.61 to 1.21) 0.52
Seronegative RA (n = 273 cases)
Cases/person-years 69/507,726 91/712,650 65/594,594 48/337,715
Age-adjusted* 1.00 (Ref) 0.96 (0.71 to 1.32) 0.83 (0.59 to 1.16) 1.06 (0.73 to 1.53) 0.90
Multivariable model 1 (main)** 1.00 (Ref) 0.97 (0.71 to 1.33) 0.83 (0.59 to 1.17) 1.08 (0.74 to 1.56) 0.96
Multivariable model 2† 1.00 (Ref) 0.95 (0.69 to 1.30) 0.80 (0.57 to 1.14) 1.03 (0.70 to 1.50) 0.76

The cutpoints for EDIP quartiles were based on baseline data pooled from both cohorts
*Adjusted for age, questionnaire cycle, and cohort
**Adjusted for age, questionnaire cycle, cohort, total energy intake, census-tract household income, smoking pack-years, hormone use (premenopausal,
postmenopausal/never postmenopausal hormone use, postmenopausal/current postmenopausal hormone use, and postmenopausal/past postmenopausal
hormone use)
†
Additionally adjusted for BMI categories (< 25, 25–29.9, ≥ 30 kg/m2 )
‡p trend was derived from tests of linear trend across categories of EDIP using the median value of each category as a continuous variable
***EDIP-age interaction for all RA: p = 0.03

We found statistically significant findings only among the in metabolism after menopause may affect the relationship be-
subset with RA diagnosed ≤ 55 years of age. We pre-specified tween dietary and metabolic factors and RA risk. The patho-
this hypothesis based on previous papers showing that BMI and genesis of older-onset RA may be different than younger-onset
other dietary factors were important for earlier onset, but not RA since these subsets have differences in clinical presentation
later onset, RA [6]. There were more cases diagnosed at > and response to treatment. Our study analyzed two closed co-
55 years of age, so the lack of association in older women is horts, so we are limited in investigating secular trends based on
less likely related to reduced power. It is possible that changes calendar time since women in both studies were aging in
Clin Rheumatol

tandem so there were few younger women in the later years of approved the final version to be published. Dr. Sparks had full access to
all of the data in the study and takes responsibility for the integrity of the
the analysis. It is possible that the availability of novel diagnos-
data and the accuracy of the data analysis.
tic biomarkers, advanced imaging, treatment options, and the Study conception and design: Sparks, Karlson, Lu
paradigm to diagnose RA and treat early to prevent disease- Acquisition of data: Sparks, Barbhaiya, Tedeschi, Leatherwood,
related damage may have affected the RA phenotype. Tabung, Costenbader, Karlson, Lu
Analysis and interpretation of data: Sparks, Barbhaiya, Tedeschi,
Therefore, it is possible that the results may have been explained Leatherwood, Tabung, Speyer, Malspeis, Costenbader, Karlson, Lu
related to secular changes of RA diagnosis in the Bbiologic era^
starting around 2000. Similarly, there have been secular trends Funding This work was supported by the National Institutes of Health
over calendar time related to lifestyle factors, such as diet and (grant numbers R01 AR061362, K24 AR052403, L30 AR066953, R01
smoking. However, we did not observe the association of EDIP AR049880, UM1 CA186107, UM1 CA176726, K99 CA207736, K23
AR069688, T32 AR007530, P30 AR070253, and P30 AR072577).
with RA risk differ before or after the year 2000. Further research
is needed to understand possible differences both in the biologic
response to dietary exposures and differences in disease pheno- Compliance with ethical standards
types based on age and calendar time.
Disclosures None.
Strengths of our study include the prospective cohort study
design with large sample size and lengthy follow-up. We had
up to 8 repeated measures of dietary intake collected prior to
References
RA onset available and investigated EDIP, a validated dietary
measure correlating food/beverage groups with systemic in- 1. Karlson EW, Chibnik LB, Tworoger SS, Lee IM, Buring JE, Shadick
flammatory biomarkers derived using agnostic methods. We NA, Manson JAE, Costenbader KH (2009) Biomarkers of inflamma-
analyzed cumulative average dietary intake to best reflect tion and development of rheumatoid arthritis in women from two
long-term dietary patterns. We identified incident RA occur- prospective cohort studies. Arthritis Rheum 60(3):641–652
2. Di Giuseppe D, Crippa A, Orsini N, Wolk A (2014) Fish consump-
ring after baseline and all cases met accepted research criteria tion and risk of rheumatoid arthritis: a dose-response meta-analysis.
for RA. We had a relatively large number of incident RA cases Arthritis Res Ther 16(5):446
so were able to detect modest effect sizes, as expected from 3. Lu B, Solomon DH, Costenbader KH, Karlson EW (2014) Alcohol
dietary factors. We had detailed covariates available to adjust consumption and risk of incident rheumatoid arthritis in women: a
prospective study. Arthritis Rheumatol 66(8):1998–2005
for time-varying confounders such as smoking and quantify 4. Benito-Garcia E, Feskanich D, Hu FB, Mandl LA, Karlson EW
the mediating of BMI. We were able to determine serologic (2007) Protein, iron, and meat consumption and risk for rheumatoid
status during medical record review to classify cases as sero- arthritis: a prospective cohort study. Arthritis Res Ther. 9(1):R16
positive or seronegative RA. However, many of the cases 5. Hu Y, Costenbader KH, Gao X, Al-Daabil M, Sparks JA, Solomon
DH et al (2014) Sugar-sweetened soda consumption and risk of
were diagnosed with RA prior to the clinical use of anti- developing rheumatoid arthritis in women. Am J Clin Nutr
CCP, so there may have been some misclassification of RA 100(3):959–967
serologic status. Since dietary exposures typically have mod- 6. Hu Y, Sparks JA, Malspeis S, Costenbader KH, Hu FB, Karlson
est effects on chronic disease risk, we pooled the two cohorts EW, Lu B (2017) Long-term dietary quality and risk of developing
rheumatoid arthritis in women. Ann Rheum Dis 76(8):1357–1364
to increase the power to be able to detect an association be- 7. Yuan C, Spiegelman D, Rimm EB, Rosner BA, Stampfer MJ,
tween EDIP score and RA risk, particularly among RA sub- Barnett JB, Chavarro JE, Subar AF, Sampson LK, Willett WC
groups by serologic status and age at onset. Future studies are (2017) Validity of a dietary questionnaire assessed by comparison
needed to replicate our findings in independent datasets. with multiple weighed dietary records or 24-hour recalls. Am J
Epidemiol 185(7):570–584
In conclusion, an inflammatory dietary pattern was associated 8. Tabung FK, Smith-Warner SA, Chavarro JE, Wu K, Fuchs CS, Hu
with increased RA risk, particularly seropositive RA for younger FB, Chan AT, Willett WC, Giovannucci EL (2016) Development
and middle-aged women. These results suggest that a long-term and validation of an empirical dietary inflammatory index. J Nutr
dietary pattern that decreases systemic inflammation may lead to 146(8):1560–1570
9. Karlson EW, Sanchez-Guerrero J, Wright EA, Lew RA, Daltroy LH,
lower RA risk. These findings add to the evidence of metabolic Katz JN, Liang MH (1995) A connective tissue disease screening
and inflammatory markers contributing to autoimmunity. Further questionnaire for population studies. Ann Epidemiol 5(4):297–302
research is needed to understand the timing, impact, and biologic 10. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF,
mechanisms of dietary behaviors and RA risk. Cooper NS et al (1988) The American Rheumatism Association
1987 revised criteria for the classification of rheumatoid arthritis.
Arthritis Rheum 31(3):315–324
Acknowledgements We thank the participants of the NHS and NHSII for 11. Aletaha D, Neogi T, Silman AJ, Funovits J, Felson DT, Bingham
their dedicated participation in this longitudinal study as well as the staff
CO 3rd et al (2010) 2010 Rheumatoid arthritis classification
members at the Channing Division of Network Medicine (Department of
criteria: an American College of Rheumatology/European League
Medicine, Brigham and Women’s Hospital and Harvard Medical School).
Against Rheumatism collaborative initiative. Arthritis Rheum
62(9):2569–2581
Author contributions All authors were involved in drafting the article or 12. Lu B, Hiraki LT, Sparks JA, Malspeis S, Chen CY, Awosogba JA,
revising it critically for important intellectual contact, and all authors Arkema EV, Costenbader KH, Karlson EW (2014) Being
 Clin Rheumatol

overweight or obese and risk of developing rheumatoid arthritis Association of dietary inflammatory potential with colorectal can-
among women: a prospective cohort study. Ann Rheum Dis cer risk in men and women. JAMA Oncol 4(3):366–373
73(11):1914–1922 16. Tabung FK, Huang T, Giovannucci EL, Smith-Warner SA,
13. Lin DY, Fleming TR, De Gruttola V (1997) Estimating the propor- Tworoger SS, Poole EM (2017) The inflammatory potential of diet
tion of treatment effect explained by a surrogate marker. Stat Med and ovarian cancer risk: results from two prospective cohort stud-
16(13):1515–1527 ies. Br J Cancer 117(6):907–911
14. Gan RW, Demoruelle MK, Deane KD, Weisman MH, Buckner JH, 17. Bengtsson C, Malspeis S, Orellana C, Sparks JA, Costenbader KH,
Gregersen PK, Mikuls TR, O'Dell JR, Keating RM, Fingerlin TE, Karlson EW (2017) Association between menopausal factors and
Zerbe GO, Clare-Salzler MJ, Holers VM, Norris JM (2017) the risk of seronegative and seropositive rheumatoid arthritis: re-
Omega-3 fatty acids are associated with a lower prevalence of au- sults from the Nurses’ Health Studies. Arthritis Care Res (Hoboken)
toantibodies in shared epitope-positive subjects at risk for rheuma- 69(11):1676–1684
toid arthritis. Ann Rheum Dis 76(1):147–152 18. Chiuve SE, Fung TT, Rimm EB, Hu FB, McCullough ML, Wang
15. Tabung FK, Liu L, Wang W, Fung TT, Wu K, Smith-Warner SA, M et al (2012) Alternative dietary indices both strongly predict risk
Cao Y, Hu FB, Ogino S, Fuchs CS, Giovannucci EL (2018) of chronic disease. J Nutr 142(6):1009–1018