Ultra-Processed Food Intake and Risk of Systemic Lupus Erythematosus among Women followed in the Nurses’ Health Study Cohorts

Abstract

Objective:

We assessed ultra-processed food (UPF) intake and SLE incidence within the prospective Nurses’ Health Study cohorts.

Methods:

204,175 women were followed, NHS: 1984 – 2016; NHSII: 1991 – 2017. Semi-quantitative food frequency questionnaires were completed every 2–4 years. UPF intake was determined as per the Nova classification. Nurses self-reported new doctor-diagnosed SLE, confirmed by medical records. Time-varying Cox regressions estimated hazard ratios (HR, 95% confidence intervals) for incident SLE and SLE by anti-double stranded DNA antibody (dsDNA) at diagnosis, according to cumulatively-updated daily: a) UPF servings, b) total intake (gms + mls), and c) % total intake. Analyses adjusted for age, race, cohort, caloric and alcohol intakes, household income, smoking, body mass index (BMI), physical activity, menarchal age, and oral contraceptive use. We tested for interaction with BMI and examined UPF categories.

Results:

Mean baseline age was ~50 years (NHS) and ~36 years (NHSII); 93% self-reported White race. 212 incident SLE cases were identified. SLE risk was higher in 3^rd vs. 1^st UPF tertile (servings/day pooled multivariable [MV] HR 1.56 (1.04–2.32); p trend 0.03). Results were stronger for dsDNA+ SLE (servings/day pooled MV HR 2.05 (1.15–3.65); p trend 0.01), and for absolute (servings or total) than % total intake. Sugar/artificially-sweetened beverages were associated with SLE risk (3^rd vs. 1^st tertile MV HR 1.45 (1.01–2.09). No BMI interactions were observed.

Conclusion:

Higher cumulative-average daily UPF intake was associated with >50% increased SLE risk, and with doubled risk for anti-dsDNA+ SLE. Many deleterious effects on systemic inflammation and immunity are postulated.

INTRODUCTION

Systemic lupus erythematosus (SLE) is a multi-system autoimmune disease predominantly affecting women, causing cellular and humoral immune system abnormalities and high levels of systemic inflammation with potential progressive and irreversible organ damage^1,2. Like that of many complex autoimmune diseases, the pathogenesis of SLE involves a combination of inherited genetic factors and environmental and lifestyle factors, some of which have been identified to date, including cigarette smoking, obesity at younger ages, oral contraceptive and postmenopausal hormone use, and alcohol intake^3–10. In recent decades, an increase in the prevalence of SLE has been observed in several countries, perhaps paralleling dietary and lifestyle changes, and increasing obesity in Western countries^11,12.

One of the most profound changes in the American diet in recent decades has been the replacement of unprocessed and minimally processed foods with ultra-processed foods (UPF), largely lacking nutritional benefit^13–15. UPF are ready-to-heat/eat formulations of processed food substances and cosmetic additives, that result from a series of industrial processes. They generally contain high levels of processing and flavors, colors, stabilizers, emulsifiers, and other non-nutritional additives^16,17. High consumption of UPF, such as chips, soda, and donuts, has been associated with obesity and chronic diseases, including cardiovascular disease, type II diabetes, cancer, inflammatory bowel disease, and even depression, as well as increased overall mortality^13,18–28. Although consumption of some UPFs has been inversely associated with risks of major health outcomes, for example whole grain breads and breakfast cereals and commercial yogurt, overall UPF dietary intake and its increase over time have been hypothesized to be related to disease pathogenesis via several potential biologic mechanisms, including the stimulation of systemic inflammation, central to the development of SLE, a multisystem autoimmune inflammatory disease²⁹.

Our past studies of prospective cohorts of women have not revealed strong associations between SLE risk and several dietary patterns, including the Western and prudent dietary patterns, the Alternative Healthy Eating Index, Empirical Inflammatory Diet, and the Dietary Interventions to Stop Hypertension (DASH) diet^30–32. We have found, however, that high carbohydrate intake was associated with increased SLE risk in a prospective study in the Black Women’s Health Study³¹. The association between UPF consumption, in many ways the most extreme form of environmental exposure through dietary intake, and risk of developing SLE, or that of other autoimmune rheumatic disease, has not yet been investigated. Thus, we examined whether consumption of UPF was associated with an increased incidence of SLE among the women followed in the large prospective Nurses’ Health Study cohorts.

MATERIALS AND METHODS

Study Participants and Data Collection

We analyzed data from two prospective cohorts of U.S. registered nurses: the Nurses’ Health Study (NHS), begun in 1976, including 121,700 women aged 30 to 55 years, and the Nurses’ Health Study II (NHSII), which began in 1989 and enrolled 116,670 women aged 25 to 42³³. Extensive lifestyle, exposure, dietary, and medical information has been assessed via repeated self-administered questionnaires. For this analysis, we included participants who self-reported dietary intake on the validated semi-quantitative food frequency questionnaire (SQFFQ), administered every four years, starting in 1984 in NHS and in 1991 in NHSII³⁴. We excluded those who did not answer the baseline food frequency questionnaire or who died prior to baseline. Deaths were reported by next-of-kin and confirmed by searches of the National Death Index and causes of death are validated by medical record review. Participants who died or who self-reported connective tissue diseases that were not confirmed as SLE (as below) were censored. Participants with prevalent SLE or connective tissue disease at baseline and those without diet exposure information were excluded at baseline.

Dietary Exposure Assessment

The validated self-reported SQFFQ assessed dietary intake approximately every 4 years, inquiring about food and beverage consumption during the previous year, and rating the frequency of each food or drink on a scale ranging from never or <1/month to ≥ 6 times/day^34–36. The validated SQFFQ included 133 food and beverage items from baseline until 2010, when it was updated to increase the list to 152 items³⁷. The Nova food processing classification was used to categorize foods into four food processing categories: 1- unprocessed and minimally processed foods, 2- processed culinary ingredients, 3- processed foods, and 4- UPF¹⁶. UPF represented 50.2% of the food-list assessed by the SQFFQs³⁸. Briefly, the following foods categories were classified as UPF: a) ultra-processed breads and breakfast foods, b) meat, poultry, and seafood-based ready to eat products, c) packaged sweet snacks and desserts, d) fats, condiments, and sauces, e) sugar and artificially-sweetened beverages, f) dairy-based desserts, g) ready-to-eat/heat mixed dishes, h) packaged savory snacks, and i) ultra-processed artificial sweeteners. (Supplementary Table 4S)

UPF intake was calculated and analyzed as three different exposure assessments: 1) the number of servings of UPF per day, 2) the absolute number of grams and ml of UPF intake per day, and 3) the percentage of total intake in grams and milliliters per day represented by UPF. Dietary exposure variables were included in models as cumulative averages to reflect long-term dietary habits and to reduce measurement error³⁹. Nine food items lacked sufficient information for classification (popcorn, soy milk, pancakes or waffles, pie, beef, pork, lamb sandwich, and tomato sauce). These nine items were categorized as non-UPF in the primary categorization and as UPF in a sensitivity analysis³⁸. Further details on Nova food processing categorization are available elsewhere^16,22,38.

Case Ascertainment

Incident doctor-diagnosed SLE was self-reported and then confirmed using a two-stage case validation procedure as previously described⁶. Participants who reported any new- doctor diagnosed rheumatic connective tissue disease, including SLE, on one of the biennial questionnaires were asked to complete the Connective Tissue Disease Screening Questionnaire (CSQ) and a medical record release form. Medical records of those participants screening positive for SLE on the CSQ were then assessed by two board-certified rheumatologists, who confirmed the self-reported diagnosis the 1997 Updated American College of Rheumatology (ACR) criteria for SLE diagnosis^40–42. As the SLE case identification process has been ongoing since the start of the cohorts, it is not feasible to reclassify all self-reports using more recent SLE classification criteria.

Covariates

Covariates were self-reported on mailed questionnaires administered every two years. Race was self-reported in both cohorts in categories. Household income was derived from the U.S. Census-tract median income and classified as median household income ≥ $60,000.00 USD (vs < $60,000.00 USD) per year. Smoking status was categorized as never or distant past (> 4 years) vs. current smoker or recent quitter (≤ 4 years), based on past association with SLE risk in this cohort⁴. Body mass index (BMI) was classified according to the World Health Organization categories as underweight (< 18.5 kg/m²), healthy (18.5 to < 25 kg/m²), overweight (25 to <30 kg/m²), and obese (≥ 30.0 kg/m²). For our BMI-stratified analysis, participants were grouped as underweight / healthy (< 25 kg/m²) and overweight / obese (≥ 25.0 kg/m²) by their time-varying updated BMI. Recreational physical activity was included as a continuous variable in weekly metabolic equivalents (METs per week). Age at menarche was classified as early menarche (≤ 10 years) versus > 10 years, again based on the risk previously identified in these cohorts⁶. Participants were classified according to their self-reported race (White vs. not White given small numbers of those reporting non-White race), updated use of oral contraceptives, and menopausal hormone status (pre-menopausal, post-menopausal hormone never use, post-menopausal current use, and post-menopausal past use). Alcohol consumption and total caloric intake were assessed on the SQFFQs. Alcohol intake was converted into grams per day (13.2 grams per bottle or can of beer, 10.8 grams per glass of wine, and 15.1 grams per standard drink of liquor), and total alcohol intake was the sum of grams of beer, wine, and liquor³. (No alcohol was included in the ultra-processed foods indicator for this analysis.)

Statistical Analyses

Long-term dietary intake was assessed using the cumulative average of daily UPF intake to reduce within-person variation, to represent a regular level of exposure, and to define a coherent time of exposure preceding the SLE onset. Cumulative average of UPF intake was calculated by averaging repeated dietary intake measures from baseline up to 2 to 4 years before SLE diagnosis. For example, to predict SLE incidence in the cycle of 1994–1998, the cumulative UPF intake was calculated between 1980 and 1990, excluding the 1994 measure, which was the most recent cycle of exposure [(1980 + 1984 + 1986 +1990)/4 assessments]. Cumulative average UPF intake was ranked in tertiles to compare the highest to the lowest tertile, using the lowest tertile (T1) as the reference group. Tertiles were derived within cohort and period to preserve differences and did not overlap within cohort. Tertile ranges reported here using minimum and maximum of ranges across both cohorts. We assessed associations of incident SLE with cumulative average UPF intake quantified in three ways: a) UPF servings per day, b) total intake (gms + mls) per day, and c) percentage of total intake per day.

We employed time-varying Cox regression models separately in the NHS and NHSII cohorts and then in combined pooled cohort data to estimate multivariable-adjusted hazard ratios (HRs) with 95% confidence intervals (95% CIs) for incident SLE. Multivariable models adjusted for age, race, age at menarche, and updated body mass index, menopausal status, total daily caloric intake, smoking, alcohol intake, oral contraceptive use, postmenopausal hormone use, and U.S. census tract median household income. We assessed risk of overall SLE and of SLE with and without anti-double-stranded DNA (dsDNA positive or negative) at the time of diagnosis. We assessed significance of trend in the HRs across increasing tertiles of UPF intake. To investigate potential effect modification by BMI, we conducted stratified analyses among participants who were underweight / healthy (< 25 kg/m²) and overweight / obese (≥ 25.0 kg/m²), adjusting for total caloric intake. We tested for multiplicative interactions between time-updated BMI categories and increasing UPF intake tertiles in the association with risk of SLE. Sensitivity analyses of all three assessments of daily UPF intake were performed to assess the alternative categorization of nine difficult to classify foods listed above as UPF³⁸. To investigate whether women changed their UPF intake with the onset of their SLE, we calculated mean UPF intake in servings per day among these women who developed SLE in the periods prior to SLE diagnosis and compared it to that in first three questionnaire cycles in which the FFQ was administered following diagnosis (mean UPF in servings/day, compared with t test). Additionally, we conducted analyses to examine the nine UPF groups separately in the relationship between cumulative updated servings per day and risk of incident SLE, in the same fully adjusted Cox models, comparing the highest T3 to T1 of intake. A p-value threshold of less than 0.05 indicated statistically significant association of UPF with SLE incidence. Statistical analyses were performed using SAS software, version 9.4.

All aspects of this study were approved by the Mass General Brigham Institutional Review Board.

RESULTS

We studied 98,614 women followed from 1984 to 2016 from the NHS and 105,561 women followed from 1991 to 2015 from the NHSII. The study population characteristics according to the lowest and highest tertile of UPF intake in the baseline years are shown in Table 1. The average age was ~50 years old in NHS and ~36 years old in NHSII across all tertiles of UPF consumption. Participants predominantly self-reported White race (>90%), with slightly fewer women reporting non-White race in the highest tertile of UPF intake. More women in the highest UPF intake tertile were current smokers, and they had slightly higher BMI, higher daily caloric intake, less physical activity, lower median household income, and higher use of oral contraceptives than did the women in the lower tertiles.

Table 1.

Age-standardized characteristics in Nurses’ Health Study (NHS) 1984 and Nurses’ Health Study II (NHS II) 1991, according to Ultra-Processed Food (UPF) Intake in Servings per Day in Tertiles^* (N=169,910)

	Tertile 1	Tertile 2	Tertile 3
No. of participants	56,549	56,683	56,678
Age, years, mean (SD) **	42.92 (9.29)	42.45 (9.25)	42.25 (9.15)
White race, % ¥	91.30	93.90	94.72
Never smoker, %	55.02	56.59	55.39
Past smoker (> 4 years), %	21.45	21.21	20.75
Current smoker or quit < 4 years, %	23.53	22.20	23.86
Body mass index, kg/m2, mean (SD)	24.22 (4.58)	24.73 (4.93)	25.54 (5.58)
Physical activity, ≥ 19 MET-h/week, mean (SD)	42.95 (32.95)	41.30 (31.31)	40.97 (28.49)
Alcohol consumption, g/day, mean (SD)	5.01 (8.59)	4.61 (7.92)	4.42 (7.79)
Total calorie intake, kcal/day, mean (SD)	1,407 (375)	1,715 (411)	2,065 (508)
Early menarche (≤10 years), %	6.92	6.69	7.25
Oral contraceptive ever or current use, %	68.14	68.58	68.75
Premenopausal hormone use, %	71.41	71.30	71.46
Postmenopausal or no use, %	14.24	14.46	14.33
Postmenopausal hormone or ever use, %	13.26	13.17	13.12
Median household income ≥ $60k, %	18.97	17.18	15.36

Means (SD) or medians (Q25, Q75) for continuous variables; Percentages or ns or both for categorical variables, standardized to age distribution of study population.

^*Tertiles were derived within cohort and period to preserve differences and did not overlap within cohort. Tertile ranges (T1: 0–5, T2: 5–7, T3: 7–28 servings/day) use the minimum and maximum of ranges across both cohorts.

Values of polytomous variables may not sum to 100% due to rounding

^**Not age-adjusted

We identified 95 incident cases of SLE from 1984 to 2016 in NHS and 117 incident cases in NHSII from 1991 to 2017. In age-adjusted models, UPF intake in servings/day was more strongly associated with SLE among the women followed in the NHSII cohort, but in the pooled analyses of the two cohorts, a significant association was demonstrated for the highest vs. the lowest tertile of UPF in number of servings per day with overall SLE diagnosis (pooled multivariable hazard ratio [MV HR] 1.41; 95% CI 1.01–1.97; p trend across categories 0.04) (Table 2). After multivariable adjustment, the association of UPF in servings per day and overall SLE remained significant (Tertile 3 [T3] vs. T1 pooled MV HR 1.56, 95% CI 1.04–2.32; p trend across categories 0.03). The association was stronger for the risk of anti-dsDNA antibody positive SLE: for women in the highest vs. lowest tertile of UPF in servings per day the pooled MV HR was 2.05; 95% CI 1.15–3.65, p trend 0.01, and no significant association was found between UPF in servings per day and SLE without anti-dsDNA antibodies.

Table 2.

Risks (Hazard Ratios with 95% Confidence intervals) of Developing Systemic Lupus Erythematosus (SLE) among Women in the Nurses’ Health Study (NHS) and Nurses’ Health Study II (NHS II), according to Cumulatively-Updated Ultra-Processed Food Intake in Servings per Day in Tertiles, Overall SLE and SLE with or without Anti-Double Stranded DNA Antibodies

	NHS				NHSII				Pooled
	Tertile 1	Tertile 2	Tertile 3	p trend	Tertile 1	Tertile 2	Tertile 3	p trend	Tertile 1	Tertile 2	Tertile 3	p trend
SLE
No. of cases	27	31	37		32	37	48		59	68	85
Person-years	778,118	796,014	781,832		790,373	801,253	787,817		1,568,491	1,597,267	1,569,649
Age-adjusted Model	1.00 (ref)	1.10 (0.66–1.85)	1.29 (0.79–2.13)	0.30	1.00 (ref)	1.14 (0.71–1.83)	1.51 (0.97–2.37)	0.06	1.00 (ref)	1.12 (0.79–1.59)	1.41 (1.01–1.97)	0.04
Multivariable-adjusted Model	1.00 (ref)	1.21 (0.71–2.08)	1.48 (0.82–2.68)	0.19	1.00 (ref)	1.18 (0.72–1.94)	1.63 (0.95–2.81)	0.07	1.00 (ref)	1.19 (0.82–1.71)	1.56 (1.04–2.32)	0.03
Anti-dsDNA Positive SLE
No. of cases	10	13	17		16	18	26		26	31	43
Person-years	777,910	795,783	781,672		790,220	801,009	787,574		1,568,130	1,596,792	1,569,246
Age-adjusted Model	1.00 (ref)	1.23 (0.54–2.82)	1.62 (0.74–3.54)	0.22	1.00 (ref)	1.10 (0.56–2.17)	1.61 (0.86–3.01)	0.11	1.00 (ref)	1.15 (0.68–1.95)	1.61 (0.99–2.63)	0.04
Multivariable-adjusted Model	1.00 (ref)	1.49 (0.63–3.51)	2.19 (0.88–5.48)	0.09	1.00 (ref)	1.22 (0.60–2.46)	1.98 (0.94–4.18)	0.06	1.00 (ref)	1.32 (0.77–2.27)	2.05 (1.15–3.65)	0.01
Anti-dsDNA Negative SLE
No. of cases	17	18	20		16	19	22		33	37	42
Person-years	777,977	795,837	781,593		790,176	801,023	787,500		1,568,153	1,596,860	1,569,093
Age-adjusted Model	1.00 (ref)	1.02 (0.53–1.99)	1.10 (0.58–2.11)	0.76	1.00 (ref)	1.18 (0.61–2.30)	1.41 (0.74–2.70)	0.29	1.00 (ref)	1.10 (0.69–1.76)	1.25 (0.79–1.98)	0.33
Multivariable-adjusted Model	1.00 (ref)	1.07 (0.53–2.16)	1.12 (0.51–2.44)	0.79	1.00 (ref)	1.12 (0.56–2.25)	1.31 (0.60–2.86)	0.50	1.00 (ref)	1.07 (0.65–1.75)	1.19 (0.69–2.07)	0.53

Tertile 1 as reference category, with p for trend across categories. Tertiles were derived within cohort and period to preserve differences and did not overlap within cohort. Multivariable models adjusted for age, cohort, questionnaire cycle, median household income, smoking status, body mass index, physical activity, alcohol consumption, total calorie intake, age at menarche, and oral contraceptive use.

When examining total intake of UPF per day in grams and milliliters, we found similarly strong results for overall SLE (T3 vs. T1 pooled MV HR 1.60; 95% CI 1.10–2.32, p trend 0.02), as well as for dsDNA positive SLE (T3 vs. T1 pooled MV HR 1.73; 95% CI 1.01–2.98, p trend 0.05) (Table 3). When examined as a percentage of total intake in grams and milliliters per day, however, the association of UPF intake with risk of SLE was somewhat lower (T3 vs. T1 pooled MV HR 1.41; 95% CI 1.00–1.99, p trend 0.04), with no significant associations with either dsDNA positive or negative SLE (Table 4).

Table 3.

Risks (Hazard Ratios with 95% Confidence intervals) of Developing Systemic Lupus Erythematosus (SLE) among Women in the Nurses’ Health Study (NHS) and Nurses’ Health Study II (NHS II), according to Cumulatively-Updated Ultra-Processed Food Intake in Grams and Milliliters per Day in Tertiles, Overall SLE and SLE with or without Anti-Double Stranded DNA Antibodies

	NHS				NHSII				Pooled
	Tertile 1	Tertile 2	Tertile 3	p trend	Tertile 1	Tertile 2	Tertile 3	p trend	Tertile 1	Tertile 2	Tertile 3	p trend
SLE
No. of cases	21	35	39		29	39	49		50	74	88
Person-years	771,557	794,565	789,842		802,847	804,529	772,067		1,574,404	1,599,094	1,561,909
Age-adjusted Model	1.00 (ref)	1.42 (0.82–2.45)	1.50 (0.87–2.59)	0.20	1.00 (ref)	1.34 (0.83–2.18)	1.76 (1.11–2.79)	0.02	1.00 (ref)	1.38 (0.96–1.98)	1.64 (1.15–2.34)	<0.01
Multivariable-adjusted Model	1.00 (ref)	1.48 (0.84–2.59)	1.52 (0.85–2.71)	0.24	1.00 (ref)	1.31 (0.80–2.14)	1.65 (1.01–2.68)	0.05	1.00 (ref)	1.39 (0.96–2.00)	1.60 (1.10–2.32)	0.02
Anti-dsDNA Positive SLE
No. of cases	9	13	18		14	21	25		23	34	43
Person-years	771,346	794,367	789,652		802,654	804,304	771,844		1,574,000	1,598,671	1,561,496
Age-adjusted Model	1.00 (ref)	1.12 (0.47–2.65)	1.50 (0.66–3.38)	0.29	1.00 (ref)	1.49 (0.76–2.94)	1.84 (0.95–3.55)	0.08	1.00 (ref)	1.34 (0.79–2.29)	1.71 (1.02–2.85)	0.04
Multivariable-adjusted Model	1.00 (ref)	1.23 (0.51–2.95)	1.63 (0.69–3.86)	0.25	1.00 (ref)	1.49 (0.75–2.96)	1.81 (0.91–3.62)	0.11	1.00 (ref)	1.39 (0.81–2.39)	1.73 (1.01–2.98)	0.05
Anti-dsDNA Negative SLE
No. of cases	12	22	21		15	18	24		27	40	45
Person-years	771,450	794,330	789,626		802,688	804,282	771,730		1,574,138	1,598,612	1,561,356
Age-adjusted Model	1.00 (ref)	1.66 (0.81–3.36)	1.49 (0.72–3.10)	0.44	1.00 (ref)	1.20 (0.60–2.39)	1.68 (0.88–3.22)	0.10	1.00 (ref)	1.41 (0.87–2.31)	1.59 (0.98–2.58)	0.09
Multivariable-adjusted Model	1.00 (ref)	1.65 (0.79–3.43)	1.41 (0.65–3.07)	0.62	1.00 (ref)	1.15 (0.57–2.33)	1.50 (0.76–2.99)	0.22	1.00 (ref)	1.37 (0.83–2.26)	1.46 (0.88–2.45)	0.20

Tertile 1 as reference category, with p for trend across categories. Tertiles were derived within cohort and period to preserve differences and did not overlap within cohort.

Multivariable models adjusted for age, cohort, questionnaire cycle, median household income, smoking status, body mass index, physical activity, alcohol consumption, total calorie intake, age at menarche, and oral contraceptive use.

Table 4.

Risks (Hazard Ratios with 95% Confidence intervals) of Developing Systemic Lupus Erythematosus (SLE) among Women in the Nurses’ Health Study (NHS) and Nurses’ Health Study II (NHS II), according to Cumulatively-Updated Ultra-Processed Food Percent of Total Grams and Milliliters per Day in Tertiles, Overall SLE, and SLE with or without Anti-Double Stranded DNA Antibodies

	NHS				NHSII				Pooled
	Tertile 1	Tertile 2	Tertile 3	p trend	Tertile 1	Tertile 2	Tertile 3	p trend	Tertile 1	Tertile 2	Tertile 3	p trend
SLE
No. of cases	25	31	39		31	36	50		56	67	89
Person-years	776,885	793,234	785,845		802,694	804,578	772,171		1,579,579	1,597,812	1,558,016
Age-adjusted Model	1.00 (ref)	1.09 (0.64–1.86)	1.26 (0.75–2.11)	0.37	1.00 (ref)	1.16 (0.72–1.88)	1.68 (1.07–2.65)	0.02	1.00 (ref)	1.13 (0.79–1.62)	1.48 (1.06–2.09)	0.02
Multivariable-adjusted Model	1.00 (ref)	1.13 (0.66–1.95)	1.25 (0.74–2.12)	0.42	1.00 (ref)	1.10 (0.68–1.79)	1.54 (0.97–2.44)	0.05	1.00 (ref)	1.12 (0.78–1.60)	1.41 (1.00–1.99)	0.04
Anti-dsDNA Positive SLE
No. of cases	13	9	18		11	26	23		24	35	41
Person-years	776,682	793,032	785,651		802,488	804,368	771,947		1,579,170	1,597,400	1,557,598
Age-adjusted Model	1.00 (ref)	0.55 (0.23–1.31)	1.00 (0.48–2.08)	0.72	1.00 (ref)	2.37 (1.17–4.80)	2.21 (1.07–4.55)	0.07	1.00 (ref)	1.35 (0.80–2.28)	1.56 (0.94–2.60)	0.10
Multivariable-adjusted Model	1.00 (ref)	0.58 (0.24–1.39)	0.99 (0.47–2.10)	0.75	1.00 (ref)	2.27 (1.12–4.64)	2.05 (0.98–4.26)	0.12	1.00 (ref)	1.35 (0.79–2.28)	1.47 (0.87–2.47)	0.18
Anti-dsDNA Negative SLE
No. of cases	12	22	21		20	10	27		32	32	48
Person-years	776,710	793,087	785,610		802,551	804,291	771,857		1,579,261	1,597,378	1,557,467
Age-adjusted Model	1.00 (ref)	1.71 (0.84–3.48)	1.52 (0.73–3.17)	0.39	1.00 (ref)	0.50 (0.23–1.07)	1.39 (0.78–2.49)	0.12	1.00 (ref)	0.97 (0.59–1.58)	1.43 (0.90–2.26)	0.08
Multivariable-adjusted Model	1.00 (ref)	1.77 (0.86–3.64)	1.49 (0.71–3.15)	0.46	1.00 (ref)	0.47 (0.22–1.02)	1.26 (0.69–2.29)	0.21	1.00 (ref)	0.95 (0.58–1.56)	1.35 (0.85–2.15)	0.14

Tertile 1 as reference category, with p for trend across categories. Tertiles were derived within cohort and period to preserve differences and did not overlap within cohort.

Multivariable models adjusted for age, cohort, questionnaire cycle, median household income, smoking status, body mass index, physical activity, alcohol consumption, total calorie intake, age at menarche, and oral contraceptive use.

Although we hypothesized that UPF intake might have decreased among women after the onset of SLE, among the 193 women with these data (missing for 19 women), we found that 52% had a decrease in UPF intake, whereas 48% had stable or increase UPF intake. However, when examined as mean UPF servings per day as a continuous value, the difference between before vs. after was −0.53 (95% CI −0.48- + 0.37, p 0.81) servings per day.

In BMI-stratified analyses to investigate whether the association of UPF intake with risk of SLE differed by time-varying updated BMI, multivariable models revealed that overall SLE risk was most elevated in the highest vs. lowest UPF tertile intake in servings per day, but non-significantly so, among the overweight / obese participants (T3 vs. T1 pooled MV HR 1.65; 95% CI 0.94–2.89, p trend 0.08; Figure 1). However, we did not detect any formal multiplicative interactions across the strata of BMI (all p values for interaction were >0.05 for tertiles of UPF and updated BMI on risk of SLE). And for none of the three assessments of UPF intake were interactions with BMI seen for risk of dsDNA positive or negative SLE (data not shown).

Figure 1.

Multivariable Cox Regression Models for Ultraprocessed Food (UPF) Intake and Risk of Incident SLE among Women in the Nurses’ Health Study Cohorts (N=169,910)

In sensitivity analyses including the nine SQFFQ food items that were difficult to classify by the Nova framework as UPF (not included as UPF in the main analyses), we found somewhat attenuated results for the measures of cumulatively updated UPF intake (e.g. servings/day T3 vs. T1 pooled MV HR 1.34; 95% CI 0.88–2.03, p trend 0.19; but stronger for dsDNA positive SLE 1.64 95% CI 0.91–2.97, p trend 0.09). (Tables 1S–3S). The results of the nine UPF food groups separately in cumulative updated servings per day and SLE risk in the pooled NHS cohorts, in the same MV Cox models, comparing T3 vs. T1 are shown in Table 4S. In particular, sugar-sweetened and artificially-sweetened beverages were most strongly associated with the risk of developing SLE, although smaller sample sizes in each category likely precluded any significant findings (T3 vs. T1 pooled MV HR: 1.45 95% CI 1.01–2.09, p trend 0.07).

DISCUSSION

Intake of many types of UPF, low in fiber, antioxidants and vitamins, but high in salt, saturated fat, processed sugar, stabilizers, and emulsifiers, increases systemic inflammation, as well as risks of obesity, diabetes, cancer, cardiovascular disease, and mortality^{13,16,19,22–26}. In this prospective longitudinal analysis, we found evidence of a positive association between higher daily consumption of UPF and higher SLE incidence, which was stronger among the younger women in more recent years followed in NHSII, but present when the cohorts were examined together as well. The risk was also somewhat stronger for dsDNA positive SLE and among those with high BMI, although no formal interactions were seen. To our knowledge, this is the first analysis of UPF intake and risk of developing SLE and the results have strong biologic plausibility. Prior analyses using the Nova categorization of UPF in the NHS cohorts have demonstrated multiple outcomes, including inflammatory bowel disease, diabetes, colorectal cancer, and even incident depression^{19,22,24–26}. Among categories of UPF, diabetes and to a certain extent depression were most strongly associated with artificially and sugar-sweetened beverages, animal-based products, and ready-to-eat mixed dishes^24,25. In relation to the risk of developing SLE, we also found that both artificially and sugar-sweetened beverages appeared to be most strongly associated, in keeping with our past finding that sugar-sweetened soda was associated with an increased risk of rheumatoid arthritis among women followed in these cohorts⁴³. We investigated whether UPF in servings per day changed after diagnosis of SLE among the women with new onset disease, but did not find a significant difference.

While there is increasing acknowledgement that diet and nutrient consumption have important influences on inflammation and disease progression among patients with SLE, as well as on the risk of developing disease, there are only a few prospective studies of diet and risk of SLE to date^44,45. In past NHS cohort studies, we have found no associations between adherence to higher diet quality scores, including the Alternate Healthy Eating Index (AHEI-2010), the alternate Mediterranean diet (aMed), the DASH diet^30,32, and risk of SLE. In those past studies, we also found no significant associations between the Empirical Dietary Inflammatory Pattern and the Western vs. prudent food patterns^30,32. We did find, however, that women in the highest AHEI-2010 tertile of nut/legume intake had a 41% lower SLE risk (HR 0.59; CI 0.40–0.87) compared women in the lowest tertile³⁰. Moreover, in the Black Women’s Health Study, women in the highest quintile of carbohydrate intake had a 96% higher risk of developing SLE than those in the lowest quintile, and fried foods such as fried chicken were associated with increased risk³¹. Dietary intake and UPF in particular in relation to risk of SLE or other autoimmune connective tissue disease have not yet been examined in other prospective cohorts.

Basic and animal studies of SLE pathogenesis have identified several important mechanisms that are influenced by environmental exposures, potentially including diet. These include oxidative stress and mitochondrial dysfunction, gut microbiome alterations, metabolites interacting with the aryl hydrocarbon receptor and other sensors that influence immune cell function and phenotypes, and epigenetic changes induced by diet^46–49. Interestingly, in a transgenic murine model of SLE, those fed a micronutrient (especially methyl)-depleted diet developed demethylation of the second X chromosome and lupus nephritis symptoms, whereas those fed a micronutrient-sufficient diet did not⁵⁰. It is plausible that micronutrient-poor polysaccharides in UPF may induce deleterious epigenetic changes⁵¹. Moreover, UPF contains higher “pro-inflammatory” content of sugar, carbohydrate, saturated fat, and sodium^52–54. UPF processing also adds many non-nutritional ingredients including emulsifiers, stabilizers, and artificial colors, flavors and sweeteners, including simple sugars and high fructose corn syrup, known to be toxic to the gut epithelia and to influence gastrointestinal microbiomes, and to stimulate inflammatory pathways, obesity, metabolic syndrome and type II diabetes^54–58. Emulsifiers, such as lecithins, mono- and diglycerides, polysorbates, carrageenans, guar gums, and carboxymethylcellulose, are biologically and metabolically harmful, affecting the gut microbiota, and implicated in causing “leaky gut”, with resulting systemic inflammation and metabolic syndrome, while additives such as nitrates in processed meats are both inflammatory and carcinogenic^58–60. UPF consumption, production of which has been supported in large part by government subsidies to the U.S. agrifood industry, has been increasing at an alarming rate, and now represents well over half of total caloric intake in the average U.S. diet^15,61,62. In fact, average UPF intake increased from 53% to 57% of caloric intake for U.S. adults from 2001–2018 (p trend < 0.001) with alarming disparities: the highest intake is seen among older Americans, African Americans, and those with lower education (representing > 60% of total daily calories on average among those without high school education)^15,63.

Obesity as a result of high UPF intake may also increase SLE risk by triggering a systemic inflammation cascade and increasing adipocyte-produced sex hormones and adipokines, and we have previously found obesity to be positively associated with SLE risk^7,10,27. In the current study, the risk of SLE was 65% higher among women with overweight or obesity (BMI ≥ 25kg/m²), who were in the highest tertile of UPF intake, but we did not observe a formal statistical interaction as there was also an increased risk of SLE among those with BMI < 25 kg/m² associated with increasing cumulative intake of UPF. The relationship between UPF dietary intake and obesity in influencing SLE risk definitely deserves further study.

The NHS cohorts both enrolled almost entirely self-reported White U.S. female registered nurses, restricting the generalizability of the findings to other populations, including people of more diverse racial and ethnic backgrounds and males, and similar studies should be performed in other populations, if possible, to validate and extend these results. Importantly, not only do individuals of African descent have higher risk of SLE and more severe SLE with higher mortality, but in the U.S., African American populations have also been found to consume more sugar-sweetened beverages, and more fast food and UPF than do White population^1,15. Moreover, due to the cohort structure, the ages at enrollment the baseline age (~50 years in NHS; ~36 years in NHSII) were relatively high for SLE, which has peak incidence at ages 15–45⁵. Thus, again we suggest that these studies should be replicated in younger and more diverse cohorts. Additionally, UPF intake is highest in the same population groups that may have multiple other risk factors for SLE and other chronic diseases, including obesity, stress, poverty, and cigarette smoking^15,52. Although we have adjusted our analyses for all available known confounders among women in these cohorts, there may in fact be a conflation or synergy of risk factors in these high-risk populations. As UPF intake has also been associated with a wide range of adverse health outcomes, there may be competing risks of cancer, cardiovascular disease, and early mortality that we have not taken into account, leading to an underestimation of the magnitude of the UPF and SLE risk relationship.

The large sample size and long period of prospective exposure and detailed covariate assessment prior to the onset of disease are strengths of this study. Despite the large size of the cohorts, SLE was a relatively rare disease in these populations, limiting our statistical power to detect smaller effect sizes and to perform subgroup analyses. To our knowledge, this is the first study assessing the association of UPF intake with SLE risk. We found that high cumulative daily intake of UPF, in particular as expressed as an absolute intake value (total grams and milliliters/day vs. relative percentage/day) was associated with SLE development. The risk of SLE was 56% higher, and that of anti-dsDNA positive SLE was > double, for women in the highest tertile of UPF intake vs. lowest tertile of daily UPF servings. The strong association with anti-dsDNA antibody positive SLE lends biologic plausibility as individuals with anti-dsDNA antibodies tend to have severe disease and organ involvement^1,5. The potential association of SLE with sugar-sweetened and artificially sweetened beverages was intriguing and deserves further study as well.

SIGNIFICANCE AND INNOVATIONS.

• Past studies of dietary intake and patterns and the subsequent risk of developing systemic lupus erythematosus (SLE) have been largely unrevealing, except for a potential relationship between high nut intake and decreased risk of SLE, and high carbohydrate intake with increased risk.

• No prior studies of the relationship between ultra-processed food (UPF) intake and risk of SLE have been performed, although UPF intake has been linked epidemiologically to increased risks of obesity, type 2 diabetes, cancer, cardiovascular disease, and overall mortality.

• In these two large cohorts of women without SLE followed prospectively for many years who completed detailed dietary and covariate assessment questionnaires biennially, we found that SLE risk was over 50% higher in the highest vs. lowest tertile of cumulative UPF intake over time.

• Our results were stronger for the risk of anti-dsDNA SLE, with a doubling of risk observed, and among individual UPF foods, sugar/artificially-sweetened beverages were associated with a 45% increased SLE risk, but we did not detect a formal interaction with body mass index.