Influence of Alcohol Consumption on the Risk of Systemic Lupus Erythematosus among Women in the Nurses’ Health Study Cohorts

Abstract

Objective

Moderate alcohol consumption has anti-inflammatory properties and is associated with reduced cardiovascular disease and rheumatoid arthritis risks. We investigated the association between alcohol consumption and SLE risk among women followed in the Nurses’ Health Study (NHS) cohorts.

Methods

We conducted a prospective cohort analysis among 204,055 women in NHS (1980–2012) and NHSII (1989–2011), free of connective tissue disease and providing alcohol information at baseline. Alcohol consumption was assessed using a semi-quantitative food frequency questionnaire every 2–4 years. We validated incident SLE through medical record review after self-report. Cox proportional hazards models estimated hazard ratios for SLE based on cumulative average alcohol intake, adjusting for potential confounders. Results were meta-analyzed using DerSimonian and Laird random effects models. We further investigated SLE risk associated with wine, beer, and liquor intake.

Results

We identified 125 incident SLE cases in NHS and 119 in NHSII. Mean age at SLE diagnosis was 55.8 (SD 9.5) years in NHS and 43.4 (SD 7.7) years in NHSII. Compared to no alcohol intake, the meta-analyzed multivariable HR for cumulative alcohol consumption ≥5 grams/day was 0.61 (95%CI 0.41–0.89). When limiting alcohol exposure to >4 years prior to SLE diagnosis, the multivariable HR was similar, 0.61 (95%CI 0.41–0.91). Women who drank ≥2 servings/week of wine had significantly decreased SLE risk (HR 0.65; 95%CI 0.45–0.96), compared to women who did not drink wine.

Conclusion

In these large prospective cohorts, we demonstrated an inverse association between moderate alcohol consumption (≥5 grams or 0.5 drink/day) and SLE risk in women.

Introduction

Environmental exposures likely play an important role in the development of systemic lupus erythematosus (SLE) within genetically-susceptible individuals (1, 2). Oral contraceptives, postmenopausal hormone use, early menarche, and crystalline silica exposure have been associated with increased SLE risk, while other potentially modifiable environmental factors, including alcohol consumption, have not been well studied(3–6).

Alcohol contains components (e.g., ethanol and antioxidants) that suppress systemic inflammation and decrease the synthesis of pro-inflammatory cytokines, such as tumor necrosis factor (TNF), interleukin (IL)-6, and IL-8 (7). Alcohol intake has been associated with reduced cardiovascular disease and rheumatoid arthritis risks (8, 9). Past epidemiologic studies of alcohol consumption and SLE risk have had conflicting results, possibly due to variation in the definitions of alcohol intake and SLE, inclusion of both incident and prevalent SLE, lack of repeated measurements, and biases inherent in the case-control study design.

We aimed to investigate the association between alcohol consumption and risk of developing SLE among women using cumulative average intake to represent long-term alcohol consumption. We hypothesized that long-term moderate alcohol consumption, compared to no alcohol consumption, is associated with decreased risk of SLE among women.

PATIENTS AND METHODS

Study Population

The Nurses’ Health Study (NHS) cohort was established in 1976 and enrolled 121,700 female registered nurses, aged 30 to 55 years, residing in 11 large U.S. states. NHSII began in 1989 and enrolled 116,670 female registered nurses, aged 25 to 42 years, residing in 14 U.S. states. Based on self-report, both cohorts are predominantly White (>90%), reflecting the ethnic background of women entering the nursing profession in the U.S. during the recruitment years. In both NHS and NHSII, all women completed an initial questionnaire and have been followed biennially to update exposures, lifestyle, health practices, and disease diagnoses. Response rates to follow-up questionnaires are >90% in each cycle, and only 5% of person-time has been lost to follow-up (10). Deaths are reported by participant family members and also ascertained via National Death Index searches. Causes of death are validated by medical record review. All aspects of this study were approved by the Partners’ HealthCare Institutional Review Board.

Identification of Incident SLE

The primary endpoint was the diagnosis of confirmed SLE. SLE case identification in the NHS cohorts is a two-stage procedure previously described, involving the connective tissue screening questionnaire and medical record review by two independent rheumatologists (4, 11). “Definite” SLE cases were those fulfilling at least four ACR SLE classification criteria and confirmed by medical record review(12, 13). “Probable” SLE consisted of 3 ACR criteria for SLE and both reviewers’ consensus of SLE diagnosis. In a secondary analysis, we evaluated the association between alcohol consumption and definite or probable SLE.

Alcohol Consumption

Alcohol consumption was assessed using a semi-quantitative food frequency questionnaire (FFQ), with separate items for beer, wine, and liquor intake. A previous study showed that correlation of alcohol intake on the FFQ with alcohol intake on dietary records was high (r=0.9) (14). In NHS, alcohol consumption was assessed using the FFQs in 1980 (baseline for this study), 1984, 1986, 1990, 1994, 1998, 2002, 2006, and 2010. In NHSII, alcohol data were collected on the 1989 questionnaire (baseline for this study) and FFQs in 1991, 1995, 1999, 2003, and 2007. Standard portions for each beverage type were defined as a glass, bottle, or can of beer; a 4-oz glass of wine; and a shot of liquor. The estimated alcohol content of each beverage was 13.2 g per bottle or can of beer, 10.8 g per glass of wine, and 15.1 g per standard drink of liquor. For each beverage type, participants were asked to estimate their average consumption over the past year. Response categories in NHS included: “almost never”, “1–3/month”, “1/week”, “2–4/week”, “5–6/week”, “2–3/day”, “4–6/day”, “6+/day”. Similar response categories were used in NHSII. Alcohol consumption was calculated as the amount of alcohol in grams per day (g/day) and number of drinks per week from each FFQ. Total alcohol intake was the sum of these three beverages. All women in our analysis were required to provide data on alcohol consumption on the baseline questionnaire.

We used cumulative average estimates of alcohol consumption which reduce within-person variation, minimize the inaccuracy of exposure information, and are more representative of regular alcohol exposure than a one-time alcohol measure (15). Cumulative average intake was calculated by averaging repeated alcohol measures from baseline up to two to four years before SLE diagnosis. Since early SLE symptoms may alter alcohol intake, we excluded the most recent FFQ before SLE onset to reduce the possibility of spurious associations from reverse causation. For example, for SLE incidence during the 1998 to 2000 exposure window, we averaged consumption from baseline until the 1994 FFQ (excluding the 1998 measure). We examined the possibility of a non-linear relationship between cumulative alcohol consumption and the hazard ratio of SLE, non-parametrically with restricted cubic splines (16). Tests for non-linearity used the likelihood ratio test, comparing the model with only the linear term to the model with the linear and the cubic spline terms. In our primary analysis, based on the results of the restricted cubic splines and for ease of interpretability, we defined alcohol consumption using an ordinal variable (0 [reference], >0 to <5 g/day, ≥5 g/day). Drinking in moderation for women has been defined by the Dietary Guidelines for Americans as ≤1 drink (10–15 g/day of pure alcohol) (17). In a secondary analysis, we also evaluated alcohol intake by alcoholic beverage type, including beer, wine, and liquor in cumulative average number of drinks/day.

Study Covariates

Time-varying data on covariates were assessed by self-report on biennial questionnaires. We selected potential confounders based on prior studies in NHS cohorts, or other studies showing associations between diet or lifestyle and alcohol intake. We included age, race and questionnaire cycle, in addition to the total daily energy intake (total kilocalories per day), measured by the FFQ. Given the high correlation of smoking with alcohol intake, we examined adjustment for time-varying smoking status in multiple ways, including as never/past/current smokers and continuous pack-years of smoking. Reproductive covariates including oral contraceptive (OCP) use and menopausal status and postmenopausal hormone use were all examined as potential confounders and included if they were related to alcohol intake. We included the U.S. Census tract-based median household income as a measure of area socioeconomic status (SES) and also examined husband’s educational level (categorized as <high school, high school, college, >college) as a proxy for individual SES.

Statistical Analysis

We assessed the relationship between cumulative average updated alcohol consumption and incident SLE. For these analyses, in order to define an SLE-free cohort at the analysis baseline, we excluded all participants who reported prevalent SLE or other connective tissue diseases (CTD) at study baseline (1980 in NHS and 1989 in NHSII). Person-years of follow-up accrued from the return date of the baseline questionnaire to date of SLE diagnosis, the end of follow-up, death, or date of censor, whichever came first. Participants were censored for self-reported CTD not subsequently validated as SLE or loss-to-follow-up, at last questionnaire return date.

Baseline characteristics across categories of alcohol consumption in each cohort were examined. We used Cox proportional regression models to calculate the hazard ratios (HR) and 95% confidence intervals (CIs) of incident SLE associated with alcohol consumption, controlling for time-varying covariates. Missing covariates were carried forward one cycle. For missing covariate data beyond one cycle, we included a separate category for missing data. We constructed age-, questionnaire period-, and total energy-adjusted models, followed by fully-adjusted models. We tested the proportional hazards assumption by including covariate-time interactions terms in the models. Based on the generalized Wald test for a joint hypothesis on all interactions, the proportional hazards assumption was not violated. We tested for linear trends across categories of increasing alcohol intake using the median value of each alcohol category as a continuous variable to minimize the influence of outliers. Analyses were performed separately in NHS and NHSII and then hazard ratio estimates were meta-analyzed using DerSimonian and Laird random effects models(18).

In secondary analyses, we examined separate alcoholic beverage types in Cox proportional hazards models, including cumulative average consumption of beer, wine and liquor, as well as covariates above, and adjusting for other alcoholic beverage types. We measured the consumption frequency of each alcoholic beverage type using ordinal variables (0 to ≤1 drink/month [reference], >1 month to <1 drink/week, ≥2 drinks/week). In stratified analyses, we also investigated whether the effect of alcohol consumption on risk of SLE was modified by smoking or oral contraceptive use, both of which increased SLE risk in past studies(4, 6). To test for possible effect modifications, we included interaction terms for smoking (never vs. ever) and alcohol consumption (0 to <5 g/day and ≥5 g/day vs. none) and, separately for oral contraceptive use (never vs. ever) and the alcohol consumption in our multivariable models.

Sensitivity analyses were performed to assess the robustness of our results. First, we performed a lagged analysis using cumulative average alcohol consumption, allowing 4–6 years of lag time prior to SLE diagnosis (compared to 2–4 years in the primary analysis), as pre-diagnosis symptoms may influence alcohol intake. Second, to ensure that this association was not driven by outliers, we excluded “heavy” drinkers (≥30 g/day of alcohol daily) to examine whether this altered the association of cumulative average alcohol intake on SLE risk. Third, we also investigated the association of cumulative average alcohol consumption with either definite or probable SLE. We evaluated short-term alcohol consumption and SLE risk in a simple updated analysis, which used the most recent alcohol intake values. Finally, we examined whether baseline alcohol consumption was associated with SLE risk. These techniques have been used in NHS studies to assess whether the timing or duration of alcohol exposure (e.g. short-term versus cumulative exposure) is associated with incident disease (19, 20). Data analyses were performed using SAS 9.3 (SAS Institute, Inc., Cary, North Carolina, USA) with a two-sided alpha of 0.05 as statistical significance.

Results

Among 90,728 women with 2.4 million person-years of follow-up time in NHS, we identified 125 incident SLE cases occurring between 1980–2012. In the NHSII cohort, among 113,327 women with 2.2 million person-years of follow-up, we identified 119 incident SLE cases between 1989–2011. Age-adjusted incidence rates of SLE within categories of increasing daily alcohol consumption (none, >0 to <5 g/day, ≥5 g/day) for NHS were 6.3, 6.1, and 3.6 per 100,000 person-years, and for NHSII were 5.9, 5.6, and 4.3 per 100,000 person-years. Among 244 SLE cases identified, the overall incidence rate for SLE was 5.3 per 100,000 person-years, which as expected is slightly lower than recent reported incidence rates of SLE among women in the U.S. due to our stringent case definition, predominantly White subjects and ages of the cohort participants(21–23).

Age-adjusted baseline characteristics of study participants categorized by daily alcohol consumption are shown in Table 1. Across increasing categories of consumption, median family income, rates of smoking and OCP use increased in both cohorts. Proportions of women who were postmenopausal and had used postmenopausal hormones were similar across increasing levels of alcohol intake.

Table 1.

Baseline age-standardized characteristics of participants in the Nurses’ Health Study in 1980 and Nurses’ Health Study II in 1989 categorized by level of cumulative average daily alcohol intake

Characteristics	Cumulative average daily alcohol intake in grams/day (g/day)
	NHS (N=90,728)			NHSII (N=113,327)
	None	>0 to <5 g/day	≥5 g/day	None	>0 to <5 g/day	≥5 g/day
No. of participants (%)	29,074 (32.0)	30,415 (33.5)	31,239 (34.4)	42,831 (37.8%)	47,796(42.2%)	22,700 (20.0%)
Age in years, mean (SD)a	46.6 (7.3)	45.7 (7.3)	46.8 (7.0)	34.6 (4.6)	34.2 (4.7)	34.3 (4.8)
White race (%)	95.9	97.0	98.0	89.2	92.8	94.2
Median family Income ≥ $60K (%)	38.6	50.2	55.9	45.0	53.2	55.4
Husband’s education level ≥ high school (%)	62.5	66.2	67.3	76.0	78.4	78.2
Calorie intake (Kcal/day, SD)	1563.7 (516.1)	1535.4 (495.7)	1597.0 (491.1)	1469.8 (842.3)	1486.1 (833.8)	1504.8 (840.0)
Smoking status (%)
Never smokers	57.0	44.5	29.9	74.1	65.0	48.3
Past smokers	20.6	28.6	33.8	15.9	22.2	30.3
Current smokers	22.5	27.0	36.3	10.0	12.8	21.4
Oral contraceptive use, ever (%)	45.6	48.6	52.3	79.4	84.6	88.3
Postmenopausal (%)	46.8	45.0	44.8	6.9	5.9	5.3
Any postmenopausal hormone use %	18.9	18.5	19.3	3.5	3.0	2.5
Alcohol, mean (g/d, SD)b	0.0 (0.0)	1.9 (1.2)	16.6 (12.6)	0.0 (0.0)	1.8 (1.1)	11.1 (8.5)
Beer ≥ 1 serving/month (%)	0.0	6.7	15.8	0.0	36.8	73.7
Wine ≥ 1 serving/month (%)	0.0	81.2	82.1	0.0	77.2	76.8
Liquor ≥ 1 serving/month (%)	0.0	49.2	78.4	0.0	41.1	61.3

Values are means (SD) or percentages and are age-standardized to the age distribution of the study population.

^aValue is not age-standardized

^bCumulative average daily alcohol consumption

The presenting manifestations of incident SLE cases are shown in Table 2. The mean age at SLE diagnosis was 55.8 (SD 9.5) years in NHS and 43.4 (SD 7.7) in NHSII. The majority of women with SLE in both cohorts were of Caucasian descent. Over 95% of SLE cases had a positive ANA test, with a mean number of ACR SLE classification criteria of 4.8 (SD 0.9) among cases in NHS and 4.9 (SD 1.1) among cases in NHSII. The majority of SLE cases in both cohorts were diagnosed by a physician who was a member of the American College of Rheumatology. Results of multivariable-adjusted analyses are displayed in Table 3. There was an inverse association between cumulative average alcohol consumption and SLE risk among NHS participants who consumed ≥5 g/day of alcohol (HR 0.57 [95%CI 0.34–0.96], p for trend <0.01), compared to non-drinkers. A similar, yet non-significant, trend for reduced SLE risk was observed in the highest category of intake of NHSII participants (HR 0.66 [95%CI 0.37–1.16], p for trend 0.17). The meta-analysis revealed a 39% decrease in SLE risk among NHS/NHSII participants who consumed ≥5 g/day (HR 0.61 [95%CI 0.41–0.89], p for trend <0.01) compared to non-drinkers. We examined the effect of individual alcoholic beverage type and SLE risk in secondary analyses (Table 4). Participants in NHS/NHSII who consumed ≥2 servings/week of wine had a reduced SLE risk (HR 0.65, [95%CI 0.45–0.96], p for trend 0.03) vs. non-wine drinkers. A similar, but weaker, trend towards an inverse association with ≥2 servings/week of beer compared to no beer was observed, after controlling for other beverage types and covariates (HR 0.71 [95%CI 0.42–1.20], p for trend 0.19). No association was demonstrated for cumulative hard liquor consumption and SLE risk. In stratified models to assess for effect modification by OCP use and smoking, we found no significant interactions (results not shown). Similarly, interaction terms in our final multivariable models were not significant (alcohol intake 0 (ref,), >0 to<5, or ≥5 g/day x ever (ref.)/never OCP use [p 0.44 and p 0.22]; alcohol intake in the same three categories x ever (ref.)/never smoking [p 0.47 and p 0.93]).

Table 2.

Characteristics of participants at SLE diagnosis in Nurses’ Health Study and Nurses’ Health Study II

Characteristics at SLE diagnosis	NHS (N=125)	NHSII (N=119)
Age at diagnosis in years, mean (SD)	55.8 (9.5)	43.4 (7.7)
White race (%)	96.0	93.3
Anti-nuclear antibody positive (%)	96.0	99.2
Anti-double stranded DNA antibody positive (%)	46.4	57.1
Arthritis (%)	76.8	68.1
Hematologic involvement (%)	56.0	66.4
Renal involvement (%)	17.6	9.2
Number of ACR SLE criteria met, mean (SD)	4.8 (0.9)	4.9 (1.1)
Seen by an ACR member rheumatologist (%)	72.8	89.1

Table 3.

Association between daily alcohol intake and risk of incident SLE among participants in Nurses’ Health Study and Nurses’ Health Study II

	Category of Cumulative Average Alcohol Intake
	None	>0 to <5 g/d	≥5 g/d	p trendc
NHS
Cases/Person-Years	33/524,022	62/1,013,665	30/823,011
Age-Adjusted HR (95%CI)a	1.00 (ref)	1.09 (0.71–1.67)	0.61 (0.37–1.00)	0.011
Multivariable-Adjusted HR (95%CI)b	1.00 (ref)	1.05 (0.68–1.63)	0.57 (0.34–0.96)	0.007
NHSII
Cases/Person-Years	35/594,522	63/1,134,816	21/484,830
Age-Adjusted HR (95%CI)a	1.00 (ref)	0.95 (0.62–1.44)	0.77 (0.45–1.33)	0.339
Multivariable-Adjusted HR (95%CI)b	1.00 (ref)	0.86 (0.56–1.32)	0.66 (0.37–1.16)	0.167
NHS/NHSII (Meta-analyzed)
Cases/Person-Years	68/1,118,544	125/2,148,481	51/1,307,841
Age-Adjusted Pooled HR (95%CI) a	1.00 (ref)	1.01 (0.75–1.37)	0.68 (0.47–0.98) p het 0.541	0.008
Multivariable-Adjusted Pooled HR (95%CI)b	1.00 (ref)	0.95 (0.70–1.29)	0.61 (0.41–0.89) p het 0.568	0.002

^aAge adjusted models adjusted for age (months), questionnaire cycle, and total daily energy intake (kcal, continuous)

^bMultivariable models additionally adjusted for race, oral contraceptive use (ever, never), smoking (ever, never), median family income from US census (continuous).

^cp trend was computed by assessing the linear trend assigning the median value in each alcohol category

Abbreviations: HR = hazard ratio; CI = confidence interval, p het= p for heterogeneity between the cohorts

Table 4.

Association between type of alcohol intake and risk of incident SLE among participants in Nurses’ Health Study and Nurses’ Health Study II^a (Meta-Analyzed)

Alcohol Intake by Serving Frequency
	Alcohol Type	None	>1/month to <1/week	≥2/week	p trendc
Beer	Cases/Person-years	113/2,007,139	9/226,423	3/127,137
	Age-adjusted HR (95%CI)a	1.00 (ref)	0.97 (0.65–1.45)	0.76 (0.45–1.29) p het 0.827	0.302
	Multivariable-adjusted HR (95%CI)b	1.00 (ref)	0.94 (0.63–1.40)	0.71 (0.42–1.20) p het 0.900	0.191
Wine	Cases/Person-years	58/944,883	37/642,808	30/773,008
	Age-adjusted HR (95%CI)a	1.00 (ref)	0.86 (0.63–1.18)	0.69 (0.47–1.00) p het 0.924	0.056
	Multivariable-adjusted HR(95%CI)b	1.00 (ref)	0.84 (0.61–1.15)	0.65 (0.45–0.96) p het 0.918	0.033
Liquor	Cases/Person-years	74/1,382,374	32/473,718	19/504,607
	Age-adjusted HR (95%CI)a	1.00 (ref)	1.11 (0.70–1.74)	1.03 (0.67–1.58) p het 0.180	0.829
	Multivariable-adjusted HR (95%CI)b	1.00 (ref)	1.09 (0.67–1.77)	0.99 (0.64–1.52) p het 0.178	0.636

^aAge adjusted models adjusted for age (months), questionnaire cycle, total daily energy intake (kcal, continuous). Additionally, the different alcohol types were included in the same model and mutually adjusted for each other (with missings in reference category). Results from the two cohorts were meta-analyzed using a DerSimonian and Laird random effects model.

^bMultivariable models additionally adjusted for race, oral contraceptive use (ever, never), smoking (ever, never), median family income from US census (continuous).

^cp trend was computed by assessing the linear trend assigning the median value in each alcohol category

Abbreviations: HR = hazard ratio; CI = confidence interval, p het= p for heterogeneity between the cohorts. Bold typeface indicates significant results.

In additional multivariable-adjusted analyses, after excluding participants who ever consumed ≥30 grams/day of alcohol (11 cases of SLE and 164,419 person-years), a slightly stronger inverse association between cumulative average alcohol consumption and SLE risk was demonstrated (HR 0.54 [95%CI 0.36–0.81], p for trend <0.01), compared to our main analyses (HR of 0.61 [95%CI 0.41–0.89], p for trend <0.01). When we incorporated probable SLE cases, a total of 280 definite or probable SLE cases were identified; 151 cases in NHS and 129 cases in NHSII. A similar inverse association was demonstrated (HR 0.71 [95%CI 0.50–0.99], p for trend 0.01). In a sensitivity analysis allowing 4–6 years before SLE diagnosis, there was also a similar inverse association: HR 0.61 [95%CI 0.41–0.91], p for trend <0.01 (Table 5). In an analysis using simple updating of alcohol measures (to assess the instantaneous hazard), we found the same inverse association between alcohol intake and SLE risk (HR 0.68 [95%CI 0.48–0.96], p for trend 0.04). Baseline alcohol consumption also was inversely associated with SLE risk (HR 0.60 [95%CI 0.42–0.86], p for trend <0.01).

Table 5.

Association between daily alcohol intake and risk of incident SLE among participants in Nurses’ Health Study and Nurses’ Health Study II in lagged analyses (last alcohol measurement 4–6 years prior to SLE diagnosis)

	Categories of Cumulative Average Alcohol Intake
	None	>0 to <5 g/d	≥5 g/d	p trendc
NHS
Cases/Person-Years	31/489,690	52/932,144	25/762,138
Age-Adjusted HR (95%CI)a	1.00 (ref)	1.00 (0.63–1.57)	0.56 (0.33–0.96)	0.013
Multivariable Adjusted HR (95%CI)b	1.00 (ref)	0.98 (0.62–1.55)	0.54 (0.31–0.94)	0.010
NHSII
Cases/Person-Years	32/567,486	55/1,036,358	19/439,303
Age-Adjusted HR (95%CI)a	1.00 (ref)	0.94 (0.60–1.46)	0.80 (0.45–1.42)	0.460
Multivariable Adjusted HR (95%CI)b	1.00 (ref)	0.87 (0.56–1.36)	0.70 (0.39–1.28)	0.283
NHS/NHSII (Meta-analyzed)
Cases/Person-Years	63/1,057,176	107/1,968,502	44/1,201,441
Age-Adjusted HR (95%CI)a P het 0.503	1.00 (ref)	0.97 (0.71–1.33)	0.66 (0.45–0.98)	0.013
Multivariable Adjusted HR (95%CI)b P het 0.572	1.00(ref)	0.92 (0.67–1.27)	0.61 (0.41–0.91)	0.006

^aAge adjusted models adjusted for age (months), questionnaire cycle, and total daily energy intake (kcal, continuous).

^bMultivariable models additionally adjusted for race, oral contraceptive use (ever, never), smoking (ever, never), median family income from US census (continuous).

^cp trend was computed by assessing the linear trend assigning the median value in each alcohol category

Discussion

In these two large, prospective cohorts of women followed for many years prior to the onset of SLE, we found a robust inverse association between alcohol consumption and SLE risk. These analyses demonstrated a 39% reduction in SLE risk among women who consumed an average of ≥5 g/day of alcohol (approximately ½ drink per day). Intake of wine, the most frequently consumed alcoholic beverage among these women, was associated with significantly reduced SLE risk in both cohorts. In this study, we used cumulative average intake to assess long-term alcohol exposure over a time period as long as up to 32 years. To our knowledge, this is the largest and longest prospective study to date, including 244 incident SLE cases, evaluating SLE risk using repeated measures of alcohol consumption.

Our results are consistent with some, but not all, of the prior studies (24–26). For example, a strong protective effect of alcohol intake of >150 grams per month compared to no consumption (OR 0.2, 95%CI 0.1–0.5) was found in a Swedish study involving 85 SLE cases and matched controls(25). However, a past Japanese case-control study suggested that heavy alcohol consumption (>4–5 days per week) was associated with increased SLE risk (27). In an internet-based case-control study of 114 SLE cases and propensity-score matched controls, no association was demonstrated between drinking alcohol and SLE risk (28). Case-control studies, however, are prone to recall bias, as subjects are asked to recall their alcohol intake history and their responses may be influenced by existing disease. Reverse causation bias, in which disease development alters alcohol intake behavior, is also possible in case-control studies and is particularly concerning for SLE diagnosis, which may have an insidious onset. The only prospective cohort study to previously examine this association was in the Black Women’s Health Study, which found no association between alcohol intake and SLE risk (RR 1.0, 95%CI 0.4–2.4) (29). That study was limited by sample small size (34 confirmed SLE cases), and a single baseline alcohol assessment. A subsequent meta-analysis of six case-control studies and one cohort study, demonstrated a protective effect of moderate alcohol intake on SLE risk (OR 0.72, 95%CI 0.55–0.95) (30). However, that study was limited by varied definitions of both alcohol intake and SLE, and inclusion of patients with prevalent SLE treated for <10 years, in addition to the biases mentioned above.

An inverse relationship between alcohol intake and SLE risk is biologically plausible. Alcohol diminishes cellular responses to immunogens, and suppresses synthesis of pro-inflammatory cytokines, such as tumor necrosis factor (TNF), interleukin (IL)-6, IL-8, both in vivo and in vitro in alveolar macrophages and human blood monocytes(31). Increased daily alcohol consumption is associated with a decline in urinary neopterin, a macrophage activation marker and indicator of SLE disease activity(32). Additionally, antioxidants such as resveratrol or humulones in wine and beer influence cytokines such as interferon-gamma in vitro and may inhibit key enzymes involved in DNA synthesis (33, 34). Moderate alcohol intake may also reduce serum levels of IgG (7). Finally, alcohol consumption may induce epigenetic changes, resulting in altered gene expression that could affect immune homeostasis(35).

A study from Japan has reported an interaction between alcohol intake and N-acetyltransferase 2 (NAT2) genotype (NAT2 rapid acetylators have a lower risk of SLE than do NAT2 slow acetylators) in SLE risk(36). This study suggests that the presence of a genotype associated with rapid acetylation (vs. slow) and ever consuming alcohol (vs. never), reduced SLE risk to an OR of 0.08 (95%CI 0.03–0.28) (p interaction 0.026). Although to our knowledge, no prior studies have been done to demonstrate whether NAT2, an important xenobiotic-metabolizing enzyme, directly metabolizes ethanol, this interaction does suggest that alcohol intake and polymorphisms in hepatic metabolizing enzymes may help determine individual susceptibility to SLE.

J- or U-shaped associations between chronic moderate alcohol consumption and risk of several outcomes, including rheumatoid arthritis (RA), cardiovascular outcomes (such as congestive heart failure and stroke), diabetes, and all-cause mortality (9, 37, 38), have been observed. Our group has shown a U-shaped association between alcohol intake and plasma biomarkers of inflammation, including IL-6 and soluble TNFR2 (a proxy for TNF) among women with pre-clinical RA (19). Moderate alcohol intake of 5.0–9.9 g/day was associated with the lowest concentrations of these biomarkers. Furthermore, we have demonstrated that NHS and NHSII participants who consumed 5.0–9.9 g/day of alcohol (<1 drink per day) had a reduced RA risk compared to women with no alcohol intake(8). In the current study, there were few ‘heavy’ drinkers, limiting our ability to directly investigate the effects of higher levels of alcohol consumption on SLE risk. However, in analyses in which we excluded women who consumed ≥30 grams of alcohol per day (<5% of the cohort), the inverse association with SLE risk was slightly stronger, suggesting a similar U-shape curve.

We found no evidence that the association between alcohol and SLE risk was modified by other SLE risk factors, including smoking or oral contraceptive use. Given the insidious onset of SLE, along with the possibility that pre-clinical disease may affect patient behavior, results from the lagged analysis were useful in ruling out reverse causation bias. We demonstrated an association similar to the primary analysis when using definite and probable SLE cases, and when studying moderate short-term alcohol consumption (shown in the simple updated mode) or baseline alcohol consumption. We investigated the possibility of residual confounding related to smoking and SES, both of which are highly associated with alcohol intake. Our results did not significantly change when we assessed smoking as current/past/never smokers or in pack-years, or when we replaced median family income with another SES covariate, husband’s education level. Alcohol consumption among women in NHSII has remained relatively consistent over time, where 54% reported the same category of total alcohol intake and 91% reported intake within one category higher or lower in 2003 compared to 1991(39). Therefore, given similar alcohol categories utilized in our study, results from our sensitivity analyses likely reflect the stability of drinking patterns among these women, such that similar results were seen in our long-term, short-term, baseline, and lagged alcohol analyses.

An important strength of our study is the use of data from cohorts with up to 2.4 million person-years of prospective follow-up. The use of cumulative average alcohol intake to assess long-term exposure along with detailed data on potential time-varying confounders reduced the within-subject variation, minimized inaccuracy of exposure data, and decreased the potential for reverse causation and recall biases. Detailed alcohol data collected over time on FFQs allowed for the calculation of alcohol in grams per day, enhancing the precision of our exposure. We were also able to evaluate an association with types of alcoholic beverages and SLE risk. Repeated exposure measurements allowed us to conduct several sensitivity analyses to assess the robustness of the main results.

Our study has some limitations. Given our strict SLE definition, we may have excluded possible SLE cases that may later become definite cases. As we are investigating SLE etiology, our aim however, was to study validated SLE although this may have resulted in a limited number of incident SLE cases. Furthermore, among the 5% of nurses lost to follow-up, it is possible but unlikely that any participants subsequently died from previously unreported SLE as deaths and causes of death are confirmed through correspondence with family members and review of medical records.

Additionally, some misclassification of alcohol may be introduced using self-reported questionnaires. However, validation studies have indicated a high correlation of alcohol intake (r=0.9) on the questionnaire with the dietary records(14). Notably, our study was not powered to investigate the association of different types of wine (e.g. white wine versus red wine) on SLE risk. Finally, given that the NHS cohorts include mostly White US women who were healthy and working in advanced nursing professions at the inception of the study, there is the potential for lack of generalizability to younger women, males, non-Whites, and populations with lower SES. Additionally, although NHS cohort questionnaires assess alcohol intake during follow-up in detail, they do not capture all alcohol intake prior to cohort entry.

In conclusion, we found a significant inverse association between moderate long-term alcohol consumption and SLE risk. However, the potential benefits of moderate alcohol consumption among premenopausal and postmenopausal women have to be weighed against the other possible health risks, with cancer risk (especially breast cancer) being a major concern even for light to moderate alcohol consumption (40). Our findings have implications for SLE prevention in that identifying risk factors, and those associated with decreased risk, enables risk factor modification and provides insight into disease pathogenesis.

Significance and Innovation.

• Alcohol suppresses systemic inflammation and has been associated with reduced risks of both cardiovascular disease and rheumatoid arthritis. However, past epidemiologic studies of alcohol consumption and SLE risk have demonstrated conflicting results.

• This prospective cohort study with detailed measures of alcohol intake and other lifestyle exposure data is the largest and longest investigation of the association of alcohol consumption and risk of SLE.

• We report an inverse association between long-term moderate alcohol consumption (≥5 grams or 0.5 drinks/day) and SLE risk (HR 0.61 [95%CI 0.41–0.89]).

• The identification of modifiable factors, such as alcohol intake, associated with SLE risk has implications for SLE prevention and provides insight into disease pathogenesis.