Association of Depression With Risk of Incident Systemic Lupus Erythematosus in Women Assessed Across 2 Decades

Key Points

Question  Is depression associated with an increased incidence of systemic lupus erythematosus, a paradigmatic, systemic autoimmune disease?

Findings  In this 20-year cohort study of 194 483 women, those with a history of depression had a more than 2-fold subsequent increased risk of systemic lupus erythematosus compared with women with no depression. Adjustment for health behaviors (eg, body mass index, cigarette smoking, and oral contraceptive or postmenopausal hormone use) slightly attenuated the associations.

Meaning  Depression may be associated with increased risk of systemic lupus erythematosus, and the association is not fully explained by the measured health behaviors.

Importance  It has long been hypothesized that depression may increase the risk of developing autoimmune disease; however, rigorous empirical evidence is sparse.

Objective  To evaluate whether an association exists between depression and risk of incident systemic lupus erythematosus (SLE), a paradigmatic, systemic autoimmune disease.

Design, Setting, and Participants  This 20-year prospective, longitudinal cohort study evaluated data collected from 2 cohorts of women participating in the Nurses’ Health Study (1996-2012) and the Nurses’ Health Study II (1993-2013). Data analyses were conducted from August 2017 to May 2018.

Main Outcomes and Measures  Incident SLE with 4 or more American College of Rheumatology criteria was ascertained by self-report and confirmed by medical record review. Depression was assessed repeatedly throughout follow-up according to whether women reported having received a clinician’s diagnosis of depression, regular antidepressant use, or a score of less than 60 on the 5-item Mental Health Inventory (MHI-5). Whether longitudinally assessed health risk factors (eg, cigarette smoking, body mass index, oral contraceptive use, menopause or postmenopausal hormone use, alcohol use, exercise, or diet) accounted for increased SLE risk among women with vs without depression was examined. Cox proportional hazards regression models were used to estimate risk of SLE. In addition, the association of depression lagged by 4 years, and depression status at baseline with incident SLE throughout follow-up was assessed.

Results  Data from 194 483 women (28-93 years of age; 93% white) were included. During 20 years of follow-up, 145 cases of SLE occurred. Compared with women with no depression, women with a history of depression had a subsequent increased risk of SLE (HR, 2.67; 95% CI, 1.91-3.75; P < .001). Adjustment for body mass index, cigarette smoking, and oral contraception and postmenopausal hormone use slightly attenuated associations (adjusted HR, 2.45; 95% CI, 1.74-3.45; P < .001). The SLE risk was elevated with each of the 3 following depression indicators modeled separately: clinician’s diagnosis of depression (HR, 2.19; 95% CI, 1.29-3.71), antidepressant use (HR, 2.80; 95% CI, 1.94-4.05), and MHI-5 scores indicating depressed mood (HR, 1.70; 95% CI, 1.18-2.44). Associations remained strong when depression status was lagged by 4 years with respect to the outcome (HR, 1.99; 95% CI, 1.32-3.00) and when depression status at baseline was used as the exposure (HR, 2.28; 95% CI, 1.54-3.37).

Conclusions and Relevance  This study contributes to increasing evidence that depression may be associated with increased risk of SLE and suggests that the association is not fully explained by measured health factors or behaviors.

Several lines of evidence suggest that depression may be a risk factor for autoimmune disease. First, many cross-sectional studies have found a high prevalence of depression among persons with autoimmune diseases.^1-6 Second, depression has been associated with greater disease severity among persons with autoimmune disease.^7-9 Finally, increased immune activation associated with depression is a plausible biological pathway that could trigger autoimmune disease.^10-16

However, only a few studies have examined the temporal associations between depression and subsequent development of autoimmune disease. Longitudinal studies have found depression to be associated with subsequent onset of any autoimmune disease^17,18 and specifically rheumatoid arthritis,¹⁹ psoriasis,²⁰ or Crohn disease.²¹ Here, we examine the association of depression with systemic lupus erythematosus (SLE), a paradigmatic, systemic autoimmune disorder. Systemic lupus erythematosus is a serious disease characterized by widespread organ system involvement and is associated with early death.²² Investigators have posited that depression is a risk factor for SLE on the basis of the high prevalence of depression among existing patients with SLE.^4-6 The largest cross-sectional study found that women with SLE reported a 1.7-fold higher prevalence of depression than women in the general population.²³

Quiz Ref IDTo establish whether depression increases risk of SLE, it is crucial to consider the following alternative possible causal structures: (1) depression could be an early symptom of SLE;^24,25 (2) prediagnostic SLE symptoms could lead to depression; and (3) SLE could be initially misdiagnosed as depression.²⁶ If depression indeed contributes to SLE risk, then it is important to consider plausible mechanisms. Depression leads to increased chronic inflammation as marked by higher levels of inflammatory biomarkers, which is partly explained by health factors, such as higher body mass index (BMI) and cigarette smoking.^10,15,16,27 Thus, a higher prevalence of health risk factors among persons with depression may account for an increased risk of developing SLE. To our knowledge, no studies to date have examined the role of health behaviors in the potential association between depression and incident SLE risk. Therefore, we examined whether an association exists between risk of incident SLE and time-updated depression status, and we investigated the potential role of health risk factors using data from 2 large, longitudinal studies of women.

The Nurses’ Health Study (NHS) is a cohort of 121 700 US registered nurses established in 1976.²⁸ The Nurses’ Health Study II (NHS II) is a cohort of 116 430 US registered nurses from 25 to 42 years of age at enrollment in 1989. Both cohorts are followed up biennially, with SLE follow-up through 2012 (NHS) and 2013 (NHS II). We included women who responded to the first questionnaire on which depression was queried (100 920 women in the baseline 1996 NHS; 107 499 women in the 1993 NHS II). After exclusions, 194 483 women (28-93 years of age; 93% white) with complete data were included over 20 years of follow-up. Given the far higher incidence of SLE among women vs men, a study of women is warranted.²⁹ The institutional review board of Brigham and Women’s Hospital (Boston, Massachusetts) approved the study. Return of the questionnaire by the respondent via US mail constituted implied informed consent.

Quiz Ref IDDepression was assessed via the following 3 indicators from responses on biennial questionnaires: (1) self-reported clinician diagnosis, (2) self-reported use of antidepressants, and (3) a 5-item depressive symptom questionnaire. Women were asked whether they had recently had clinician-diagnosed depression and the year of diagnosis (NHS, biennially 2000-2012; NHS II, biennially 2003-2011). Women were also asked whether in the past 2 years they had regularly used an antidepressant, with a variety of antidepressants listed (eg, Prozac and Zoloft). Antidepressant use was queried biennially in the NHS beginning in 1996 and in NHS II beginning in 1993. Depressive symptoms were assessed with the 5-item Mental Health Inventory (MHI-5) of the 36-item Short-Form Health Survey (for NHS in 1992, 1996, and 2000; for NHS II in 1993, 1997, and 2001).³⁰ The MHI-5 has good predictive value in identifying depression, with an area under the curve of 0.88 to 0.91 for detection of mood disorders or major depression.^30,31 Several cutoff values for the MHI-5 score have been suggested to indicate probable depression.^31-34 Here, we followed work assessing healthy populations and dichotomized the MHI-5 score at 60, with scores less than 60 indicating probable depression.³³ We conducted a sensitivity analysis using a more stringent score of less than 52 to indicate depression. We coded “history of depression” as a time-updated covariate and considered women who reported antidepressant use, had a depression diagnosis, or had an MHI-5 score indicating probable depression as having a history of depression at the first year in which they reported one of these indicators and for all subsequent years.

Systemic lupus erythematosus case identification according to the American College of Rheumatology 1997 classification criteria has been reported.^35-37 Two rheumatologists independently reviewed the medical records of women who reported SLE to confirm whether 4 of these 11 criteria were met. We excluded 10 853 women who reported SLE or other connective tissue disease before cohort enrollment. We censored 6660 women who reported SLE not confirmed by medical record review or other connective tissue disease during follow-up. Dates of SLE diagnosis and first SLE symptoms were ascertained from medical records.

We considered factors that have been associated with SLE as potential covariates.^38-41 We examined the following time-updated health factors, all assessed via biennial questionnaire: oral contraceptive use (coded as never, past, or current); BMI, calculated as weight in kilograms (assessed biennially by self-report and previously validated in the NHS) divided by height in meters squared (from height assessed at cohort inception);⁴² cigarette smoking (coded as never, past, or current); alcohol use (coded as none, >0 and <5 g/d, or ≥5 g/d); exercise (coded as <10 or ≥10 metabolic equivalents per day); diet assessed every 4 years with the Alternative Healthy Eating Index and coded in quintiles; and postmenopausal hormone use (coded as any or none). Because these factors could both be a consequence of depression^43-45 and affect risk of subsequent depression^46,47 and because many of these behaviors are patterned in early adulthood before cohort enrollment, we conceptualized them as either potential mediators or confounders and did not conduct formal mediation analyses.

We considered the following factors as potential confounders because they would likely not be sequelae of depression: race (white or nonwhite), early menarche (<10 years of age),^48,49 US region of residence, and census tract median income.⁵⁰ Respondents indicated their major ancestry from among 7 choices; race was considered because it has been associated with the incidence of SLE.²⁹ Of these potential confounders, only race changed our risk estimates by more than 2%; therefore, we included only race in subsequent models.

First, we calculated the prevalence of demographic covariates and health risk factors by history of depression in 1996 (NHS) or 1997 (NHS II), the first questionnaire cycles in which all 3 measures of depression had been queried. Quiz Ref IDTo ascertain whether women with vs without a history of depression were at higher risk of incident SLE, we calculated hazard ratios (HRs) for SLE using time-updated depression status as the independent variable in Cox proportional hazards regression models, with age in months as the time measure, separately for each cohort and adjusted for race. Depression and SLE were chronologically sequenced such that depression status was used to estimate SLE status reported in the subsequent biennial questionnaire (eg, depression in 2002 was used to estimate SLE in 2004). The date of SLE diagnosis from the medical records was used to date the onset of SLE. Women were censored at death, loss to follow-up, end of follow-up, or report of connective tissue disease not confirmed as SLE. We conducted tests of heterogeneity, and the results indicated similar associations in the 2 cohorts (P = .81); therefore, we pooled data for all analyses.

We then examined the extent to which increased prevalence of health risk factors among women with a history of depression might account for a possible association between depression and SLE. To construct a parsimonious model, we examined the effect of adding each risk factor individually, in separate models, on the association of depression with incident SLE. We then calculated HRs for SLE in a Cox proportional hazards regression model, with depression as the independent variable and all risk factors that altered the β coefficient of the association of depression with SLE by more than 2% as covariates.⁵¹ Because time-varying covariates may lie on the pathway from depression to SLE, we also estimated a model adjusted for covariates at baseline only without time-updating them.

To reduce the likelihood that associations occurred because SLE caused depression or because depression was an early symptom of SLE rather than because depression increased risk of SLE, we conducted the following 3 additional analyses: (1) we lagged depression status by an additional 4 years; thus, only women with depression at least 4 years prior to SLE were considered to have a history of depression at SLE incidence (eg, depression status as of 1995 would be the exposure for SLE incidence between 1999 and 2001); (2) we examined depression status characterized only for the first year in which all 3 depression measures had been queried (depressive symptoms, depression diagnosis, and antidepressant use; 1996 for NHS and 1997 for NHS II) in association with SLE throughout follow-up without time-updating depression status; and (3) we conducted analyses in a subsample of 120 women with data on the date of first SLE symptoms using the date of the first symptoms rather than the date of diagnosis to indicate SLE onset. In addition, we separately examined the associations of each of our 3 indicators of depression with SLE incidence. To ascertain whether the use of antidepressant medication might be driving an association of depression with incident SLE, we calculated HRs for SLE, with depression status as the independent variable, censoring women at first report of antidepressant use. In addition, we examined the association of the use of specific antidepressants with SLE incidence in 2 separate models: (1) selective serotonin reuptake inhibitors (SSRIs) and (2) tricyclic and other antidepressants. Finally, we examined associations of depression with incidence of double-stranded DNA (dsDNA)–positive SLE and dsDNA-negative SLE. Double-stranded DNA–positive SLE is a subtype of SLE characterized by more active systemic inflammation; some environmental exposures have been associated with only dsDNA-positive SLE.^52,53All data analyses were conducted from August 2017 to May 2018 using SAS, version 9.4 (SAS Institute Inc). A 2-sided P < .05 was considered statistically significant.

Compared with women with no depression history, women with a history of depression were more likely to be current cigarette smokers (14.4% vs 10.5%) and to have used oral contraceptives (72.5% vs 69.2%) and postmenopausal hormones (38.7% vs 33.3%) but were less likely to exercise regularly (43.1% vs 48.2%) (Table 1). Nearly all of the women receiving a diagnosis of SLE during follow-up tested positive for antinuclear antibodies (99.3%); most of these women had hematologic involvement (67.6%), and 51.7% tested positive for dsDNA at diagnosis (eTable 1 in the Supplement).

Quiz Ref IDDepression was associated with a more than 2-fold increased risk of incident SLE in a proportional hazards regression model adjusted for race (HR, 2.67; 95% CI, 1.91-3.75) (Table 2). The median time from depression to incident SLE was 4.5 years (interquartile range, 2-8 years). In models examining individual health risk factors (ie, BMI, cigarette smoking, and oral contraceptive and postmenopausal hormone use), each factor altered the association of depression with SLE by 2% or more. Diet, exercise, or alcohol use did not alter the association of depression with SLE; therefore, we did not include these covariates in subsequent models (results including all covariates are given in eTable 2 in the Supplement). Depression remained strongly associated with risk of incident SLE in models further adjusted for BMI, cigarette smoking, and oral contraceptive and postmenopausal hormone use whether at baseline only (HR, 2.53; 95% CI, 1.80-3.57) or using time-updated covariates (HR, 2.45; 95% CI, 1.74-3.45) (Table 2).

In sensitivity analyses, depression status lagged 4 years (HR, 1.99; 95% CI, 1.32-3.00) and depression status in 1996 or 1997 (HR, 2.28; 95% CI, 1.54-3.37) were both associated with increased risk of developing SLE. In the subsample of women with a date of first SLE symptoms, the date of first symptoms preceded SLE diagnosis by a mean of 1.0 years (range, 0-15 years). In a model using the date of first symptoms to indicate SLE onset, the association of depression with risk of SLE symptom onset was nearly identical to risk estimates in the main analysis (HR, 2.60; 95% CI, 1.79-3.77) (Table 2). In models separately examining each indicator of depression, history of depressive symptoms, depression diagnosis, and antidepressant use were each associated with higher risk of incident SLE. Of these, the magnitude of the association with increased risk of SLE was strongest for antidepressant use (HR, 2.80; 95% CI, 1.94-4.05) and weakest for depressive symptoms (HR, 1.70; 95% CI, 1.18-2.44). In analyses using the MHI-5 with a more stringent cutoff score of less than 52 to indicate significant depression, associations were stronger than among those found using less than 60 as the cutoff score (HR, 2.17; 95% CI, 1.37-3.46) (Table 3).

In analyses censoring women at first report of antidepressant use, a clinician’s diagnosis of depression or high depressive symptoms on the MHI-5 remained strongly associated with increased risk of developing SLE (HR, 2.11; 95% CI, 1.32-3.38). In addition, the associations of SSRIs and tricyclic and other antidepressants with risk of incident SLE were similar (SSRIs: HR, 2.96; 95% CI, 1.88-4.64; tricyclic antidepressants: HR, 3.19; 95% CI, 1.72-5.93). The results did not substantially differ in models including only anti-dsDNA antibody–positive SLE or anti-dsDNA antibody–negative SLE (75 dsDNA-positive cases: HR, 2.58; 95% CI, 1.66-3.99; 70 dsDNA-negative cases: HR, 2.50; 95% CI, 1.54-4.06) (Table 3).

In this long-term prospective study, a history of depression was associated with a risk 2.5 times as great as no history of depression in developing SLE. This association was retained and of similar magnitude when depression was lagged by 4 years, when depression status at study baseline only was considered, and when SLE onset was dated from symptom onset. Accounting for the higher prevalence of health risk factors among women with depression, including higher BMI, cigarette smoking, and oral contraceptive and postmenopausal hormone use, only slightly attenuated associations between depression and incident SLE. Of our 3 indicators of depression, depressive symptoms (MHI-5) were the least strongly associated with SLE. This result may have occurred because depressive symptoms were assessed only 3 times during follow-up compared with 5 (NHS II) or 7 times (NHS) for depression diagnosis and 9 (NHS) or 10 times (NHS II) for antidepressant use. Therefore, the measure of depressive symptoms is more likely than the other depression indicators to have misclassified women with depression as having no depression, potentially attenuating the effect estimate toward no association.⁵⁴ Alternatively, these results may indicate a dose-response association with depression if women with depressive symptoms as defined when based on the use of a screening tool are not as severely depressed as those who receive a depression diagnosis or antidepressant medication; research on depression and the risk of other diseases, such as coronary heart disease, has found similar patterns.⁵⁵ This alternative hypothesis is supported by the stronger associations in analyses using a more stringent cutoff for the MHI-5. It is likely that, for some women, antidepressants were prescribed for anxiety disorders or chronic pain, conditions frequently associated with SLE,^56-58 which could explain why antidepressant use was associated most strongly with SLE. However, we found an association of depression with SLE risk in analyses censoring women at first antidepressant use, suggesting that antidepressant use was not a primary factor driving associations between depression and SLE. Moreover, associations between antidepressant use and SLE were not specific to antidepressant type, further suggesting that the antidepressants themselves were not causing the risk of SLE.

Although our results are consistent with several underlying causal relationships, our findings provide support for the hypothesis that depression is a causal risk factor for developing SLE, perhaps via altered immune function. In prospective studies, depression has been associated with increases in concentrations of biomarkers of systemic inflammation, including interleukin 6 and C-reactive protein.^10-16 Compared with healthy individuals, people with depression have higher concentrations of indicators of cell-mediated immune system activation (such as neopterin), proinflammatory cytokines (such as tumor necrosis factor), and coagulation factors (such as fibrinogen, plasminogen activator inhibitor type 1, and tissue-type plasminogen activator antigen).^16,27,59 In SLE, immune complexes activate the complement cascade and cause inflammatory cells to attack tissue-specific cells, resulting in organ damage (eg, lupus nephritis). Individuals susceptible to developing SLE have increased blood concentrations of inflammatory cytokines and chemokines; further elevations could fuel autoimmunity in pre-SLE states.⁶⁰

A common cause of depression and SLE could account for some of the association observed in the present study between depression and SLE, although this explanation is somewhat unlikely because we considered the possible effects of numerous potential confounders, including age at menarche, race, and socioeconomic status. Moreover, further adjustment for a wide range of health risk factors plausibly associated with depression and SLE did not greatly attenuate associations. In addition, depression may be an early symptom of SLE because nervous system involvement (including mood disorders, such as depression, as well as headache, seizure, and stroke) is common in SLE, occurring in 50% to 90% of patients.^24,25 In addition, pain, fatigue, and other symptoms of SLE may lead to depression. In related research, the experience of pain partly explained the higher prevalence of depression in persons with rheumatoid arthritis vs those without.⁶¹ Symptoms of SLE, including fatigue, headache, confusion, memory loss, and pain, may also mirror symptoms of depression,⁶² and early SLE can be initially misdiagnosed as depression. However, in our analyses using the date of first symptoms to indicate SLE onset, the results were nearly identical. Therefore, depression would need to precede other early SLE symptoms to account for our findings.

Although some earlier studies included SLE as one of many autoimmune diseases examined in association with depression,¹⁸ to our knowledge, only a single study has separately examined risk of SLE onset.¹⁷ In a prospective study using Danish medical registries, treatment of depression was associated with a significantly increased risk of any subsequent autoimmune disease diagnosis and with SLE specifically (359 total cases; incident rate ratio for SLE, 1.63; 95% CI, 1.26-2.11).¹⁷ However, that study did not provide data about the timing of depression with respect to SLE; thus, it is possible that depression was temporally proximate to SLE diagnosis and may, in fact, have been an early symptom of SLE. An additional study found that US military veterans with psychiatric diagnoses were at elevated risk of a new diagnosis of SLE throughout follow-up compared with veterans with no psychiatric diagnoses.⁶³ However, that study did not examine risk of SLE in association specifically with depression. Moreover, the median time between first psychiatric diagnosis and autoimmune diagnosis was 220 days, leaving open the possibility that depressive and other psychiatric symptoms resulted from nervous system involvement in undiagnosed SLE rather than psychiatric illness increasing risk of subsequent development of SLE.

Quiz Ref IDOur study has important limitations. Nurses may be healthier than the general population, which may limit generalizability. It is likely that some women had depression prior to cohort enrollment and were misclassified as having no history of depression, which may have biased our results. Our data lacked adequate detail to examine subtypes of depression or depression remission in association with risk of SLE.

Our study contributes to increasing evidence that depression may constitute a causal risk factor for autoimmune disease. Screening of patients with depression for family history and symptoms of SLE may improve its early detection. In addition, lifestyle interventions to reduce inflammation in persons with depression may reduce risk of autoimmune disease as well as other negative health sequelae of depression, including cardiovascular disease.^64-67

Accepted for Publication: July 6, 2018.

Correspondence to: Andrea L. Roberts, PhD, Department of Environmental Health, Harvard T. H. Chan School of Public Health, 677 Huntington Ave, Boston, MA 02115 (aroberts@hsph.harvard.edu).

Published Online: September 12, 2018. doi:10.1001/jamapsychiatry.2018.2462

Author Contributions: Ms Malspeis had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Roberts, Kubzansky, Feldman, Costenbader.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Roberts, Costenbader.

Critical revision of the manuscript for important intellectual content: Kubzansky, Malspeis, Feldman, Costenbader.

Statistical analysis: Roberts, Malspeis, Feldman, Costenbader.

Obtained funding: Costenbader.

Administrative, technical, or material support: Costenbader.

Supervision: Roberts, Kubzansky, Costenbader.

Conflict of Interest Disclosures: None reported.

Funding/Support: This work was supported by a Midcareer Investigator Award in Patient-Oriented Research (K24 AR066109) (Dr Costenbader), a Research Project Grant (RO1 AR057327) (Dr Costenbader), and a Mentored Patient-Oriented Research Career Development Award (K23 AR071500) (Dr Feldman), all from the National Institute of Arthritis and Musculoskeletal and Skin Diseases.

Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Information: Walter Willet, MD, DrPH, received a Research Project With Complex Structure Cooperative Agreement award (UM1CA176726-05) from the National Cancer Institute in support of the Nurses’ Health Studies.