Sex hormones, body mass index, and related comorbidities associated with developing Sjögren's disease: a nested case-control study.

A case–control study was conducted, nested within the Marshfield Clinic Health System’s Virtual Data Warehouse (MC-VDW) population. This source population included adult females who were MC-VDW members for ≥ 1 year between 01/01/1990 and 12/31/2019. The MC-VDW is a research resource used as part of the Health Care Systems Research Network. It is composed of members of Security Health Plan (SHP) of Wisconsin (health insurance affiliated with MC) and residents of the Marshfield Epidemiologic Study Area (MESA). MESA includes 24 postal codes within the umbrella of MCs’ core service areas in north-central Wisconsin. The MC-VDW population tracks person-time on members of SHP and MESA residents. Over 90% of inpatient and outpatient medical care is captured among MESA residents [ 6 ], and virtually all claims-based diagnoses, labs, anthropometric values, and pharmacy prescriptions are captured among SHP members (e.g., 99% prescription capture) [ 7 ]. As described further below, cases included all patients in the source population with first known primary SjD codes during the study timeframe. All procedures were approved in advance by the MC Institutional Review Board (IRB-20–699) with a waiver of individual informed consent and HIPAA authorization due to the retrospective nature of this medical record study design. To be included as a confirmed incident primary SjD case, patients met at least one of the following three definitions [ 8 , 9 ]: ≥ 2 SjD International Classification of Diseases diagnostic codes in 710.2 or M35.0 × made by a rheumatology provider and separated by at least 28 days apart. ≥ 1 SjD diagnostic codes in 710.2 or M35.0 × made by a rheumatology provider and a positive anti-SS-related antigen A (SS-A) antibody test or lip biopsy indicative of inflammatory/lymphocytic infiltrate (per confirmation via manual chart review). ≥ 2 SjD diagnostic codes in 710.2 or M35.0 × made by a provider outside of rheumatology at least 8 days apart, as well as a positive SS-A antibody test or lip biopsy indicating inflammatory/lymphocytic infiltrate (per confirmation via manual chart review). To avoid overlap SjD cases, we excluded patients who met any case definition above, but with a positive cyclic citrullinated peptide antibody test or with a diagnostic code indicative of other systemic autoimmune disorders, including rheumatoid arthritis (A714.0, 714.1, 714.2, 714.30, 714.31, 714.32, 714.33, 714.81, M05.XX-M06.XX) [ 10 ], systemic lupus erythematosus (710.0, M32), myositis (710.3, 710.4, M33.03, M33.13, M33.2, M33.9, M33.93), psoriatic arthritis (696.0, L40.52), ankylosing spondylitis (720.0, M45.0), or systemic sclerosis (710.1, M34) after the index date of enrollment in the MC-VDW. For each SjD case, three controls were randomly selected from the source population of SjD-free patients and matched on 2-year age increments and a clinical encounter within 2 years of the index case’s SjD date. Several sociodemographic and clinical covariates were collected from electronic health record (EHR) data, including age, race/ethnicity, health insurance, tobacco use, and alcohol disorders. We generated a composite score to estimate lifetime endogenous and exogenous hormone exposure. Our modified composite estrogen score (mCES) consisted of five variables, each assigned one point if present: (1) any hormone replacement therapy used for at least 90 days; (2) more than three pregnancies; (3) menopause by 55 years or older; (4) obesity; (5) hysterectomy. This mCES was informed by previous composite scores and modified to include factors that could be abstracted from the EHR. Previous composite scores used estrogen replacement therapy [ 11 - 14 ], parity [ 11 , 14 - 16 ], late menopause [ 12 - 14 ], BMI [ 16 , 17 ], and hysterectomy [ 12 , 13 ]. Our mCES score included events associated with unopposed estrogen like obesity. Raised BMI has been established as correlating with endogenous estrogen levels after menopause [ 18 ] and is thus considered a marker of endogenous estrogen exposure. Similar to previously developed composite scores, we also included hysterectomy in our mCES, because this procedure is most commonly indicated for fibroids, benign tumors dependent on estrogen for growth [ 19 ]. The type, dose, and duration of exogenous estrogen and progesterone exposure were tested. Timing of exposure was documented as early (< 40 years old) vs. late (≥ 40 years old) and exposure duration was documented at the SjD date of diagnosis. Oral contraceptives were tested as a separate variable to evaluate if they have an effect on SjD risk. Estrogen dose within each oral contraceptive was documented. Other sex hormone–related variables collected included menarche, fibroids, menstrual irregularity, anovulation, menorrhagia, polycystic ovarian syndrome (PCOS), hirsutism, ovarian cysts, amenorrhea, endometriosis, hysterectomy (with or without oophorectomy), bilateral oophorectomy, breast cancer, endometrial cancer, any cancer, anorexia/bulimia, aromatase, or selective estrogen receptor modulators (SERM) use. Because PCOS, hirsutism, and ovarian cysts are representative of the same underlying phenomenon, we combined these variables into an umbrella “PCOS” [ 20 ]. Health status and medical comorbidities are tied to menstrual cycling [ 21 ], so we also collected data on related comorbidities. The comorbidities, including components of the Charlson Comorbidity Index [ 22 ], were myocardial infarction, congestive heart disease, peripheral vascular disorder (PVD), cerebrovascular disease, chronic pulmonary disease, diabetes, diabetes with chronic complications, renal disease, malignancy, lymphoma, and pelvic fracture. We also included fibromyalgia given the prominence of this comorbidity among SjD patients. Exposure variable definitions are included in Supplemental Table A . Diagnosis or medication exposures that were missing were considered absent (for example, we assumed a missing endometriosis diagnosis code signified a subject did not have endometriosis). Analytical procedures were conducted using R for statistical computing version 4.0. Univariable comparisons utilized t -tests, ANOVAs, or chi-square tests when appropriate. Relative risk ratios (RRs) were estimated from Poisson regression models. Multivariable models included age, Caucasian status, ethnicity, and insurance type as covariates. Subjects with missing data for a variable necessary in a particular analysis were removed from that analysis. The development of the predictive model followed the methods laid out in Austin and Tu [ 23 ]. In brief, two-thirds of the cases and their corresponding controls were randomly selected for inclusion in a derivation dataset. The remaining one-third was then used in a validation dataset. Each candidate variable (see Supplemental Table B ) was tested for univariable association with case–control status via t -test or chi-square test. Those that had univariable p -values < 0.25 were considered possible predictor variables. A bootstrap resampling backwards logistic model selection with p < 0.05 as the inclusion threshold was conducted 1000 times and the frequency of each possible predictor in the 1000 models was counted. Once the final list of predictor variables was set, a logistic regression model with those predictors was fit to the validation dataset with case–control status as the primary predictor. Goodness of fit was examined using the Hosmer–Lemeshow goodness-of-fit statistic [ 24 ], while predictive ability was assessed by the area under the curve (AUC) of the ROC curve for the model-predicted probability of SjD. This is also known as Harrell’s c -index [ 25 ].

The final analytical sample included 546 SjD cases and 1637 age- and sex-matched SjD-free controls. As outlined in Table 1 , cases and controls had a similar sociodemographic profile. Significantly fewer SjD cases were obese than controls (33% vs. 44%; < 0.001). At the end of our study period in 2019, there were 275 females meeting criteria for SjD among 73,616 current, living female members of the population-based source population. Accordingly, the estimated point prevalence of primary SjD in this population was 37 per 10,000 adult females (0.4%). Our primary exposure of interest was the mCES. Given few patients with mCES of 4 ( n = 8), we combined mCES levels 3 and 4. Patients with mCES of ≥ 3 were older and had different medical insurance (e.g., more Medicare; Table 2 ). Other covariates, including race, ethnicity, tobacco use, and alcohol use, were not significantly different between mCES levels. BMI was highest in mCES ≥ 3 with 85% of these subjects categorized as obese, compared to mCES = 0 with 0%, mCES = 1 with 55%, and mCES = 2 with 63% ( Table 2 ), which was an expected finding because BMI ≥ 30 was a variable included in the mCES. The mCES was not significantly associated with SjD ( Table 3 ). As outlined in Table 4 , our secondary analysis examined sex hormone exposure individually. Several hormone exposure events were associated with greater RR for SjD including the following: any estrogen-only systemic hormone replacement therapy (HRT) (1.78 [1.47–2.14]), PCOS (including ovarian cysts and hirsutism) (1.65 [1.28–2.12]), hysterectomy without bilateral oophorectomy (1.51 [1.13–2.03]), menopause (RR 1.46 [95% CI 1.20–1.77]), fibroids (1.44 [1.19–1.73]), oral contraceptive use (1.40 [1.09–1.79]), hysterectomy (with or without bilateral oophorectomy) (1.37 [1.07–1.74]), menorrhagia (1.36 [1.09–1.68]), any combined HRT ever (1.36 [1.00–1.86]), menstrual irregularity (1.34 [1.07–1.67]), endometriosis (1.26 [1.06–1.52]), and estrogen dose within oral contraceptives (1.01 [1.00–1.02]). In addition, two exposures were associated with lower RRs for SjD including continuous BMI (0.98 [0.97–0.99]) and age at menopause (0.98 [0.97–1.00]). Non-hormone comorbidities with a significantly greater RR for SjD included the following: fibromyalgia (2.09 [1.75–2.50]), lymphoma (1.99 [1.06–3.74]), osteoporosis (1.92 [1.50–2.46]), PVD (1.79 [1.13–2.84]), renal disease (1.73 [1.13–2.64]), chronic pulmonary disease (1.40 [1.09–1.80]), and COPD (1.26 [1.04–1.54]) ( Table 5 ). Diabetes was associated with significantly lower RR for SjD (0.48 [0.31–0.76]). The final stepwise multivariable risk ratio model included five factors: (1) fibromyalgia (odds ratio [OR] 2.50 [95% CI 1.93–3.25]), (2) diabetes (0.27 [0.13–0.50]), (3) osteoporosis (1.84 [1.27–2.66]), (4) BMI (0.97 [0.95–0.99]), and (5) HRT ever (1.61 [1.22–2.12]) ( Table 6 ). When the final model was applied to the validation dataset, the AUC was 0.67 (0.63–0.72; Supplemental Figure A ), indicating moderate performance for classifying SjD. The goodness of fit from the Hosmer–Lemeshow test was p = 0.956 ( Supplemental Table B ), which indicates a very strong fit between the predicted and true data.

To our knowledge, this was the largest nested case–control study to date on the association between sex hormone exposures and SjD. Key findings of our study included novel individual sex hormone and comorbidities significantly associated with SjD that we leveraged to develop a preliminary prediction model. Specifically, we describe new sex hormone exposures protecting against SjD risk, such as BMI, and exposures increasing SjD risk, such as HRT and PCOS. Additionally, we described new comorbidities preceding and potentially predicting increased SjD risk. These exposures may advance our insights into the sexual dimorphism of SjD and reveal novel pathogeneses to inform development of targeted therapies. Our composite measure of sex hormone exposure, the mCES, did not predict incident SjD. The SjD predictors identified by our optimized model included fibromyalgia, diabetes, osteoporosis, BMI, and estrogen-only HRT use, providing moderate predictive value for SjD (AUC of 0.67). To our knowledge, this is the first reported prediction model for SjD built from pre-existing comorbidities and hormone exposures. Notably, this model achieved moderate predictive ability for SjD based on ICD and medication codes alone—all EHR extractable variables. One could envision that adding symptoms or signs of dryness or complications of dryness could further enhance the predictive value of our model. We report a number of endogenous hormone exposure variables that predicted SjD development. First, menstrual irregularity was more common in SjD, which has previously been reported in population-based studies [ 26 ]. PCOS, a common cause of menstrual irregularity, was similarly increased among SjD compared to matched control patients [ 27 ]. These findings would support that an aberrant estrogen to androgen ratio precedes SjD. Menorrhagia and endometriosis have previously been reported as more frequent in SjD patients than controls [ 5 ]. Interestingly, endometriosis correlated with anti-SSA and anti-SSB antibodies, and SjD patients required surgical intervention for endometriosis more often than controls [ 5 ]. Previous publications theorize that endometriosis might represent an autoimmune disease because (i) there is associated tissue damage; (ii) autoantibodies target endometrium, ovary, phospholipids, and histones; (iii) there is decreased cell apoptosis and abnormalities in T- and B-cells; and (iv) endometriosis is associated with other autoimmune diseases [ 28 , 29 ]. We also identified an earlier age of menopause and higher frequency of post-menopausal status among SjD patients, which persisted after multivariable analysis. This suggests protection against SjD with greater menses duration. Notably, this contrasts with previous studies that were smaller, not population-based, and did not find a difference in menopause status between groups [ 2 , 30 ]. We also report gynecologic surgical interventions associated with SjD. We stratified hysterectomies with or without bilateral oophorectomy assuming patients who undergo bilateral oophorectomy will use estrogen replacement, whereas those with remaining ovaries rely on endogenous estrogen production. We found that SjD risk was 50% greater after hysterectomy without bilateral oophorectomy, indicating HRT after bilateral oophorectomy might provide some protection. Our results contrast with a prior study evaluating hysterectomy among SjD than sicca-controls that reported fewer hysterectomies in SjD than sicca-controls versus our findings with healthy controls [ 2 ]. Referral bias due to dryness after hysterectomy might contribute discrepant findings [ 2 ]. Herein, we found ERT and HRT were greater in SjD than controls. The independently protective associations of obesity and diabetes were somewhat surprising. Prior studies on these risk factors are mixed. For example, high BMI preceded the onset of other autoimmune diseases such as RA in one study [ 31 ], but a small population-based study observed a non-significant trend toward lower BMI in patients with SjD [ 32 ]. Interestingly, in one study, SjD subjects with the lowest total body water had higher BMI, longer disease duration, and the highest ocular symptoms; BMI might increase as SjD progresses in some [ 33 ]. A large epidemiologic study of adults in Sweden found a significant association between concurrent SjD and type 2 diabetes, but no significantly increased risk of incident SjD over time in adults following their type 2 diabetes diagnosis [ 34 ]. That study used ICD codes alone, so their SjD diagnosis used old ICD definitions that included sicca symptoms, a common diabetes comorbidity. It is possible that the lower BMI occurs before diagnosis and this pro-inflammatory state or symptoms lead to reduced BMI. A unifying inflammatory pathway associated with both diabetes and BMI, including changes in the adaptive immune response, might provide a common protective mechanism [ 35 ]. Related therapies may also be insightful, as diabetes treatments such as metformin have been proposed as a possible strategy to treat SjD via its activation of 5′-adenosine monophosphate–activated protein kinase (AMPK) and resulting inhibition of mammalian target of rapamycin (mTOR) [ 36 ]. Murine studies show improved salivary gland inflammation and function with metformin [ 37 ]. Although COPD, PVD, renal disease, lymphoma, and osteoporosis are known to be increased in SjD, we report that they preceded SjD diagnosis and may help serve as predictors for SjD. In a small study of 41 subjects with SjD, 37% had COPD [ 38 ]. Posited theories for increased COPD in SjD are that (i) SjD patients had occult disease, including pulmonary-parenchymal changes, before diagnosis, which our data support, or (ii) SjD pathogenesis exacerbates adverse effects of smoking history. Likewise, SjD is associated with increased arterial stiffness and subclinical atherosclerosis [ 39 ], and we demonstrated increased PVD predating SjD compared to controls. Previous studies have shown increased frequencies of renal disease [ 40 ], lymphoma [ 41 ], and osteoporosis [ 42 ] in SjD compared to controls, but have not shown these morbid conditions are associated with prospective risk of developing SjD. Because SjD diagnosis is typically delayed by years, it is likely that subacute inflammation drives morbidity prior to a formal SjD diagnosis [ 43 ]. Previous population-based cohorts have only described fibromyalgia as a risk factor for development of SjD with adjusted hazard ratios of twofold healthy controls [ 44 ]. Although there were significant strengths to this study including the defined source population, large sample size, and rigorous case definition that was specific to primary SjD, we acknowledge limitations. Limitations included the retrospective design that precludes causal conclusions and the sole focus on female risk factors for SjD. In addition, the source population is mostly White and rural, which limits generalizability to other races and regions. At the index date, patients averaged 55 years old; thus, it is uncertain if the absence of pre-menopausal exposures, like oral contraceptive use, might have represented no use versus missing data. In contrast, because the cohort had decades of EHR data since 1991, including periods when HRT was common, we collected fairly rich HRT data. Although we excluded 205 subjects with missing data from our multivariable analyses, the univariable and multivariable results were very similar, suggesting minimal impact of missing data on study conclusions. Given the inherent and, for some patients, relatively long diagnostic delay of SjD, reverse causation could also be a source of bias if a given exposure occurred in close proximity to the clinical recognition of SjD. However, this could be informative in itself, as early indicators of disease for improved SjD identification using EHR data. Risks of measurement or disease misclassification bias were likely minimal given the use of EHR data and chart-verified indicators of SjD under a rigorous case definition. Selection bias, which can be particularly problematic in the control group, was also minimized by selecting both cases and controls from the same, defined source population. We attempted to control for potential confounders through statistical adjustment for race, ethnicity, and socioeconomic status, as well as all other exposures in our final multivariable model, though unmeasured confounding risks remain. In addition, the matching procedure helped balance age and recent medical visits between cases and controls. We excluded other autoimmune diseases from our SjD cases, so we cannot extrapolate our findings to patients with overlap SjD. We describe multiple novel SjD comorbidities, but did not examine whether multimorbidities coexist in the same subject; future studies should compare populations with the described multimorbidities among subjects with and without SjD and explore other relevant comorbidities, such as small fiber neuropathy [ 45 ]. In summary, we report a novel algorithm that, if confirmed in future prospective studies, can improve prediction of SjD. We describe sex hormone exposures that suggest estrogen contributes to SjD pathogenesis. Finally, we report new comorbidities that might be associated with decreased risk of SjD including diabetes and high BMI, beyond those associated with increased risk like lymphoma and osteoporosis, among other factors.

Sjögren’s disease (SjD) is a highly female-predominant systemic autoimmune disease. Though often underdiagnosed, estimates of SjD prevalence vary widely depending on region and case definitions, ranging from 0.01% to more than 3% of the adult population [ 1 ]. The variable prevalence of SjD reflects diagnostic challenges, such as the nonspecific nature of dry eyes and mouth, delaying the time-to-diagnosis of SjD, and contributing to reduced well-being. SjD is associated with immutable risk factors (e.g., genetics, female sex, and older age) and also modifiable exposures (e.g., sex hormones and infections), but distinct causal pathways are poorly understood. This is a major barrier to SjD prevention and management, as even modest gains in understanding modifiable factors that predict SjD development could help postpone disease onset, or at least minimize SjD diagnostic delays. SjD is a female-predominant autoimmune disease. The extreme sexual dimorphism of SjD is thought to be related to female sex hormone exposure [ 2 ] and X chromosome–dependent gene expression [ 3 ]. Previous studies evaluating endogenous or exogenous hormone exposure suggested a link between sex hormones and SjD development, but were conducted using small samples or lacked healthy controls [ 2 , 4 , 5 ]. Using a case–control sample from a defined population cohort, the objectives of this study were to identify factors associated with SjD risk, focusing on endogenous and exogenous sex hormone exposures to inform risk prediction modeling.