Systemic Lupus Erythematosus
Combined oral contraceptive use and the risk of systemic lupus erythematosus
To assess whether the risk of incident systemic lupus erythematosus (SLE) is associated with the use of combined oral contraceptives (COCs), because studies of the link between exogenous hormonal exposure and the risk of SLE have produced conflicting results.
We conducted a population-based nested case–control study among women ages 18–45 years, using the UK's General Practice Research Database. All incident cases of SLE from 1994–2004 (n = 786) were identified in the database and matched with up to 10 controls (n = 7,817) among women without SLE at the time of the case's diagnosis.
The adjusted rate ratio (RR) of incident SLE associated with any use of COC was 1.19 (95% confidence interval [95% CI] 0.98–1.45), whereas with current use it was 1.54 (95% CI 1.15–2.07). The rate was particularly increased in current users who had only recently started COC use (RR 2.52, 95% CI 1.14–5.57) compared with longer-term current users (RR 1.45, 95% CI 1.06–1.99). The risk appeared to be particularly elevated with current exposure to first- or second-generation contraceptives (RR 1.65, 95% CI 1.20–2.26), and increasing with the dose of ethinyl estradiol (RR 1.42, 1.63, and 2.92 for ≤30 μg, 31–49 μg, and 50 μg, respectively).
The use of COCs is associated with an increased risk of SLE. This risk is particularly elevated in women who recently started contraceptive use, suggesting an acute effect in a small subgroup of susceptible women.
Systemic lupus erythematosus (SLE) is a systemic autoimmune disease. The ratio of women to men with SLE is 9 to 1, and the incidence increases after puberty. Thus, endogenous hormonal factors are believed to play an important part in the etiology of SLE (1, 2). Estrogen contained in contraceptives has been also questioned. In the UK, ∼25% of women under the age of 50 years use a contraceptive pill as a method of contraception (3, 4). First, second, and third generations of combined oral contraceptives (COCs) are widely used today, with doses of estrogen varying from 15–50 μg.
However, the role of exogenous estrogen as a trigger of SLE remains controversial because studies of the link between exogenous hormonal exposure and SLE risk have produced conflicting results. Several case reports described a temporal association between the start of oral contraceptives (OCs) containing estrogen and the onset of SLE, especially in predisposed individuals (5). Four case–control studies failed to find such a risk (6–9). On the other hand, a slight increase in the risk of SLE with ever OC use was found in the Nurses' Health Study (10, 11).
In view of these conflicting data, we conducted a large population-based observational study to assess whether COC use increases the risk of incident SLE, with particular consideration to the timing of COC use.
SUBJECTS AND METHODS
Subjects and source of data.
The data were obtained from the UK population-based General Practice Research Database (GPRD), which has been described in detail elsewhere (12, 13). More than 6 million people in the UK are enrolled with over 400 general practitioners who use office computers and have agreed to provide data for research purposes. General practitioners have been trained to record medical information, including demographic data, medical diagnoses, details of hospital stays, and death, using a standard anonymous form. The physicians generate prescriptions directly with their study computer; this information is automatically transcribed into the computer record. A modification of the Oxford Medical Information System (OXMIS) classification (similar to the International Classification of Diseases, Eighth Revision) is used to enter medical diagnoses, and a coded drug dictionary based on the UK Prescription Pricing Authority Dictionary is used for recording prescriptions. The recorded information on drug exposure and diagnoses has been validated and proven to be of high quality (14–16). The study was approved by the Scientific and Ethical Advisory Group of the GPRD.
We used the GPRD to form a population-based cohort of all women 18–45 years of age during the period January 1, 1994 to December 31, 2004 who had at least 2 years of continuous recording in the database. Cohort entry was defined as January 1, 1994, the patient's 18th birthday, or the date of joining the practice, whichever was latest. We restricted our study to women ≤45 years of age because the date of menopause is not systematically recorded in the GPRD. We conducted a case–control analysis within this cohort with up to 10 matched controls per case.
We identified all women in the study cohort with a first recorded diagnosis of SLE after cohort entry, using the GPRD medical codes that have been used and validated in previous studies (17, 18). As validated in the study by Somers et al, OXMIS and Read codes for SLE were identified from a coding dictionary (18). Patients with ≥1 SLE code in their medical record were identified. The date corresponding to the first SLE record represented the date of diagnosis. In order to confirm the incident nature of SLE, in contrast with prevalent cases who were experiencing an exacerbation, all women necessarily had ≥2 years of data available, during which no SLE diagnosis was made, prior to this first diagnosis. All cases with premature artificial menopause, defined as a medical code of bilateral oophorectomy or a pelvic radium treatment any time prior to the SLE diagnosis, were excluded.
Each case was matched with up to 10 controls selected from the study cohort. The controls were matched on age (±1 year) at the date of the case's diagnosis and came from the same general practice as the case in order to control for possible physician-related and geographic variations. To control for trends over time in the use of drugs, they were also matched on the calendar year of study cohort entry. Controls had to be at risk for the event, i.e., active in the practice and free of SLE on the date of SLE diagnosis of their matched case. Thus, cases and their matched controls had equal followup, which provides equal delay of disease occurrence. The date of the case's SLE diagnosis was assigned to the matched controls and called the index date. All controls with premature artificial menopause any time prior to the index date were also excluded.
COC use was defined by prescriptions for estrogen and progestative hormones in the same contraceptive. We classified contraceptives into first- and second-generation contraceptives containing ethinyl estradiol combined with the progestatives norethisterone, levonorgestrel, and norgestrel. Third-generation contraceptives contained ethinyl estradiol and either gestodene, desogestrel, or norgestimate. All such prescriptions given prior to the index date were identified from the computerized medical records.
Conditional logistic regression for matched case–control data was used to estimate the adjusted odds ratios (ORs) of incident SLE and their 95% confidence intervals (95% CIs) associated with the ever, current, and past use of COCs. The OR is an accurate estimate of the rate ratio (RR), particularly with a rare event such as SLE. Current use of COCs was defined as ≥1 prescription of COC in the 3 months prior to the index date, and past use was defined as the last COC prescription having been written >3 months prior to the index date. Non-use of COCs at all times during the 3 years prior to the index date was used as the reference.
To assess the effect of the duration of current use, we defined new, short-term use as starting COC use in the 3 months prior to the index date, i.e., current use with no other prescriptions in the 3 years prior to the index date. All other current use, i.e., current use with COC prescriptions preceding the 3 months prior to the index date, was called long-term current use. Additionally, an analysis was performed to compare the effects of the newer third-generation COCs with first- or second-generation COCs. Finally, for the first- or second-generation COCs, an analysis of the ethinyl estradiol dose was performed, comparing COCs containing ≤30 μg of ethinyl estradiol with those containing 30–50 μg and those containing exactly 50 μg.
In addition to the inherent adjustment by the matched factors, we adjusted for smoking and drinking status, as well as for comorbidity. Variables related to comorbidities, present any time before the index date, were considered as dichotomous covariates (yes/no). All comorbidity variables were entered in the model for the adjustment procedure. The common autoimmune diseases considered were type 1 diabetes mellitus and thyroiditis. Among vascular diseases, hypertension, stroke, and venous thrombosis were considered. Smoking or drinking status was classified as ever, never, or missing value before the index date.
The cohort consisted of 1,723,781 women ages 18–45 years. During followup, 786 of the women presented a first-time medical diagnosis of SLE, corresponding to an incidence rate of 5.7 cases per 100,000 per year. The cases were matched to 7,817 controls from the cohort. Characteristics of the cases and controls are reported in Table 1. The women were ∼33 years old at the index date, with 81%younger than 40 years, and had an average of 8 years of followup. Other autoimmune diseases and vascular diseases were more prevalent in the cases than in the controls. Thrombotic events occurring within the 3 months prior to the index date were present in 20 cases and 4 controls, all non-users of COCs. Smoking and drinking habits were more frequent in cases than in controls.
Table 1. Comparison of baseline characteristics of cases and controls (women ages 18–45 years)*
|Age, mean ± SD years†||33.1 ± 7.3||33.1 ± 7.3|
|Followup, mean ± SD years†||8.0 ± 3.7||8.0 ± 3.7|
|Lifestyle habits|| || |
| Smoking|| || |
| Never||276 (35.1)||2,710 (34.6)|
| Ever||275 (35.0)||1,742 (22.3)|
| Missing||235 (29.9)||3,365 (43.1)|
| Regular alcohol use|| || |
| Never||105 (13.4)||743 (9.5)|
| Ever||382 (48.6)||3,174 (40.6)|
| Missing||299 (38.0)||3,900 (49.9)|
|Autoimmune disease prior to the index date|| || |
|Type 1 diabetes||13 (1.7)||51 (0.7)|
|Thyroiditis||47 (6.0)||107 (1.4)|
|Vascular disease prior to the index date|| || |
|Hypertension||29 (3.7)||105 (1.4)|
|Stroke||7 (0.9)||1 (0.0)|
|Venous thrombosis||32 (4.1)||79 (1.0)|
The adjusted RR of SLE associated with any use of COCs relative to non-use was 1.19 (95% CI 0.98–1.45) (Table 2). However, whereas current use of COCs was associated with a higher rate of SLE (RR 1.54, 95% CI 1.15–2.07), this was not observed with past use (RR 1.06, 95% CI 0.85–1.33). Moreover, the rate of SLE was higher with newly started, short-term current use of COCs (RR 2.52, 95% CI 1.14–5.57) as compared with longer-term current use (RR 1.45, 95% CI 1.06–1.99).
Table 2. Crude and adjusted RRs of incident SLE associated with use of COCs in the 3-year period prior to the index date, relative to never use*
|Non-use (reference)||600 (76.3)||6,395 (81.8)||1.00||1.00 (reference)|
|Any use||186 (23.7)||1,422 (18.2)||1.43||1.19 (0.98–1.45)|
|Current use||65 (8.3)||385 (4.9)||1.85||1.54 (1.15–2.07)|
| Short-term||8 (1.0)||31 (0.4)||2.79||2.52 (1.14–5.57)|
| Long-term||57 (7.3)||354 (4.5)||1.77||1.45 (1.06–1.99)|
|Past use||121 (15.4)||1,037 (13.3)||1.28||1.06 (0.85–1.33)|
Use of first- or second-generation contraceptives was associated with an increased rate of SLE (RR 1.65, 95% CI 1.20–2.26), but use of third-generation contraceptives was not (RR 1.12, 95% CI 0.57–2.19) (Table 3). Additionally, the rate of SLE appeared higher with the highest (50 μg) dose of ethinyl estradiol (RR 2.92, 95% CI 1.45–5.89) compared with the lower doses (31–49 μg RR 1.63, 95% CI 0.90–2.95; ≤30 μg RR 1.42, 95% CI 0.94–2.15). Analyses restricted to women with ≥3 years of information in the database before the index date (721 cases and 7,169 controls) yielded similar results. Furthermore, when we performed the sensitivity analysis using the cases defined by ≥2 separate clinical diagnoses of SLE (422 cases and 4,197 controls), the adjusted RR of incident SLE associated with any use of COCs was 1.29 (95% CI 0.98–1.67), and with current use it was 1.78 (95% CI 1.20–2.64), but there were too few short-term users to provide stable estimates.
Table 3. Crude and adjusted RRs of incident SLE associated with type and dose of ethinyl estradiol content of COC use, relative to never use*
|Non-use (reference)||600 (76.3)||6,395 (81.8)||1.00||1.00 (reference)|
|Past use||121 (15.4)||1,037 (13.3)||1.28||1.06 (0.85–1.33)|
|Current use by COC type|| || || || |
| Third generation||10 (1.3)||83 (1.1)||1.34||1.12 (0.57–2.19)|
| First or second generation||55 (7.0)||302 (3.9)||2.00||1.65 (1.20–2.26)|
|Current use by COC dose of ethinyl estradiol|| || || || |
| Low (≤30 μg)||30 (3.8)||189 (2.4)||1.73||1.42 (0.94–2.15)|
| Medium (31–49 μg)||14 (1.8)||78 (1.0)||1.96||1.63 (0.90–2.95)|
| High (50 μg)||11 (1.4)||35 (0.5)||3.38||2.92 (1.45–5.89)|
In this study, we found that the use of COCs was associated with a significant increased risk of incident SLE. This risk appeared to be mostly limited to the first 3 months of use with first- and second-generation contraceptives containing higher doses of ethinyl estradiol, suggesting an acute effect in susceptible women and possibly a dose-response effect of estrogen on SLE onset.
Our finding of a significant increased risk of incident SLE with current COC use, mostly in patients who had started use recently, suggests that estrogen may have an acute effect in susceptible women. Estrogen is able to directly modulate the immune response (2, 19–21). This modulation could complete the action of some sex-linked genes contributing to the genetic predisposition for the disease. Indeed, the estrogen receptor α codon 594 genotype was shown to influence the development of SLE (19). Furthermore, estrogen plays an important role in B cell maturation, selection, and activation, and has been shown to have an effect on the breakdown of immune tolerance seen in SLE (2, 22).
Previous studies designed to assess the risk of SLE following exposure to OCs have provided conflicting results. OC use was found to be unrelated to SLE in 4 case–control studies (6–9). In addition, 2 clinical trials showed that COCs did not increase the risk of disease flares in SLE patients (23, 24). On the other hand, the Nurses' Health Study reported an OR of SLE of 1.4 (95% CI 0.9–2.1) with ever OC use (10). A recent extended analysis of this cohort, including over 200,000 women followed between 1976 and 2003, reported a significant OR of 1.5 (95% CI 1.1–2.1), with the risk highest among women with <2 years of OC use, but no effect of contraceptive type and a nonsignificant increased risk with current use (11).
The discrepant results of these studies may be explained by differences in study design, patient selection, and assessment of exposure of interest. The case–control studies had a limited number of cases, ranging from 85–240; a retrospective assessment of the exposure by interview subject to recall bias; and incomplete assessments of the formulations of OCs used (6–9). However, we do not believe that the discrepant results can be explained by different doses and formulations of estrogen and different study periods (old and recent), because there was considerable overlap in these respects in several of the studies.
Our results are consistent with those of the Nurses' Health Study, both in terms of the presence and magnitude of an association. An additional strength of our study was that we found a possible dose-response effect of estrogen on SLE onset within a recent period of exposure, supporting the causality of the effect. However, given the relatively homogenous cohort in the Nurses' Health Study, consisting of professional, mostly white women above the age of 25 years, it has been argued that those results may not be generalizable to other women with SLE, in particular the young black women with low socioeconomic status who represent a large group at risk of particularly severe disease (25). Even though racial and socioeconomic status are not available in our study, considering the population-based recruitment in the GPRD database, we studied a representative sample of the UK population, with a multiethnic representation. Further studies on specific ethnic groups will have to be performed.
An alternative explanation of our results could be that COCs are associated with an increased risk of thrombotic events, and this risk is increased in patients with genetic and acquired thrombophilia (26). This risk tends to be highest in the first year of use. Because thrombophilia is a feature of SLE, new COC use may indirectly predispose susceptible patients to thrombotic events, which become the first recognized symptom of SLE. However, we did not observe any thrombotic event between the start of COCs and the date of SLE diagnosis in users that had recently started COCs.
Another explanation of our results could be that unrecognized symptoms of SLE may induce women to take COCs, which is rather unlikely because COCs are prescribed mainly for contraception in premenopausal women. Other indications of COCs in premenopausal women could be irregular menstrual cycles, endometriosis, dysfunctional uterine bleeding, or menstrual disorders. No clear association between endometriosis or other menstrual conditions and SLE is known (6, 8), and no strong association between irregular menstrual cycles and SLE has been reported (11).
Our findings that longer-term use of contraceptives is associated with an increased risk of incident SLE (albeit of lower magnitude) and that current use of contraceptives with higher doses of ethinyl estradiol is associated with an increased risk of incident SLE, suggest a possible dose-response effect of estrogen on SLE onset, which could be an alternative or additional mechanism to favor occurrence of the disease. It could be supported by biologic data collected from both human and animal studies, which have provided evidence that alteration in sex hormone levels can alter tolerance of autoreactive B cells and exacerbate disease (27). The increase in risk observed with first- and second-generation contraceptives was associated with a possible dose-response effect of estrogen on SLE onset. The absence of significant increase in risk associated with the third-generation contraceptives could be related to lower doses of estrogens comparatively with first and second generations. Hormonal contraceptives containing only progestatives were not studied because there was no concern in any previous studies about a possible association with SLE occurrence (6–11).
Our study was strengthened by its use of incident SLE cases. Identification of cases of SLE was done through the medical codes for SLE in the GPRD database. We did not use the criteria the American College of Rheumatology (28) because the general practitioners who are collecting data are unlikely to record every symptom of a chronic disease such as SLE. Two previous studies have specifically used the GPRD database to identify SLE cases in the cohort of the patients. The close comparability of the incidence of SLE in the UK using SLE diagnostic codes in the GPRD (17, 18) and the incidence found by active surveillance studies in the 1990s (one based on the Metropolitan Districts of Birmingham and Solihull, UK and the other based on the Greater Nottingham metropolitan community [29, 30], described above) supports the use of GPRD medical codes for identification of SLE.
To ensure the incident nature of the cases, we chose a 2-year diagnostic-free period before the diagnosis because the mean interval between first symptoms and effective diagnosis is ∼2 years (31). The sensitivity analysis with ≥3 years of followup before the incident diagnosis found similar results. The population-based nature of the study protects against selection bias, and the large number of SLE cases, compared with previous studies, allowed several specific risk analyses such as short-term use, dose-response, and contraceptives type.
The use of a nested case–control study in a database such as GPRD, i.e., a cohort study, has several strengths (12–16). Matching on calendar time and practice allows control for time-dependent, physician-related, and geographic variations in the outcome and exposure. The design also minimizes the bias due to inappropriate selection of controls, and the use of prospectively recorded computerized exposure records prevents recall bias. Other types of differential misclassification of exposure history are also unlikely because exposure information was gathered prospectively before the first diagnosis code for SLE. The dates of COC prescription and of SLE diagnosis based on validated database records allowed estimation of COC use effects.
The incomplete information on gynecologic history could be a potential confounding factor and a limitation of our study. However, there are discrepant results concerning the potential influence of reproductive life events. Indeed, one study found that early age at menarche (≤10 years of age) is a risk factor for SLE occurrence (RR 2.1, 95% CI 1.4–3.2) (11), whereas others reported no relationship (8, 9) or a positive relationship between late menarche (>15 years of age) and SLE onset (OR 3.82, 95% CI 1.66–8.81) (32). Additionally, breastfeeding was found to be associated with a decreased risk of SLE (OR 0.6, 95% CI 0.4–0.9) in one study (8), but not in another (11). We restricted the age of inclusion to ≤45 years to consider only premenopausal women, thereby excluding the potential effect of hormonal changes related to postmenopausal status.
We observed an increase of other associated autoimmune diseases and cardiovascular disorders in case patients. Both could be associated with the underlying inflammatory and autoimmune susceptibility without being specifically associated with SLE diagnosis criteria, as reported previously (33–37). Adjustment for these factors slightly decreased the RR as compared with the crude RR, but the results remained significant.
Smoking and drinking habits were more frequent in cases than in controls, although missing data were more frequent in controls than in cases. In order to take into account the differential percentage of missing data between cases and controls, the variables “missing status of drinking” and “missing status of smoking” are included in the analysis for adjustment. Collection of information concerning smoking or drinking habits is not systematic in the GPRD database and could not predict the quality of all the medical data collection. The data are collected by general practitioners during clinical visits and then entered in the database. Information concerning smoking status and drinking status are known to be partially reported. Moreover, as these factors were significantly more frequent in the cases, the adjustment has reduced the size of the effect, thereby attenuating the association between exposure and the disease. Therefore, any result found by this study could be considered a lower-bound estimate of the effect. Finally, smoking has been reported to be a risk factor for SLE (38), whereas a protective effect of alcohol drinking on SLE risk has been reported (9, 39), but the hypothesis for the mechanisms of action is still debated.
Although COC use may be associated with a significant increased risk of incident SLE, some have argued that the low relative risk of ∼2 is probably insufficient to explain the 9:1 sex ratio in this disease (25). Thus, although our findings add considerably to the existing literature, further studies on the acute effects of COCs will be needed to better identify the characteristics of women susceptible to developing SLE when exposed to COCs.
Dr. Suissa had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study design. Bernier, Mikaeloff, Hudson, Suissa.
Acquisition of data. Suissa.
Analysis and interpretation of data. Bernier, Mikaeloff, Hudson, Suissa.
Manuscript preparation. Bernier, Mikaeloff, Hudson, Suissa.
Statistical analysis. Bernier, Suissa.
We would like to thank Mrs. Diane Gaudreau for her participation in extracting the data from the GPRD database and A. Kezouh for his assistance in data analysis.