Objective Prior studies found conflicting results about whether lupus is likely to flare during or after pregnancy. Using a large cohort of pregnant and non-pregnant women with lupus, we estimated the effect of pregnancy on disease flares in systemic lupus erythematosus.
Methods Data were collected in the Hopkins Lupus Cohort 1987–2015. Women aged 14–45 years with >1 measurement of disease activity were included. The time-varying exposures were classified as pregnancy, postpartum or non-pregnant/non-postpartum periods. Flares were defined as: (1) change in Physician Global Assessment (PGA)≥1 from previous visit and (2) change in Safety of Estrogens in Lupus National Assessment-Systemic Lupus Erythematosus Disease Activity Index (SELENA-SLEDAI)≥4 from previous visit. A stratified Cox model estimated HRs with bootstrap 95% CIs.
Results There were 1349 patients, including 398 pregnancies in 304 patients. There was an increased rate of flare defined by PGA during pregnancy (HR: 1.59; 95% CI 1.27 to 1.96); however, this effect was modified by hydroxychloroquine (HCQ) use, with the HR of flares in pregnancy compared with non-pregnant/non-postpartum periods estimated to be 1.83 (95% CI 1.34 to 2.45) for patients with no HCQ use and 1.26 (95% CI 0.88 to 1.69) for patients with HCQ use. The risk of flare was similarly elevated among non-HCQ users in the 3 months postpartum, but not for women taking HCQ after delivery.
Conclusions Our study supports and extends previous findings that the incidence of flare is increased during pregnancy and within the 3 months postpartum. Continuing HCQ, however, appeared to mitigate the risk of flare during and after pregnancy.
- systemic lupus erythematosus
- disease activity
Statistics from Altmetric.com
Systemic lupus erythematosus (SLE) is characterised by fluctuations of disease activity, with periods of high disease activity (ie, flares) followed by periods of low activity. The effect of pregnancy on disease activity in SLE has long been debated. Previous research has found that 19%–68% of women with SLE experience a flare during pregnancy.1–11 Risk factors for flares during pregnancy include active disease at conception, prednisone use, kidney disease and previous flares.2 5 7
There are conflicting results about the effect pregnancy has on the health of SLE women. Some studies report an increased rate of flares during pregnancy, while others report no difference in disease activity.8 9 12 13 A study by Lockshin et al 14analysed flare characteristics of pregnant and non-pregnant patients with SLE and did not find a difference between women who were and were not pregnant. In contrast, Petri et al 8 found the rate of flare was greater during pregnancy than in non-pregnant controls, and a subsequent analysis by Ruiz-Irastorza et al 9 found the flare rates during pregnancy and 6 weeks postpartum were increased compared with non-pregnant, age-matched controls. However, as these studies of flares during pregnancy were published over 20 years ago, an updated analysis is warranted.
A limitation of the current literature is the inconsistency in which flares were defined, making it difficult to make comparisons across studies. Many previous studies were also limited by small sample size, which reduced power to determine differences in the rate of flares between pregnant and non-pregnant patients. Understanding the effect pregnancy has on disease activity is clinically significant for the patient, as high disease activity during pregnancy is associated with preterm births and pregnancy loss.4 15 16 Additionally, examining the rate of flares during the postpartum period is important in determining if patients need to be more closely monitored in the months following pregnancy. The objective of the current analysis was to estimate the effect of pregnancy on disease flares in SLE.
The Hopkins Lupus Pregnancy Cohort is a subset of the Hopkins Lupus Cohort, which has prospectively followed patients with SLE since 1987, with data available through February 2015. Patients meeting the American College of Rheumatology (ACR) or Systemic Lupus International Collaborating Clinics (SLICC) criteria for SLE17–19 were eligible for enrolment in the cohort following informed consent. Patients enrolled in the Hopkins Lupus Cohort and patients with SLE seen in the Hopkins Obstetrics Clinics were referred to the Hopkins Lupus Pregnancy Cohort. Other patients were referred by their rheumatologists, the Maryland Lupus Foundation and self-referral.8 Pregnant women were seen every 4–6 weeks during pregnancy by a single rheumatologist. During each visit, lupus disease activity (Physician Global Assessment of disease activity (PGA)20 and Safety of Estrogens in Lupus National Assessment-Systemic Lupus Erythematosus Disease Activity Index (SELENA-SLEDAI)21–23 was measured, medications were updated, and laboratory tests were conducted. Pregnancy outcome data were collected from patients at the first postpartum visit or by telephone or email, if a woman did not continue care at the Lupus Center.
Exposure was classified as pregnancy (yes/no), postpartum period (yes/no) or non-pregnant/non-postpartum period (unexposed). The exposure variables were included as time-varying covariates, so as to include all observations for an individual (including prepregnancy observations on women who became pregnant). The postpartum period was analysed separately as lasting 3 months and 12 months.
Disease flares during follow-up were classified by PGA and SELENA-SLEDAI:
Change in PGA≥1 from the previous visit.
Change in SELENA-SLEDAI≥4 from the previous visit.
During the study period, there were 2417 patients with SLE observed in the Hopkins Lupus Cohort, of which 2229 were female. Fifteen patients were removed due to lack of complete information on pregnancies, and an additional 350 patients were removed because SLE diagnosis occurred after age 45. In order to calculate flares, at least two disease activity measurements were required. Of the remaining patients, 1426 had more than one study visit; however, 77 of patients were observed only during pregnancy and were removed from the study population due to likely being systematically different from patients routinely followed in the cohort. The final analytic cohort consisted of 1349 women, including 304 women who had 398 pregnancies. There were 381 observed postpartum periods, with at least one visit during the postpartum period.
All women in the cohort between the ages of 14 and 45 were included in the analysis, regardless of pregnancy status. The time of entry into the cohort was considered the initial measurement for all women. Patients were right censored and removed from the risk set at age 45, menopause, loss to follow-up, death or the end of follow-up. If patients had a gap of more than 1 year in study visits, patients were considered lost to follow-up but were allowed to re-enter the cohort when study visits resumed.
If a woman had more than one pregnancy, all pregnancies and postpartum periods were included. Incidence rate ratios and 95% CI were calculated separately for pregnancy and postpartum periods compared with non-pregnant/non-postpartum periods. The HRs of flares were estimated using a stratified Cox model based on the Prentice, Williams and Peterson total time approach using the PHREG procedure.24 25 A stratified Cox model is a conditional model that does not assume independence of multiple events of flare.26 Instead, the model took into account that a patient was not at risk for a second flare without having experienced a first flare. Using the same model, relative hazard rates of flare were calculated between (1) pregnant and non-pregnant/postpartum periods and (2) postpartum and non-pregnant/postpartum periods. Due to repeated events of flares being counted in the same patient and patients being allowed to exit and re-enter the analytic cohort, 95% CI were estimated with 1000 bootstrap replications sampled with replacement.27 Potential covariates of interest included patient race, age at diagnosis, age at baseline and duration of disease at baseline. Confounders were defined by a 10% change in beta estimates when included in the model. None of these covariates were found to be confounders and were not included in any models.
Prednisone and hydroxychloroquine (HCQ) were explored as time-varying covariates. To test whether effects were similar for HCQ users and non-users, as well as prednisone users and non-users, interaction terms between each medication with the exposure were included in the model. Models exploring effect modification by HCQ, for example, would (1) estimate the HR of flare for pregnant women taking HCQ compared with non-pregnant/non-postpartum women taking HCQ, (2) estimate the HR of flare for pregnant women not taking HCQ compared with non-pregnant/non-postpartum women not taking HCQ and (3) compare these two HR to see if the HR for women taking HCQ differed from the HR for women not taking HCQ. Effect measure modification was determined by likelihood ratio test (α=0.20).
To account for the time-varying exposures (pregnancy, postpartum period and medication use) and the possibility for each individual to experience multiple events, a new patient ID was created when the exposure changed for each patient to account for the censoring that occurred in the model with a change in exposure and clustering within an individual as well as within a certain exposure group.28 Both the original ID, to account for correlation across all observations for an individual, and new ID, to account for correlation within each exposure period for an individual, were included in the models. In order to determine if all women in the cohort were an appropriate comparator group for women who became pregnant, we performed a sensitivity analysis that included only women with an observed pregnancy in the cohort (n=304). To determine if changes in clinical practice affected the results, models were stratified by time period: before and after 2000. All analyses were conducted in SAS V.9.3 (Cary, North Carolina, USA).
The median age at cohort entry was 30.6 years and 29.4 years at the first pregnancy (table 1). The median duration of SLE at cohort entry was 2.0 years, and the median follow-up was 3.9 years. Of the 398 pregnancies, 85% were live births, of which 29% were preterm. HCQ was taken during 58% of pregnancies, and 80% of patients took HCQ at some point during follow-up. Forty-five per cent of pregnancies occurred between 1987 and 2000. The median number of visits was 5 during pregnancy, 1 postpartum and 11 in non-pregnant/non-postpartum periods.
PGA flares were more common during pregnancy compared with outside of pregnancy (table 2; HR: 1.59; 95% CI 1.27 to 1.96). There was no evidence of an increased rate of flare during the 12-month postpartum period, but there was an increase in flare in the initial 3 months postpartum (HR: 1.48; 95% CI 1.07 to 1.95). In the sensitivity analysis of only patients with an observed pregnancy, the incidence of flares during non-pregnant/non-postpartum periods decreased, suggesting women who became pregnant while in the cohort had, on average, fewer flares than women without a pregnancy. During pregnancy, almost half of flares occurred during the 3rd trimester, while 24% occurred during the 1st trimester. One-third of flares during pregnancy were scored PGA 2 or higher, compared with 40% of flares during non-pregnant, non-postpartum times.
When flares were defined by SELENA-SLEDAI, results were comparable to PGA (table 3), with a higher rate of flare during pregnancy (HR: 1.57; 95% CI 1.25 to 1.92). There was no evidence of an increased rate 12 months postpartum, but there was a non-statistically significant increase in flares in the initial 3-month postpartum (HR: 1.37; 95% CI 0.94 to 1.82). Similar to models of PGA flares, the incidence of SELENA-SLEDAI flare decreased during non-pregnant/non-postpartum periods when only women with an observed pregnancy were included. SELENA-SLEDAI flares most commonly occurred during the 3rd trimester (54% of flares). Half of flares during pregnancy and 45% of flares during non-pregnant, non-postpartum time were mild, with a score of 4–8. Only 15% of flares during pregnancy and 20% of flares during non-pregnant, non-postpartum times were scored ≥12.
HCQ use was found to be an effect modifier in the association of pregnancy and flares. The increase in flares during pregnancy appeared to only be present in women not taking HCQ. When flares were measured by PGA, the HR of flares in pregnancy compared with non-pregnant/non-postpartum periods was 1.83 (95% CI 1.34 to 2.45) for patients with no HCQ use and 1.26 (95% CI 0.88 to 1.69) for patients with HCQ use (likelihood ratio P value: 0.04; table 4). While HCQ appeared to have a similar effect in the 3-month postpartum period, with the HR of flares 1.63 (95% CI 1.04 to 2.39) without HCQ and 1.25 (95% CI 0.71 to 1.87) with HCQ, the difference did not meet our statistical definition for modification. When flares were measured by SELENA-SLEDAI, there was a modest decrease in the association between pregnancy and flares for women taking HCQ, but not to the extent that HCQ would be considered an effect modifier. However, when limited to only patients with an observed pregnancy in the cohort, the HR of SELENA-SLEDAI flares in pregnancy compared with non-pregnant/non-postpartum periods was 2.09 (95% CI 1.39 to 2.97) for patients with no HCQ use and 1.49 (95% CI 0.92 to 2.08) for patients with HCQ use (likelihood ratio P value: 0.07). No differences in race, age at diagnosis, age at baseline and duration of disease were found between HCQ users and non-users.
When the cohort was limited to visits after the year 2000, results were similar for flares defined by PGA and SELENA-SLEDAI (online supplementary table 1). Prior to 2000, more person-time during pregnancy was unexposed to HCQ (63.2 person-years (PY) compared with 20.9 exposed to HCQ). After 2000, the majority of patients during pregnancy, as well as most patients during non-pregnant periods, were treated with HCQ (102.6 PY compared with 34.1 PY unexposed to HCQ during pregnancy). When flares were measured by PGA, the HR of flares during pregnancy remained higher in patients not taking HCQ in both time periods (online supplementary table 2). When flares were defined by SELENA-SLEDAI (online supplementary table 3), however, the HR of flares during pregnancy was increased for patients unexposed to HCQ in the time period prior to 2000, but an opposite effect was observed in the time period after 2000.
Supplementary file 1
Prednisone use was only found to be an effect modifier in the association of pregnancy and flares in the sensitivity cohort of patients with an observed pregnancy when flares were defined by SELENA-SLEDAI. The HR of flares in pregnancy compared with non-pregnant/non-postpartum periods was 1.91 (95% CI 1.23 to 2.81) in patients with no prednisone use and 1.44 (95% CI 0.98 to 2.01) in patients with prednisone use (likelihood ratio P value: 0.16). There was no evidence for modification by prednisone use in PGA models or other SELENA-SLEDAI models.
Previous studies found conflicting results about whether lupus was more or less likely to flare in pregnancy.8–10 13 14 29 The present analysis is the largest cohort study to date and includes data collected over almost 30 years. When compared with non-pregnant women with SLE, pregnant women and recently pregnant women did appear to flare more frequently. However, women taking HCQ did not appear to have an increased risk of lupus flare in pregnancy or the postpartum period. Prednisone may also play a role in decreasing disease activity during pregnancy, as it was found to be an effect modifier for SELENA-SLEDAI flares in the sensitivity cohort of patients with an observed pregnancy.
The results support what has previously been reported in the literature, both within the Hopkins Lupus Cohort and in other pregnancy cohorts.4 8 9 13 The initial effort to determine the impact of pregnancy on lupus activity in this cohort was completed by Petri and colleagues in 19918 and found that among the first 40 pregnant patients in this cohort, the rate of flare was greater during pregnancy (1.6 flares per PY) compared with non-pregnant controls (0.7 flares per PY). A previous analysis in this cohort by Clowse et al 4 reported that among patients seen at least 6 months prior to pregnancy, 12.5% had high disease activity (PGA≥2) compared with 21.3% of patients during pregnancy. Additionally, lupus activity was greater among patients who discontinued HCQ during pregnancy compared with patients who continued.3 The current study extended previous work in the Hopkins Pregnancy Cohort by analysing a longer follow-up period and including comparisons to postpartum periods. Interestingly, the rate of flares per PY has dramatically decreased from the initial study, with the crude flare incidence in the entire cohort averaged around 0.6 flares per PY, ranging from 0.4 with HCQ to 0.8 without HCQ, highlighting the improvements in management of lupus during pregnancy over the past 25 years. We found that the protective effect of HCQ remained for flares measured by PGA when results were stratified prior to and after 2000. Of interest, we did not observe a similar pattern for flares measured by SELENA-SLEDAI, with a protective effect of HCQ observed in the time prior to 2000 but not after 2000.
An increased rate of flare during pregnancy has been observed in other SLE cohorts. Ruiz-Irastorza et al 9 found that the rates of flare during pregnancy and 6 weeks postpartum were increased compared with non-pregnant, age-matched controls (table 5). A study of 29 pregnancies in Hong Kong estimated a higher rate of flares during pregnancy compared with non-pregnant patients.13 However, in contrast to our results, other studies have found no evidence of an increased rate of flare during pregnancy,10 14 29 potentially due to differences in patient ethnicity, study design, sample size or definition of flare.
Fewer studies have examined postpartum flares. In this same cohort, Petri et al 8 reported a lower mean rate of flare after delivery than during pregnancy among 42 patients, with the rate of flare decreasing from 1.6 flares per PY during pregnancy to 0.7 per PY in the year after delivery. A study in Argentina observed 19% of patients flared during pregnancy, compared with 4% in the puerperium.6 Ruiz-Irastorza et al 9 estimated a rate of flare during pregnancy of 0.08 per person-month, compared with 0.15 per person-month 8 weeks after pregnancy outcome, which decreased to 0.05 1-year postpartum. In our analysis, we defined the postpartum period according to two definitions and observed an increase rate of flare during a 3-month postpartum period, yet no increased rate during a 12 month postpartum period, suggesting the increased risk of flare experienced during pregnancy remains for several months postpartum.
We estimated HRs using stratified Cox models, which take into consideration the order in which flares occurred and allowed different baseline hazards based on the number of previous flares a patient had in the cohort.26 Given that a patient with no history of flares likely has a different baseline hazard of flare than a patient who has had multiple previous flares, a model that takes this into account seems more appropriate. A limitation of our study design was patients were censored in the model when the exposure changed. We accounted for this by creating a new ID variable when a patient’s exposure changed and included the original and new IDs in the model. However, this caused a patient’s stratum for previous flares to be limited to the current exposure period, which may result in residual confounding. Even so, we view this residual confounding to be preferable to the potentially biased estimates of a crude model or an unadjusted counting process Cox model that would not account for any previous flares.
The present analysis benefited from including two disease activity indices in the same analytic cohort, which allowed us to compare how results might differ depending on the flare index used. We found that, although more flares were observed by SELENA-SLEDAI, the HRs for both indices were similar. Additionally, using data from all women enrolled in the cohort allowed us to analyse more of the disease history of patients. Because all patients may not be the most appropriate comparator group for women who became pregnant, we conducted a sensitivity analysis restricted to women with an observed pregnancy. We found that non-pregnant/non-postpartum flare rates do change depending on the group of women analysed, with women who had a pregnancy having a lower incidence of flare during non-pregnant/non-postpartum periods. We also considered patients who had more than a 1 year gap between study visits to be considered lost to follow-up, although patients were allowed to re-enter the analytic cohort. This was done to include patients who were under routine care and to allow for an appropriate comparator score for the calculation of disease flare. The disease activity of these patients during unobserved periods is unknown, and if a gap in visits was due to remission of the disease, it is possible we underestimated the person-time for low disease activity periods. Although we did not find any differences between HCQ users and non-users, there remains a possibility that non-users were patients with an allergy to HCQ, intolerant to HCQ or refused to take the medication. Additionally, flares captured in the analysis were based on flares observed at the Lupus Center; therefore, we were unable to include flares that occurred during hospitalisations.
Our study supports prior data suggesting HCQ may prevent lupus flares during pregnancy and now suggests that it also may prevent postpartum flares. While in prior decades many women with lupus were expected to flare during or after pregnancy, more recent data suggest that a large proportion of women have minimal disease activity throughout the period. We hypothesise that routine continuation of HCQ in modern lupus pregnancies may be the driving force for the diminution in lupus activity during and following pregnancy. The results suggest we can be more optimistic with many women with lupus: they do not need to expect a lupus flare during or after pregnancy, particularly if they continue HCQ.
Handling editor Josef S Smolen
Contributors All authors were involved in the conception and design of the study. MP contributed to the acquisition of the data. All authors contributed to analysis and interpretation of the data. All authors participated in the drafting of the manuscript or revised it critically for intellectual content. All authors read and approved the final manuscript.
Funding The Hopkins Lupus Cohort is supported by NIH AR 43727 and AR 69572.
Competing interests None declared.
Patient consent Detail has been removed from this case description/these case descriptions to ensure anonymity. The editors and reviewers have seen the detailed information available and are satisfied that the information backs up the case the authors are making.
Ethics approval Johns Hopkins University School of Medicine Institutional Review Board and University of North Carolina at Chapel Hill Institutional Review Board.
Provenance and peer review Not commissioned; externally peer reviewed.
Presented at This work was previously presented as an oral presentation at the American College of Rheumatology Annual Meeting 2016 in Washington, DC, 11–16 November 2016. Eudy AM, Siega-Riz AM, Engel S, et al. Effect of Pregnancy on Disease Flares in Patients with Systemic Lupus Erythematosus (abstract). Arthritis Rheumatol 2016;68(suppl 10).
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.