NSAID exposure delays time-to-pregnancy in patients with spondyloarthritis: an analysis of the GR2 prospective cohort.

Spondyloarthritis (SpA) (both axial and peripheral) is among the most frequent chronic rheumatic inflammatory diseases (CRID), particularly in people of childbearing age. 1 2 This disease causes pain, stiffness and major disability in patients in their reproductive years. Left untreated, SpA may lead to severe loss of function in these young patients, with a major impact on their quality of life. SpA has been classically seen as a predominantly male disease, but recent data have indicated little to no differences in the sex ratio prevalence of the disease. 3 4 Therefore, the issue of desire for pregnancy and treatment management in this situation is an increasingly common question in clinical practice. 5 Impaired fertility and poorer ovarian function have been reported in patients with other CRID, such as rheumatoid arthritis (RA), 6 10 with longer time to pregnancy (TTP) associated with disease activity but also with preconceptional exposure to glucocorticoids and non-steroidal anti-inflammatory drugs (NSAIDs). 11 Little is known about the impact of disease activity and treatment strategies on fertility and pregnancy outcomes in SpA. 12 A systematic literature review 13 reported only four studies in women with SpA with information on fertility, and all had a high risk of bias 13 that precluded drawing any conclusions. All were based on retrospective designs, with a risk of recall bias. This was highlighted, for example, in one of them that showed a significant discrepancy between the self-reported data and the medical records. 14 Other factors are suspected to have a potentially negative influence on fertility in women with SpA. One of these such is the chronic use of NSAIDs, particularly selective COX-2 inhibitors, which can lead to anovulatory cycles due to the luteinised unruptured follicle syndrome. 15 17 However, the potential negative impact of NSAIDs on fertility outcomes has not been evaluated in women with SpA, although it seems particularly relevant as NSAIDs remain the cornerstone treatment in this disease. 18 20 More recently, the Norwegian RevNatus registry showed that TTP exceeded 12 months (the threshold for defining ‘infertility’) in 21% of 274 women with SpA followed up between 2006 and 2018. In this cohort, longer TTP was associated with older age, nulliparity and longer disease duration. 21 These points led us to conduct this study, which aims to describe TTP in a cohort of women with SpA and to evaluate the association of TTP with disease activity and preconceptional treatments.

We report data from the GR2 ( NCT02450396 ) study, a French multicentre prospective observational study of women desiring pregnancy and a rare and/or rheumatological diseases, including SpA (both axial and peripheral phenotypes). This study has been underway since October 2014 in 63 centres and is described elsewhere. 22 24 Briefly, women were included by their clinicians (internists or rheumatologists), and treatment decisions were left to these physicians’ discretion. The GR2 study is part of the European network of pregnancy registers in Rheumatology supported by Foundation for Research in Rheumatology 25 and follows European Alliance of Associations for Rheumatology recommendations regarding core data sets for these registers. 26 Investigators provided all patients with written information and obtained oral consent from them, in accordance with French regulations for observational non-interventional studies. Women with SpA (both axial and peripheral phenotypes, diagnosed by their rheumatologist) were included between December 2015 (although inclusion in the overall cohort began in October 2014, inclusion for women with RA and SpA began in December 2015) and June 2021 (this date was chosen to have at least 1 year of follow-up after the inclusion of women desiring pregnancy). Patients could be included in either the ‘preconception’ module or the ‘pregnancy’ module of the study. In the ‘preconception’ module, consecutive women with SpA and desiring pregnancy (who had recently ceased contraception or planned to do so in the near future) were included in the ‘preconception module’ and were followed up yearly until pregnancy occurred or the patient no longer desired to be pregnant or after 5 years of trying to conceive, whichever arrived first. Accordingly, one desired pregnancy corresponds to one inclusion. If a patient was included twice or more for more than one pregnancy, only the first inclusion was analysed. If during follow-up in the ‘preconception’ module a pregnancy was confirmed, patients transferred into the ‘pregnancy’ module, that is, they were followed up at least once per pregnancy trimester and at 12 months post partum. Women with SpA not included in the preconception module who had a confirmed pregnancy at less than 14 weeks of gestation could be immediately included in the ‘pregnancy’ module. Only patients included in the preconception period (ie, via the ‘preconception’ module) were included in the main analysis. TTP was also calculated in the population of patients included via both modules (‘preconception’ and ‘pregnancy’), both as a sensitivity analysis and to explore if entering the cohort by the ‘preconception’ module created a potential selection bias. That is, we examined the hypothesis that patients who became pregnant immediately after stopping contraception might not have had the time to enter the ‘preconception’ module and were included directly in the ‘pregnancy’ module), as TTP was also collected at the inclusion visit in the latter module, although retrospectively. At the first visit of the ‘preconception’ module, we collected the following data: demographics, SpA characteristics (including the items enabling calculation of the ASAS (Axial Spondyloarthritis Assessment international Society) criteria 27 ), SpA history, current SpA treatment (symptomatic, ie, analgesics, NSAIDs, glucocorticoids and steroidal injections; conventional synthetic disease-modifying antirheumatic drugs (csDMARDs): methotrexate, leflunomide, sulfasalazine, azathioprine and hydroxychloroquine and biological disease-modifying drugs ((b)DMARDs), eg, TNF inhibitors), obstetrical history (any previous pregnancies and their outcomes: early miscarriage, stillbirth or live-birth), treatments received and events during pregnancy (infections/disease flares/others), comorbidities (thrombo-embolic disease, hypertension, diabetes, dyslipidaemia) and status for hepatitis B and C viruses and HIV. At each visit (including the first one), disease activity and its severity, and ongoing treatments were collected at the first visit and every subsequent visit. The primary outcome was TTP in months, defined as the time between the cessation of contraception and confirmed pregnancy. The secondary outcome was the number of subfertile patients (ie, patients with a TTP>12 months 28 or non-achievement of pregnancy). Disease activity was collected by the Bath Ankylosing Spondylitis Disease Activity Index (BASDAI). 29 C reactive protein (CRP) was collected at each preconceptional visit. Although ASDAS 30 is the preferred disease activity outcome, we chose to consider BASDAI and CRP separately so that we could assess the potential effect on TTP of both ASDAS components—systemic inflammation with CRP and disease activity burden with BASDAI. The clinical research form included fillable fields for specifying the start date (either the exact start date if within the past 12 months or the mention of treatment for more than 12 months), the end date (only if it was stopped during the follow-up) and the dose and frequency of intake. Without a treatment start or end date, we considered it was taken before or after the follow-up period, respectively. First, we described the analysis populations (main and sensitivity analyses) in terms of demographics, disease characteristics, disease activity, severity and treatment history. The different treatment exposures during preconceptional follow-up were described and at least one dose during this period constituted exposure. Continuous variables normally and not normally distributed were expressed, respectively, by their means with SDs and medians with IQRs. Second, TTP for the main analysis was estimated with survival models (Kaplan-Meier). Moreover, to determine the factors associated with a pregnancy occurrence (ie, the event), we used shared frailty Cox models for the multivariate analysis, including a random centre effect 31 to consider potential heterogeneity of practices among centres. The explanatory variables were selected based on the univariate analyses and current knowledge of factors potentially associated with fertility disorders. The threshold of p<0.2 in univariate analyses was used determine the introduction of variable in the multivariate model. Several variables were forced into the model because of the strong evidence of their negative impact on fertility: current smoking, disease activity, nulligravidity and body mass index (BMI). Exposure to csDMARDs and TNF inhibitors was also forced into the model to examine the effect of therapeutic classes significantly associated with a longer TTP, independently of exposure to other classes. Results are presented as HRs with their 95% CIs. For the sensitivity analysis, TTP was estimated by combining the prospective observations of the TTP (months) in the ‘preconception’ module and the TTP (months) recalled by the patients, included directly in the ‘pregnancy’ module. No analysis of associated factors was performed in the sensitivity analysis population, as prospective records of disease activity were not available for them. Missing data were addressed by using multiple imputation by chained equations (MICE), which involves generating multiple imputed datasets to take into account the uncertainty associated with missing values. The imputation process was performed with 100 iterations to ensure robust and reliable estimates. All analyses were conducted with R. Studio R V.4.0.3 (2020-10-10) and used the following packages: prettyR, tidyverse, lubridate, dplyr, survival, coxme, survminer and mice.

From December 2015 to June 2021, 207 women with SpA were enrolled. Among them, 91 (44.0%) were included in the ‘preconception’ module and had a prospective follow-up during the preconception period; 88 of them were finally retained in the fertility analysis ( figure 1 ). Three patients were included twice in the cohort for two different desired pregnancies; only the first inclusion was analysed (the first pregnancy was considered the event of interest, while the follow-up was censored according to the methodology of survival analyses). In this 88-patient main analysis population, the mean age was 31.6 (±4.1) years and the mean BMI 23.8 (±4.9) kg/m 2 . 44 (50%) were nulligravid and 17 (19.3%) smokers in the preconception period ( table 1 ). SpA was axial in 36 (40.9%) cases, with a median duration of 5.0 (2.0–9.0) years at inclusion. The extra-articular manifestations included 13 (14.8%) patients with uveitis, 18 (20.5%) with psoriasis and 7 (8.0%) with inflammatory bowel disease. Severity signs were found for a history of coxitis for 3 (3.4%), total hip replacement for 1 (1.1%) and syndesmophytes in 2 (2.3%) cases. 61 women (69.3%) had received at least one TNF inhibitor for SpA, and 10 (11.4%) had at least 3 different TNFis for SpA ( table 1 ). BMIbody mass indexcsDMARDsconventional synthetic disease-modifying antirheumatic drugsNSAIDsnon-steroidal anti-inflammatory drugsSpAspondyloarthritisTNFitumour necrosis factor inhibitor Among the 116 SpA patients who were included directly in the ‘pregnancy’ module, information on TTP was available for only 55 patients. The sensitivity analysis population thus included 143 patients. That is, pregnancy was not planned for 19/116 and thus TTP could not be collected for them; overall, 42/116 patients had missing TTP data. The characteristics of the populations in the main and sensitivity analyses were relatively similar ( table 1 ). Of the 88 women included preconceptionally, 56 (63.6%) achieved a clinical pregnancy and 40 (45.4%) did not and were classified as subfertile. Among these 56, the median TTP was 16.1 months (95% CI (12.2 to 25.3)). On the whole, disease activity was controlled at the inclusion visit with a mean BASDAI of 2.9 (± 2.1). Patients were mainly treated with TNF inhibitors during the preconception period (69.3%), while 23 (26.1%) received NSAIDs. Only 12 (13.6%) were on csDMARDS, and 8 (9.1%) on glucocorticoids ( table 2 ). A description of the NSAIDs used by patients at inclusion is provided in online supplemental table 1 . CRP, C reactive proteincsDMARDs, conventional synthetic disease-modifying antirheumatic drugs; NSAIDs, non-steroidal anti-inflammatory drugs; score, Bath Ankylosing Spondylitis Disease Activity Index score; TNFi, tumour necrosis factor inhibitor Univariate analyses found p<0.2 for the associations of TTP with preconceptional NSAID exposure, disease duration, age and SpA phenotype. These variables were included in the multivariate model, to which we added csDMARDs and TNF inhibitor exposure during the preconception period, nulligravidity, BMI, disease activity and current smoking, as explained above. The multivariate shared frailty model (adjusted for age, BMI, nulligravidity, BASDAI at inclusion, disease duration, smoking, SpA phenotype—axial, peripheral or both—and exposure to NSAIDs, csDMARDs and biological treatment in the preconception period) found that TTP was significantly and independently associated with age (HR (per year) 1.22, 95% CI (1.08 to 1.40); p<0.001) and preconceptional NSAID use (HR 3.01, 95% CI (2.15 to 3.85); p=0.01) ( table 3 ). Median TTP was 31.6 months (95% CI (22.3 to 40.4)) in women who were exposed to NSAIDs during the preconceptional period, vs 12.3 months (95% CI (10.9 to 20.3)) in women who were not (p=0.01) ( figure 2 ). Statistically significant results are presented in bold. BASDAI scoreBath Ankylosing Spondylitis Disease Activity Index scoreBMI, body mass index; csDMARDs, conventional synthetic disease-modifying antirheumatic drugs; NSAIDsnon-steroidal anti-inflammatory drugsSpA, spondyloarthritis; TNFi, tumour necrosis factor inhibitor The sensitivity analysis for TTP reassessment in the overall population included patients in the ‘preconceptional’ and those included only after pregnancy onset—for whom TTP was collected retrospectively. Median TTP was 9.1 months (95% CI (6.4 to 11.6)), and the subfertility rate was 33.6%.

Our study is the first to evaluate TTP prospectively in women with SpA, including all of its types. The aim of this study was to describe TTP prospectively in patients with SpA who desired to become pregnant. In the main analysis population (ie, patients with a prospective observation of TTP), the median TTP was 16.1 months and the subfertility rate 45.4%. Factors associated with a longer TTP were age and preconceptional NSAID use. These results should be interpreted in the light of general population data: a study from 1999 showed a prevalence of subfertility of 16% in a general Western European population (17.2% in a population of 523 French women) and a median TTP of 3 months. 32 A later study based on population surveys from 25 countries found a mean prevalence of subfertility of 9%. 33 In a more recent study among a French general population of women of childbearing age who were trying to conceive, 24% did not achieve a pregnancy after 12 months. 34 Thus, subfertility appears more common in women with SpA, even though we were unable to make a direct comparison with a control group. We chose to analyse TTP for the assessment of fertility. This method is considered superior to fertility analysis with anti-Müllerian hormone (AMH), due to the low correlation between AMH and the occurrence of pregnancy in healthy populations. 35 36 AMH has proven its usefulness in cases of infertility to predict response to ovarian stimulation and its use is standard for dose-adjustment of treatment against infertility. 37 39 Studies in healthy populations, however, have not found that its utility extends to them. 40 41 Nonetheless, these elements must be clarified specifically in women with SpA, as a recent prospective cohort found lower AMH measurements and follicular counts in patients with radiographic axial SpA compared with healthy controls. 42 Some authors also suggest that TTP use for fertility assessment must always be prospective to avoid recall bias. 43 45 Only one recent study has specifically investigated TTP in women with SpA. 21 It included 274 women with axSpA and found a median TTP of 2 months, with 21.2% of the women subfertile. Ursin et al also found that age, nulliparity and disease duration were associated with a longer TTP. The differences in our results might be explained by differences in our respective populations. Women in our cohort were more likely to be smokers than those in theirs (19.3% vs 6.4%). They also excluded women with a known cause of infertility. Moreover, our cohort’s recruitment took place mainly in tertiary centres, whereas patients from Ursin’s cohort received follow-up care in all kinds of centres. This might well explain why a history of TNF inhibitor treatment was observed in nearly 70% of our cases, and more than 10% of patients had already received at least three different TNFi at inclusion. By comparison, the proportion of women on TNF inhibitors in the study by Ursin et al was 49%, a figure suggesting that the women with SpA in our study had more severe cases. Our study highlights the negative impact of NSAIDs on the fertility of women with SpA, with a median TTP threefold greater in women exposed to NSAIDs preconceptionally. Their impact on ovulation, implantation 46 and risk of miscarriage is well documented in the literature, 47 48 although less well for CRID. They could play a role in subfertility by inhibiting COX-1 and COX-2, which are necessary for follicle wall rupture and ovum release, 49 50 which may lead to luteinised unruptured follicle syndrome. Furthermore, animal studies in mice have shown that COX-2 inhibition decreases implantation after fertilisation. 51 In humans, prostaglandin synthesis (eg, by COX-2 inhibition) disruption is reported to be associated with repeated IVF (ie, suggesting that reduced prostaglandin synthesis in the human endometrium may lead to poor endometrial receptivity). 52 Two studies also found a higher risk of pregnancy loss when NSAIDs were used close to conception 47 or for prolonged treatment. 48 The American College of Rheumatology has thus conditionally recommended discontinuation of NSAIDs during preconception for patients having difficulty conceiving (and if disease control would not be compromised). 53 Moreover, Brouwer et al 11 examined fertility in 245 women with RA and found an association between NSAID use during the preconception period and longer TTP. Nonetheless, this association has never been demonstrated in women with SpA. Several articles also show a transient and reversible effect of NSAID use on fertility, with recovery of normal ovulation on cessation of treatment, which makes cessation a valuable modifiable factor. 54 56 Unfortunately, in our study, the small number (n=3) of patients exposed to Cox-2 inhibitors precluded us from performing an analysis stratified by the NSAID class. The use of biologicals in the preconceptional period was not associated with an increased TTP in our cohort, consistent with the (scarce) data in the literature about their impact on fertility. A retrospective study has even shown that in women with RA desiring to conceive, treatment with biologicals might be associated with a shortening of the TTP. 57 We also found that disease activity did not have a deleterious impact on TTP in this cohort. Disease activity was generally well controlled during the preconception period in this cohort. Nonetheless, we had only one BASDAI measurement in the preconceptional period for most patients (at the inclusion visit), which hampered our ability to conduct analyses that considered the variation in disease activity over time in patients. Finally, almost 19% of our patients were current smokers at the time of inclusion, which is in line with French smoking prevalence in the general population of 22% in women. 58 Our study has several limitations. The statistical power was limited by the relatively small patient sample. Moreover, we could not perform subgroup analyses for the different SpA phenotypes because of the number of participants. The importance of missing data, especially concerning disease activity, may also have hampered the power of our analyses. Nevertheless, we addressed this limitation by using multiple imputations. Some data were not available in the database, such as quality of life and frequency of sexual intercourse, which may affect the primary outcome. Despite the national and multicentre aspect of the cohort, patients were all recruited from tertiary rheumatology centres, which may result in a selection bias in favour of potentially more severe forms of the disease. Nevertheless, although the rheumatology follow-up took place in the tertiary centres, gynaecological and/or obstetric preconceptional and pregnancy care did not do so systematically. This information is not collected in the current version of the clinical research form. Finally, there was insufficient data on the frequency and dose of NSAIDs used in the cohort to refine the results according to these parameters. Our study also has several strengths worth mentioning. The multicentre national design allowed a more representative cohort. Prospective patient follow-up and data collection limit the biases of retrospective studies and allow us to analyse the effect of SpA treatment exposure and disease activity on fertility. This is particularly important as the literature is almost devoid of data about fertility, and most of it comprises retrospective studies at high risk of bias. The addition of a random centre effect in the statistical models allowed this factor to be considered in the analyses. We also performed a sensitivity analysis to reassess TTP in the overall population. It showed that the TTP and subfertility rate in the preconception population may indeed be overestimated by the selection bias of these patients, as those who became pregnant easily were more likely to be included after pregnancy onset. Our study shows that women with SpA appear to have a longer TTP and a higher subfertility rate, with the longer TTP associated with maternal age and preconceptional NSAID use. More research is needed to confirm these findings and understand the mechanisms behind this association, as well as to study the role of disease activity in greater depth. TFN-inhibitors, given the increasing evidence of their fertility benefits, 59 the minimal safety concerns with their use during pregnancy, 60 and their NSAID-sparing effect, 61 62 represent an effective and safe treatment alternative for patients receiving continuous NSAIDs who are trying to conceive. Our findings underline the importance of preconceptional counselling in all patients who desire pregnancy and the need to reduce or cease NSAID intake in SpA patients struggling to conceive.