Article Text

Download PDFPDF
Extended report
Risk of preterm delivery and small-for-gestational-age births in women with autoimmune disease using biologics before or during pregnancy: a population-based cohort study
  1. Nicole W Tsao1,2,3,
  2. Eric C Sayre3,
  3. Gillian Hanley4,
  4. Mohsen Sadatsafavi1,2,5,
  5. Larry D Lynd1,2,6,
  6. Carlo A Marra7,
  7. Mary A De Vera1,2,3
  1. 1 University of British Columbia Faculty of Pharmaceutical Sciences, Vancouver, British Columbia, Canada
  2. 2 Collaboration for Outcomes Research and Evaluation, Vancouver, Canada
  3. 3 Arthritis Research Canada, Richmond, British Columbia, Canada
  4. 4 University of British Columbia Faculty of Medicine, Department of Obstetrics and Gynecology, Vancouver, British Columbia, Canada
  5. 5 Centre for Clinical Epidemiology and Evaluation, Vancouver, British Columbia, Canada
  6. 6 Centre for Health Evaluation & Outcomes Sciences, Vancouver, Canada
  7. 7 Otago School of Pharmacy, Dunedin, New Zealand
  1. Correspondence to Dr Mary A De Vera, Faculty of Pharmaceutical Sciences, University of British Columbia, Vancouver, BC V6T 1Z3, Canada; mdevera{at}mail.ubc.ca

Abstract

Objectives To assess the risk of preterm delivery and small-for-gestational-age (SGA) births in women with autoimmune diseases using biologics before or during pregnancy.

Methods Using population-based administrative data in British Columbia, Canada, women with one or more autoimmune diseases who had pregnancies between 1 January 2002 and 31 December 2012 were included. Exposure to biologics was defined as having at least one biologic prescription 3 months before or during pregnancy. Each exposed pregnancy was matched with five unexposed pregnancies using high-dimensional propensity scores (HDPS). Logistic regression modelling was used to evaluate the association between biologics use and preterm delivery and SGA.

Results There were 6218 women with 8607 pregnancies who had an autoimmune disease diagnosis; of which 109 women with 120 pregnancies were exposed to biologics 3 months before or during pregnancy. In unadjusted analyses, the ORs for the association of biologics exposure with preterm deliveries were 1.64 (95% CI 1.02 to 2.63) and 1.34 (95% CI 0.72 to 2.51) for SGA. After HDPS matching with 600 unexposed pregnancies, the ORs for the association of biologics exposure and preterm deliveries were 1.13 (95% CI 0.67 to 1.90) and 0.91 (95% CI 0.46 to 1.78) for SGA. Sensitivity analyses using HDPS deciles, continuous HDPS covariate or longer exposure window did not result in marked changes in point estimates and CIs.

Conclusions These population-based data suggest that the use of biologics before and during pregnancy is not associated with an increased risk of preterm delivery or SGA births.

  • biologics
  • pregnancy
  • preterm
  • small-for-gestational-age
  • autoimmune disease

Statistics from Altmetric.com

Request Permissions

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.

Introduction

Pregnancy is a unique state of coexistence of genetically different individuals, which is possible due to dramatic shifts in maternal immune function during pregnancy, protecting the fetus from immunological attack.1 In women with chronic inflammatory diseases, this interaction between autoimmunity and pregnancy becomes complex. The pathology underscoring autoimmune diseases including rheumatoid arthritis (RA), ankylosing spondylitis (AS), psoriatic arthritis (PsA), psoriasis (Ps) and inflammatory bowel disease (IBD) are perpetuated mainly by the dysfunction of cytokines and chemokines regulating immune system activity, with tumour necrosis factor (TNF)-alpha being a key cytokine in this abnormal immune response.2–5

In pregnancy, TNF-alpha controls cyclo-oxygenases that affect blastocyst implantation, endometrial permeability and decidualisation,6 and contributes to the process of labour.7 Abnormally high levels of TNF-alpha and other cytokines have been implicated in pregnancy complications including preterm delivery, fetal growth retardation, early and unexplained spontaneous abortions, and miscarriages.7–10 As such, evidence suggests that higher autoimmune disease activity at the time of conception and during pregnancy is correlated with increased risks of adverse maternal and neonatal outcomes.11 12

Biologics work to treat autoimmune diseases by modulating the immune system by targeting key inflammatory cytokines including TNF-alpha, interleukin (IL)-1, IL-6 or receptors of these cytokines.13 With these medications available only within the last 15 years, their use by women during pregnancy has been growing and becoming more clinically acceptable.14 However, prior studies on this topic included only a small number of women enrolled in registries, and with comparison groups often selected from external sources; furthermore, majority of the studies have not implemented any methods to adjust for differences in baseline characteristics between groups.15–19 The aim of this study was to assess the risk of preterm delivery and small-for-gestational-age (SGA) births—two related outcomes that remain as leading causes of infant morbidity and mortality20—in women with autoimmune diseases exposed to biologics, compared with those who were not exposed to biologics before or during pregnancy.

Methods

Data sources

Population Data British Columbia (Population Data BC) is an extensive data repository that holds individual-level, de-identified, longitudinal data on all health services covering the entire population of BC (estimated 4.6 million residents, December 201621). These include all provincially funded physician visits, laboratory tests and diagnostic procedures (X-rays, ultrasounds and so on) from the Medical Services Plan (MSP) database,22 hospitalisations from the Discharge Abstract Database (DAD)23 and demographics and vital statistics since 1985.24–26 Population Data BC also includes the comprehensive prescription drug database, PharmaNet, which captures all prescriptions dispensed in community pharmacies regardless of payment source, since 1996.27 These data were linked to the BC Perinatal Database Registry (BCPDR),28 which contains validated information on the date of conception, antenatal, intrapartum and postpartum maternal and infant data abstracted from medical records for 99% of births in BC. Altogether, linkage of these data sources created a source population comprising women (n=305 351) in BC who had one or more pregnancies (n=449 098) ending in a live or stillbirth between 1 January 2002 and 31 December 2012 and were continuously covered by BC’s provincial health plan for at least 12 months prior to the start of pregnancy and in the 12 months following delivery. Details of these data sources are described in previous work.14

Study cohort

We created a cohort of women who had a recorded diagnosis of one or more autoimmune diseases that could be treated with a biologic, which included RA, IBD (Crohn’s disease and ulcerative colitis), Ps/PsA, AS, juvenile idiopathic arthritis and systemic autoimmune rheumatic diseases—including systemic lupus erythematosus and other connective tissue diseases. These were defined as having the same ICD-9/10 code for a specific autoimmune disease from two separate physician visits that were at least 60 days apart and within 2 years, any time prior to the date of conception; or, having at least one hospitalisation with an ICD-9/10 code for an autoimmune disease any time prior to the date of conception. Given that the unit of analysis was individual pregnancy, each pregnancy had to satisfy the above criteria in order to be included in the analyses.

Exposure ascertainment

Using dispensation dates and Canadian Drug Identity Codes for biologics in PharmaNet, we identified pregnancies in women in the autoimmune disease cohort who had at least one prescription for a biologic at any point during the drug exposure period of interest for each study outcome. For preterm delivery, this period was defined as 3 months prior to the date of conception (referred to as the preconception period) until the date of delivery or 36 weeks+6 days of gestation, which ever came first, for each pregnancy. This was to avoid classifying pregnancies as exposed if they were exposed to a biologic on or after 37 completed weeks of gestation in which by definition they would not be susceptible to the outcome occurring. For SGA, the exposure period was defined as 3 months prior to the date of conception, until the date of delivery. Disease-matched women with pregnancies who were not exposed to biologics during the drug exposure periods of interest comprised the unexposed groups. All biologics available in BC for the treatment of autoimmune diseases of interest during the study period are listed in the online supplementary table 1.

Supplemental material

Outcomes

The outcomes of interest were preterm delivery and SGA births. We had access to exact date of birth for all babies born to the women in our cohort from the BCPDR, as well as valid gestational age estimates based on information from early gestational ultrasounds or from the date of last menstrual period if an early gestational ultrasound was not performed. If neither field was recorded, gestational age was estimated from newborn clinical exam and/or chart documentation. Preterm delivery was defined as a binary outcome of delivery occurring before 37 completed weeks of gestation, regardless of the reason. We also included infants with ICD-9/10 codes for preterm births from the MSP database or DAD. SGA was defined as a newborn weighing less than the 10th percentile of gestational age-specific and sex-specific weights for neonates in BC29 using birth weights recorded in the BCPDR.

Statistical analysis

To minimise bias due to confounding by indication, we used a high-dimensional propensity score (HDPS) algorithm that incorporated investigator-specified covariates and additional factors that acted as proxy variables for unmeasured confounders.30 The HDPS was generated using logistic regression models to identify candidate covariates derived from four dimensions of data comprising aforementioned data sources: (1) MSP database; (2) DAD; (3) PharmaNet and (4) BCPDR. Within the MSP database, DAD and PharmaNet, only claims or codes that occurred during the 12 months prior to the date of conception for each pregnancy were assessed as candidate covariates to be included in the HDPS. We specified the HDPS algorithm to prioritise covariates across data dimensions by their potential for controlling confounding based on the bias term estimator proposed by Bross,31 meaning that the covariates must both be associated with the exposure and the outcome to mitigate the potential for including variables that were only associated with the exposure, which may actually introduce bias into estimates.32 The top 50 empirically derived covariates for each outcome were included along with investigator specified confounders for propensity score estimation (see the online supplementary tables 2 and 3). For each outcome, biologics-exposed pregnancies were matched with unexposed pregnancies using HDPS in a ratio of 1:5 without replacement. Match performance was evaluated using standardised mean differences in baseline characteristics of matched and unmatched cohorts.

Using logistic regression models we analysed each study outcome among biologics-exposed and unexposed women in the HDPS-matched cohort (model 1). Because the length of pregnancies does not affect the risk of exposure, exposures were not modelled as time-dependent. As sensitivity analyses for each outcome, we conducted multivariable logistic regression models with deciles of HDPS included as indicator terms (model 2) and with continuous HDPS as a covariate (model 3). As sensitivity analysis for the exposure, we defined the exposure window beginning at 12 months prior to conception for both outcomes, and used HDPS matching (model 4). Using robust variance estimators to account for correlation between multiple pregnancies within the same woman did not appreciably change CIs in the outcome models, as such, all correlation structures were omitted. All analyses were conducted using SAS statistical software V.9.3.

Results

From a source population of 305 351 women in BC who have had one or more pregnancies over the study period, approximately 2% had a diagnosis of one of the autoimmune diseases of interest resulting in 6218 women with 8607 pregnancies in the study cohort. Table 1 shows baseline characteristics for the unmatched cohorts as well as HDPS-matched cohorts for analyses of respective study outcomes. Marked imbalances between exposure groups in the distribution of autoimmune disease types and concomitant medication use, as seen with large standardised mean differences in the unmatched cohort, were mitigated in the HDPS-matched cohorts.

Table 1

Baseline characteristics in unmatched and matched samples of biologic-exposed and unexposed pregnancies

Preterm delivery

The HDPS-matched cohort for analysis of preterm delivery outcomes comprised 109 women and 120 babies exposed to biologics during 3 months preconception to the date of delivery, and 584 women and 600 babies unexposed to biologics during that time (table 1). Most of the women had a diagnosis of RA or IBD (49% and 46%, respectively) and filled prescriptions for one of three commonly prescribed TNF-alpha inhibitors (infliximab 39%, etanercept 30% or adalimumab 25%) (table 1). In the HDPS-matched cohort, 21 of the 120 babies (18%) exposed to biologics preconception or during pregnancy and 95/600 (16%) babies unexposed to biologics were born preterm. Table 2 shows the results of crude analyses of the association between biologic exposure and preterm delivery with an unadjusted OR of 1.64 (95% CI 1.02 to 2.63). In primary analyses, the OR for the association between biologic exposure and preterm delivery was 1.13 (95% CI 0.67 to 1.90) (table 2, model 1). Sensitivity analyses involving multivariable logistic regression based on the unmatched cohort adjusting for HDPS deciles (model 2) and continuous HDPS (model 3) and extending the exposure window to 12 months preconception (model 4) did not appreciably change the results. Finally, examination of the birth data showed that the mean gestational age at delivery was 38 weeks (range 27–43 weeks) among women exposed to biologics and 38 weeks (range 19–43 weeks) among those unexposed (figure 1).

Table 2

Proportion of pregnancies ending in preterm delivery or SGA births based on exposure group and timing of biologic exposure

Figure 1

Distribution of gestational age by exposure groups.

SGA births

The HDPS-matched cohort for analysis of SGA comprised 109 women and 120 babies exposed to biologics during 3 months preconception to the date of delivery, and 585 women and 600 babies unexposed to biologics during that time. RA and IBD remained the most common disease types (45% and 48%, respectively), and infliximab, etanercept and adalimumab were the most commonly prescribed biologics (table 1). In the HDPS-matched cohort, SGA births occurred in 11/120 (9%) pregnancies in the biologics-exposed group and in 60/600 (10%) pregnancies that were in the biologics unexposed group. Table 2 shows the results of crude analyses of the association between biologic exposure and SGA with an unadjusted OR of 1.34 (95% CI 0.72 to 2.51). In primary analyses, the OR for the association between biologic exposure and SGA was 0.91 (95% CI 0.46 to 1.78) (table 2, model 1). Sensitivity analyses (models 2, 3 and 4) again showed similar results. Furthermore, examination of the Apgar scores of SGA newborns showed inappreciable differences; those exposed to biologics had mean Apgar scores of 8.1 (SD 1.5) at 1 minute, and 9.0 (SD 1.0) at 5 minutes, and those unexposed had Apgar scores of 7.7 (SD 2.2) at 1 minute and 8.7 (SD 1.7) at 5 minutes.

Discussion

Our objective was to use population-based administrative health data with valid information on estimated date of conception and complete information on all dispensed prescriptions in BC to evaluate the association between biologic exposure preconception, or during pregnancy, and preterm delivery or SGA births in women with autoimmune diseases. We applied HDPS methods—primarily matching—to account for differences in baseline characteristics between women exposed and unexposed to biologics. Prior to restricting the population using HDPS matching, we found that differences in baseline characteristics in the unmatched sample led to suggestion of an association between biologics use and the risk of preterm deliveries. However, after successful implementation of HDPS to control for confounding by indication and proxies of unmeasured confounders, we did not find an association between biologics exposure and the outcomes of interest, in primary and various sensitivity analyses. While we examined all biologics used in the cohort, TNF-alpha inhibitor biologics (94%) were the most common, and as such, our results mostly apply to these biologics and less so to those that are not TNF-alpha inhibitors.

Indeed the population-based setting of this study lends more generalisability to the results, and the implementation of HDPS-based methods allows for better control of confounding compared with traditional modelling methods, thus contributing to better understanding of the use of biologics in the pregnant population. With respect to the outcome of preterm delivery, several single-centre studies using maternal medical records have reported risks of preterm delivery in those exposed to biologics during pregnancy compared with those who were not exposed, ranging from OR 2.00 (95% CI 0.19 to 20.51) to 2.71 (0.44 to 16.52).18 19 33 Registry-based studies from the British Society for Rheumatology Biologics Register in patients with RA, and the German Registry Rheumatoide Arthritis: Beobachtung der Biologika-Therapie in patients with IBD reported risk estimates of 1.42 (95% CI 0.25 to 7.73) and 2.14 (0.10 to 44.28), respectively, of preterm deliveries in women who were using biologics before or during pregnancy.15 34 These studies have relatively small sample sizes (50–80 individuals), and have not implemented methods to adjust for the effects of the underlying disease severity or effects from measured and unmeasured confounders, as such these estimates have lower generalisability and high uncertainty, as evidenced by the wide confidence intervals. At the time of this publication, only two studies have reported adjusted risk estimates, one abstract by Chambers et al 35 and one publication by Burmester et al,36 with data from the Organisation of Teratology Information Services registry and the Adalimumab Pregnancy Exposure Registry. Chambers (total N of 722) using propensity score methods found that the adjusted HR for preterm delivery was 0.82 (95% CI 0.50 to 3.84) in pregnancies exposed to adalimumab compared with those unexposed; and Burmester (total N of 373) reported an adjusted HR for preterm delivery of 1.08 (95% CI 0.41 to 2.83) in patients with RA using adalimumab during pregnancy, compared with patients with RA not using adalimumab.

With respect to SGA outcome, there are fewer studies—only two to date—with conflicting findings. Using medical records from a university hospital, Schnitzler et al reported 6% of pregnancies exposed to infliximab ending in a very SGA birth (<5th percentile) compared with 11% of unexposed pregnancies; in our study, there were no occurrences of very SGA births. Martinez et al, using medical records, reported that among women with IBD exposed to a biologic during pregnancy, 12.5% resulted in SGA births compared with 9% among unexposed pregnant women with IBD.18 19 These rates appear similar to our results, however, again neither of these studies accounted for baseline differences between exposure groups. Thus, with respect to SGA outcome among women with autoimmune disease taking biologics, our study is the first to use population-based data to conduct analyses adjusted for measured and unmeasured confounders using HDPS.

Strengths and limitations of our study bear discussion. High-quality, high coverage, population-based databases from Population Data BC, and the linkage with the perinatal registry (BCPDR) and the prescription dispensations database (PharmaNet) provided the ability to accurately determine the timing of all medication dispensations with respect to milestone pregnancy dates, for each pregnancy in the cohort, thus minimising potential biases caused by problems such as misclassification, patient recall bias and selection bias. The comprehensive BCPDR data also allowed for the ascertainment of SGA using babies’ gestational age and birth weights, whereas currently available research focus mainly on the outcome of low birth weight, which is itself confounded by gestational age whereby about two-thirds of low-birthweight infants are preterm.37 As such, SGA is not only a more useful outcome measure, but also allowed the investigation of the impact of biologics on SGA and preterm delivery outcomes independently. Using HDPS matching is another strength which lends this study high internal validity, as it allows for better adjustment of confounding by indication and adjustment of proxies of unmeasured confounders.30 Indeed addressing confounding by indication is of utmost importance in the population of women with autoimmune disease given the association between disease activity and adverse pregnancy outcomes,11 12 and the fact that those with higher disease activity are also more likely to be on biologics given the current treatment pathways. The main limitation of our study remains the relatively small sample size in the matched cohorts; however, the use of HDPS matching inherently prioritises validity over precision of estimates, of which the latter can only be overcome by accumulation of further evidence or pooling of multiple databases.

Altogether, we found no association between biologics use before or during pregnancy and preterm delivery or SGA births in women with autoimmune diseases, compared with those who had comparable propensity to receive biologics during that time but did not. As such, our findings suggest that biologics may be a safe treatment option for women with certain autoimmune diseases who, as previous research suggest, are at higher risk of adverse pregnancy outcomes due to their disease. Given that exposures and outcomes in biologics use during pregnancy remain fairly rare, relatively small samples are a continual challenge, as such our study represents an important contribution to the accumulation of evidence on the safety of the use of biologics in pregnant women, which may lead to increased prescriber comfort and patient acceptance, decreased uncertainty and improved maternal and neonatal outcomes in this population.

Acknowledgments

The authors would like to acknowledge Mr Alexander Lin for his work on the manuscript figure. The authors would like to acknowledge Dr Jeremy Rassen for his help in answering questions and confirming our understanding of the HDPS programming.

References

Footnotes

  • Handling editor Josef S Smolen

  • Contributors NWT: had full access to all the data used in this study and takes responsibility for the integrity of the data and accuracy of data analysis. MADV and NWT: concept and design. MADV, NWT and ECS: acquisition, analysis and interpretation of data. NWT and MADV: drafting of manuscript. NWT, MADV, MS, GH, LDL, CAM: critical revision of manuscript for important intellectual content. NWT and ECS: statistical analysis.

  • Funding This research was funded by an operating grant from The Arthritis Society (YIO-13-07). MADV obtained funding for this study.

  • Disclaimer All inferences, opinions and conclusions drawn in this manuscript are those of the authors and do not reflect the opinions or policies of the Data Stewards.

  • Competing interests NWT is a PhD Candidate and Canadian Institutes of Health Research Fellowship holder. ECS is a biostatistician with Arthritis Research Canada. GH is an Assistant Professor, Canadian Institutes of Health Research New Investigator, and a recipient of Canadian Cancer Society Research Institute Capacity Development Award. MS is an Assistant Professor, Canadian Institutes of Health Research New Investigator, and a Michael Smith Foundation for Health Research Scholar. LDL is a Professor, and Director of Collaboration for Outcomes Research and Evaluation. CAM is a Professor and Dean at the Otago School of Pharmacy, New Zealand. MADV is an Assistant Professor, Canada Research Chair in Medication Adherence, Utilization, and Outcomes, The Arthritis Society Network Scholar, and Michael Smith Foundation for Health Research Scholar. LDL has received honoraria from Boehringer Ingelheim and Pfizer Canada for consulting services unrelated to this study.

  • 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 This study was approved by the University of British Columbia, Behavioural Research Ethics Board.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Correction notice This article has been corrected since it published Online First. The afffiliations list for all authors and table 2 have been updated.