Azathioprine (AZA) is recognized among immunosuppressive medications as relatively safe during pregnancy for women with systemic lupus erythematosus (SLE) requiring aggressive treatment. This pilot study aimed to determine whether SLE therapy during pregnancy was associated with developmental delays in offspring.
This cohort study included SLE patients with at least one live birth postdiagnosis. Medical histories were obtained via interviews and chart review. Multiple logistic regression was used to examine associations between SLE therapy during pregnancy and maternal report of special educational (SE) requirements (as proxy for developmental delays) among offspring. Propensity scoring (incorporating corticosteroid use, lupus flare, and lupus nephritis) was used to account for disease severity.
Of 60 eligible offspring from 38 mothers, 15 required SE services, the most common indication for which was speech delay. Seven (54%) of the 13 children with in utero AZA exposure utilized SE services versus 8 (17%) of 47 nonexposed children (P < 0.01). After adjustment for pregnancy duration, small for gestational age, propensity score, maternal education level, and antiphospholipid antibody syndrome, AZA was significantly associated with SE utilization occurring from age 2 years onward (odds ratio 6.6, 95% confidence interval 1.0–43.3), and bordered on significance for utilization at any age or age <2 years.
AZA exposure during SLE pregnancy was independently associated with increased SE utilization in offspring, after controlling for confounders. Further research is indicated to fully characterize developmental outcomes among offspring with in utero AZA exposure. Vigilance and early interventions for suspected developmental delays among exposed offspring may be warranted.
Systemic lupus erythematosus (SLE) is a multisystem, autoimmune disease associated with strong female preponderance and incidence that increases during the reproductive years (1, 2). Due in large part to improved disease management, morbidity and mortality for SLE patients over the last several decades has decreased, providing more women with SLE the opportunity to achieve successful pregnancies (3).
Among immunomodulators used for SLE, azathioprine (AZA), hydroxychloroquine, cyclosporine, and glucocorticoids are considered relatively safe for use during pregnancy (4), whereas others (cyclophosphamide, methotrexate, and mycophenolate mofetil [MMF]) are contraindicated, especially during the first 2 trimesters. However, little is known about long-term effects on children born to mothers with SLE who were treated with these medications during pregnancy. Concerns have been raised that neurocognitive outcomes in these children may be influenced by maternal SLE disease activity and its treatment during pregnancy (5–10).
We performed this study to investigate potential risk factors for increased utilization of special educational (SE) services, as a proxy for developmental delays, among offspring of mothers with SLE, and in particular to identify novel associations with specific features of lupus or its treatment during pregnancy.
Significance & Innovations
Azathioprine (AZA) exposure during systemic lupus erythematosus (SLE) pregnancy in this study was independently associated with an increased requirement for special educational services among offspring, after controlling for suspected confounders.
Increased vigilance and early intervention for suspected developmental delays among children born to mothers with SLE may be warranted, since early intervention is effective in improving long-term functioning among affected children.
Our findings do not establish sufficient risk that AZA should be withheld in lupus pregnancy requiring immunosuppressive therapy, since active, untreated disease may result in worse maternal and fetal outcomes.
PATIENTS AND METHODS
Women attending rheumatology outpatient clinics at the University of Michigan and participants in the Michigan Lupus Cohort were enrolled between December 2008 and November 2010. Patients were eligible if they met ≥4 American College of Rheumatology criteria for SLE (11, 12) prior to at least one pregnancy and had at least one live birth following SLE diagnosis, with the year of delivery occurring from 1993 onward. This research was approved by the University of Michigan Institutional Review Board; written informed consent was obtained from the mothers and assent or consent of the offspring was obtained for children ages 10–17 years.
Data were collected from the mothers through a structured interview with a study team member, as well as medical record review. The following information was obtained.
Maternal medical and obstetric history.
Data comprising SLE disease duration, history of disease manifestations (e.g., lupus nephritis, hypertension), and history of antiphospholipid antibody syndrome (APS) defined according to published criteria (12, 13) were recorded. Medication exposures were recorded, including details on fluorinated and nonfluorinated corticosteroids (nonfluorinated was considered high dose if the average daily dose was >20 mg for the majority of the pregnancy). An additional “significant steroid” variable was derived to combine exposures likely to cross the placenta into a single category (>20 mg/day of nonfluorinated steroids or fluorinated steroids) (13). Laboratory profiles, including antiphospholipid antibody and renal biopsy data, were also recorded.
Children's medical and developmental history.
Antenatal complications and delivery information, as well as medication use at conception and during pregnancy, were recorded. Children's perinatal and pediatric general health histories were collected, including time spent in intensive care units, chronic medication exposures, frequency of infections (beyond typical childhood infections), and the need for regular medical followup beyond pediatrician “well child” examinations. Small for gestational age was defined as neonatal birth weight below the tenth percentile for gestational age, based on US national reference data (14). The primary outcome was maternal report of developmental delay or utilization of SE services due to any significant lag in a child's cognitive, physical, or social maturity when compared with established age-appropriate norms. Included in this definition were delays in language and hearing, or any SE need prompting referral by the child's pediatrician or teacher for SE services (e.g., occupational or speech therapy, behavioral counseling, tutoring). In addition to collecting such data in an open-ended fashion, we specifically collected data on attention deficit hyperactivity disorder. Delays were categorized according to whether they occurred during the first 2 years of life and/or from age 2 years onward.
Baseline summary statistics were computed as the mean and SD or the median and interquartile range (IQR) for continuous variables, and the frequency and proportion for categorical variables. Nonparametric statistics were used, if appropriate. Demographic characteristics of mothers among the groups with and without SE service utilization were assessed for comparability. Although the primary outcome was utilization of SE services occurring at any age during childhood, separate models were also examined according to an SE requirement prior to age 2 years and from age 2 years onward. We compared baseline characteristics (including maternal and perinatal variables) among those children with and without SE utilization using Fisher's exact and Wilcoxon's rank sum tests. Adjustment for multiple comparisons was made by Šidák-Holm's method (15, 16). Multivariable logistic regression was used to adjust for potential confounders. Variables judged to have clinical relevance based on a priori knowledge were retained in the final models regardless of significance.
Since AZA use implies a need to treat active disease and/or maintain remission after moderate to severe disease manifestations, we derived a propensity score to account for potential confounding by indication. Using multinomial logistic regression, we found the predicted probability of AZA treatment for each patient based on nonfluorinated corticosteroid dose during pregnancy, lupus nephritis (renal biopsy World Health Organization class III–V), and SLE flare during pregnancy. This predicted probability was then used as a covariate in the multivariable logistic models examining the association between AZA and SE utilization. Some mothers had more than one child included in the study; however, correlation was ignored in the primary results, since the small sample size prohibited regression methods that take correlation into account. We therefore conducted sensitivity analyses, in which we ran a simulation on 1,000 reduced samples that consisted of one randomly selected child per mother, and examined the distribution of P values from these models. Statistical analyses were performed using Stata, version 11.1 (17).
Eighty-five women were screened for the study, of whom 43 did not meet the eligibility criteria, e.g., because pregnancies occurred prior to SLE diagnosis, and 4 were excluded due to inability to obtain outcomes data. This analysis included data from the remaining 38 mothers and their 60 eligible offspring. Full medical records were available for 70% of the pregnancies. The median age of the offspring at the time of data collection was 5.7 years (IQR 3.4–9.2 years). Maternal characteristics and SLE features are shown in Table 1. There were no significant differences in maternal age, race, or level of education for mothers of the children with versus without an SE requirement (data not shown). Among the offspring, there were 56 singleton pregnancies and 2 sets of twins (delivered at 30 and 36 weeks; no SE utilization occurred in the twins). SE utilization was reported in 15 (25%) of the 60 children, with the most prevalent indication being speech delay (requiring speech therapy), which occurred in 12 of these 15 children. When categorized according to the timing of delays, 10 (17%) of the 60 children received SE in the first 2 years of life, and 14 (23%) received SE from age 2 years onward; 9 children had an SE requirement reported for both time periods. The reasons for SE utilization are enumerated in Table 2.
Table 1. Maternal characteristics and SLE features (n = 38 SLE patients)*
Values are the number (percentage) unless otherwise indicated. SLE = systemic lupus erythematosus; IQR = interquartile range; ACR = American College of Rheumatology.
Unit of analysis is the delivery (n = 60), since some women had more than one child included in this study.
Table 2. Reported developmental delays or special educational utilization, stratified by age of occurrence, among 60 children of mothers with systemic lupus erythematosus*
Total (n = 60)
AZA (n = 13)
No AZA (n = 47)
Values are the number (percentage). Four children had more than 1 delay. AZA = azathioprine.
Age <2 years
Fine motor skill deficit
Gross motor skill deficit
Age ≥2 years
Aid with reading
Attention deficit disorder
Maternal/fetal characteristics among those requiring SE are shown in Table 3. Several suspected confounders, including small for gestational age and maternal education level, were not detected in this population to be associated with SE utilization at any age in univariate analyses. However, APS was significantly associated with SE utilization at any age and from age 2 years onward. Among maternal SLE therapeutics during pregnancy, in univariate analyses corticosteroid dose and AZA were each associated with a higher proportion of reported SE utilization among offspring with in utero exposure. When accounting for multiple comparisons, the adjusted P value for AZA was 0.08, whereas the adjusted P value for nonfluorinated corticosteroids was 0.4. The proportions of SE utilization according to in utero AZA exposure are shown in Figure 1. Overall, SE requirements were reported among 7 (54%) of 13 children with in utero AZA exposure versus 8 (17%) of 47 without AZA exposure (univariate odds ratio [OR] 5.69, 95% confidence interval [95% CI] 1.50–21.50; P < 0.01). In the first 2 years of life, 5 (38%) of the 13 children with in utero AZA exposure had an SE requirement compared to 5 (11%) of 47 nonexposed children (P < 0.05); from age 2 years onward, 7 (58%) of 12 children with exposure had an SE requirement compared to 7 (15%) of 47 with no exposure (P < 0.01).
Table 3. Maternal/fetal characteristics and developmental delays/SE needs among 60 children of mothers with SLE*
Values are the number (percentage) unless otherwise indicated. SE = special educational; SLE = systemic lupus erythematosus; IQR = interquartile range; APS = antiphospholipid antibody syndrome; WHO = World Health Organization; NSAIDs = nonsteroidal antiinflammatory drugs.
Age of delay missing for 1 child.
P < 0.10 in comparison to the no delay group.
P < 0.05 in comparison to the no delay group.
Variable included in the propensity score.
P < 0.01 in comparison to the no delay group.
No patients were taking methotrexate or cyclophosphamide.
Fluorinated steroids (e.g., betamethasone, dexamethasone) administered for high risk of preterm labor.
Mycophenolate mofetil contraindicated during pregnancy.
Results from multivariable logistic regression models are shown in Table 4. When adjusting for pregnancy duration, small for gestational age, maternal education level, and maternal APS (model A), in utero AZA exposure was associated with significantly increased SE utilization at any age and for both subgroups (age <2 or ≥2 years), with ORs ranging from 6.1–10.0. When the propensity score was added as a covariate (model B), ORs for AZA were 4.4–6.6, with significance for SE utilization at age ≥2 years and borderline significance for SE utilization at any age or age <2 years (Table 4).
Table 4. Association between azathioprine use and special educational service utilization among children of mothers with SLE*
Results from multivariable logistic regression models. All models adjust for maternal education level, duration of pregnancy, small for gestational age, and antiphospholipid antibody syndrome. SLE = systemic lupus erythematosus; OR = odds ratio; 95% CI = 95% confidence interval.
Propensity score includes lupus nephritis, SLE flare during pregnancy, and nonfluorinated corticosteroid use (dose in mg).
Age <2 years
Age ≥2 years
We performed a sensitivity analysis to account for potential correlation due to multiple births from the same mother, in which we ran a simulation on 1,000 reduced samples that consisted of one randomly selected child per mother, modeling the outcome of SE utilization at any age (see Methods). From this simulation of models including the propensity score, 22% of the 1,000 random samples were significant at an alpha level of 0.1, lending support to conclusions based on models using the full sample size of 60 children.
We performed a further sensitivity analysis restricted to the first-eligible offspring for each mother (n = 38); to reduce the number of variables in this smaller model, we incorporated into the original propensity score the covariates from model A. Results from this sensitivity analysis were consistent with those from our primary analyses, although not reaching statistical significance in this reduced subset of the population (OR 2.5, 95% CI 0.3–21.7 for delay at age <2 years; OR 2.9, 95% CI 0.4–19.4 for delay at age ≥2 years).
In this pilot study, we found an association between maternal AZA therapy during pregnancy and SE utilization, a proxy for developmental delays, in offspring (particularly after age 2 years). This association remained significant when adjusting for recognized risk factors for learning disorders, e.g., prematurity and low birth weight (18, 19). We did not find an increased risk for SE utilization among male offspring, as described in studies of dyslexia and learning delays among offspring of mothers with SLE (5–9, 20, 21).
We controlled for dose of antenatal glucocorticoid exposure, both fluorinated and nonfluorinated. Fluorinated steroids (dexamethasone, betamethasone) cross the placenta and have been linked to cognitive dysfunction in children after in utero exposure, including outside the setting of threatened premature delivery (22). Only 2 offspring in our study were exposed to fluorinated steroids, precluding the ability to focus on this mode of therapy. Nonfluorinated glucocorticoids cross the placenta at a much lower rate: fetal exposure is regulated by placental 11β-hydroxysteroid dehydrogenase, which converts active cortisol into inactive cortisone (13). Most likely, ∼10% of nonfluorinated glucocorticoid crosses into the fetal circulation at doses of ≤20 mg (13, 23). Whereas in univariate analyses the dose of nonfluorinated steroids was associated with an SE requirement, this was no longer significant when adjusting for multiple comparisons, nor in multivariable modeling (OR 1.04, 95% CI 0.95–1.14).
Our study population included women with a wide range of lupus activity, a significant number of whom required immunosuppressive, antimalarial, and/or glucocorticoid therapy during their pregnancies. In the majority of cases, immunosuppressive therapy during pregnancy was with AZA (n = 13). Two subjects had exposure to MMF, one throughout the first and second trimesters (prior to Food and Drug Administration [FDA] category D prescribing a warning for MMF) and one at conception, at which time MMF was stopped. No other maternal characteristic, with the exception of APS, predicted an SE requirement in children.
AZA has been used in solid organ transplantation for more than 50 years and is used frequently for therapy of organ-threatening autoimmune diseases. For SLE, AZA is considered both a steroid-sparing agent and a “maintenance drug” for use after initial disease control is achieved with cyclophosphamide (24). Although AZA is considered relatively safe during pregnancy, it remains listed as a category D drug by the FDA, indicating that potential benefits may warrant its use in pregnant women despite potential risks. Studies of AZA and pregnancy outcomes from diverse patient populations (solid organ transplant, rheumatic disease, and inflammatory bowel disease) have revealed inconsistent results: there are case reports of malformations occurring in women treated with AZA, as well as rare reports of fetal immunologic abnormalities, including newborn hypogammaglobulinemia and pancytopenias, most of which normalize by 10 weeks (25–29). Some studies have reported increased rates of spontaneous abortions, prematurity, intrauterine fetal growth retardation, and low birth weight (27–29), but these studies often include patient populations with heterogeneous and sometimes poorly controlled underlying diseases. Therefore, findings may be confounded due to heightened disease activity or to concomitant medications required during pregnancy. Other studies in inflammatory bowel disease and transplant populations have found no association between maternal AZA use and poor perinatal outcomes (30–33). In experimental animals, however, 6-mercaptopurine (6-MP), an AZA metabolite, has been found to be teratogenic at doses similar to or greater than the therapeutic doses used in humans (34).
AZA and 6-MP inhibit synthesis of DNA and RNA precursors adenine and guanine, thereby exerting immunosuppressive and antiinflammatory effects by stopping proliferation of rapidly dividing immune cells (34). As administered, AZA is inactive, requiring nonenzymatic and enzymatic intracellular metabolism to its active metabolites, primarily 6-MP, and other inactive metabolites. 6-MP in turn is metabolized to active metabolites 6-thioguanine (6-TG) and 6-methylmercaptopurine (6-MMP). 6-TG has been demonstrated in fetal red blood cells (RBCs) at slightly lower concentrations than in the RBCs of their mothers, who were treated with AZA throughout their pregnancies for Crohn's disease (35). However, the investigators in that study did not detect 6-MMP in fetal circulation, nor were any teratogenic effects noted. The authors concluded that the presence of measurable levels of one pharmacologically active thiopurine metabolite (6-TG) but not another (6-MMP) may indicate that the placenta forms a “relative barrier” to AZA and its metabolites, and recommended that a maternal level of 6-thioguaninenucleotide (6-TGN) be obtained during pregnancy in women treated with AZA to assure that the fetus is not exposed to high levels of this metabolite. However, in their series, the highest maternal 6-TGN level observed was 291 pmoles/8 × 108 RBCs, corresponding to 6-TGN levels in the artery and vein of the umbilical cord of 65 pmoles/8 × 108 RBCs and 93 pmoles/8 × 108 RBCs, respectively, all of which are reassuringly lower than the current recommended therapeutic levels for AZA treatment of 235–400 pmoles/8 × 108 RBCs (36, 37).
The possibility that thiopurine metabolites exert adverse effects on the developing fetus cannot be excluded. AZA has not been definitively linked to birth anomalies or other complications of pregnancy, and understanding of the intrauterine effects of AZA and its metabolites is clearly incomplete. Therefore, our findings of significantly higher rates of SE utilization in children with in utero AZA exposure compared to those without exposure must be interpreted in the context of this uncertainty. It is possible that AZA metabolites may have some effect on fetal nervous system development, manifesting in early childhood as impairment in speech or hearing or delayed learning. Alternatively, AZA may act as a marker of a disease-related phenomenon.
After controlling for several potential confounders, the association of AZA exposure and SE requirement in offspring remained significant. While we cannot exclude the possibility of confounding by indication, we attempted to account for underlying disease severity by utilizing propensity scoring, and results from our models adjusting for the propensity score support an independent association between AZA exposure and SE requirement in these children.
Several methodologic considerations in our study are noteworthy. First, the retrospective nature of the data acquisition regarding childhood SE requirement and developmental delays may have resulted in recall bias. However, because all of the mothers in this study had chronic disease and were unaware of the specific hypotheses under investigation, we do not expect recall bias to have been different based on exposure to a specific therapeutic agent. Although a number of maternal and fetal characteristics were examined in a univariate fashion, we did not formally adjust for multiple comparisons, since this was a pilot study intended to inform the design of a larger, prospective study. It will be important to replicate the results related to AZA exposure in an independent population. Our small sample size limited the ability to account for correlations between offspring from the same mother; therefore, separate models that randomly selected for one child from each mother were constructed, albeit further reducing the sample size. A larger sample size would be necessary to investigate critical windows of susceptibility to AZA exposure during gestation that may be relevant to long-term developmental outcomes. There was also a large degree of overlap among different categorizations of delays, i.e., many children with a delay or special need in their first 2 years of life also had attention deficit disorder or an educational need later on, and were included in both categories. Therefore, significance across multiple categorizations for a given characteristic is expected. Another limitation was the inability to perform formal standardized testing in the full study cohort using validated instruments, which would more fully characterize the developmental spectrum and result in less susceptibility to referral biases that may exist for reasons such as provider practice or geographic location. For example, preliminary data from the UK indicated only 2 cases of learning difficulties among 132 offspring of mothers with SLE (38), in contrast to the larger proportion in our study (15 of 60 children) requiring SE services for learning or developmental delays. Prospective studies with standardized testing would enable comparisons between study populations such as children born in the UK versus the US. Finally, due to the nonrandomized nature of this study, it is not feasible to completely exclude the possibility that disease characteristics or severity may underlie the observed association between maternal AZA use and SE requirement in offspring, despite our attempts to account for the possibility of confounding by indication. For example, we were unable to assess longitudinal patterns of disease activity measures during the course of pregnancy, and use of propensity scoring is an imperfect way to disentangle the effects of underlying disease and its treatment.
Strengths of this study include access to a well-characterized cohort of patients who were motivated and interested in participating in research. Since all patients were followed at the University of Michigan, laboratory assays were standardized. The multidisciplinary research team included rheumatologists and a high-risk obstetrician who brought in-depth knowledge of complicated pregnancies and perinatal risks to the review of each case. In addition, the proxy variable for developmental delays, i.e., maternal report of SE utilization, represented a clinically relevant end point that was apparent to a provider and prompted intervention.
The association detected in this pilot study between SE needs and in utero exposure to AZA warrants further prospective study of developmental delays. We wish to emphasize that immunosuppressive therapy should not be withheld in lupus pregnancies when indicated for treatment of active disease, since highly active lupus during pregnancy is associated with poor fetal outcomes, including increase in premature birth rates and decrease in live births (39, 40). Our findings should alert pediatric providers to consider the need for early developmental screening of children born to mothers with SLE. Early identification of developmental delays, even during infancy, has been shown to increase long-term functioning of affected children with appropriate intervention and treatment (41–43). With increasing numbers of women with lupus achieving successful pregnancies, increased vigilance and understanding of long-term outcomes among their children will be important for patients and providers alike.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Somers 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 conception and design. Marder, Gordon, McCune, Somers.
Acquisition of data. Marder, Romero, Hyzy, Gordon.
Analysis and interpretation of data. Marder, Ganser, Romero, Gordon, McCune, Somers.