In Utero Antidepressants and Neurodevelopmental Outcomes in Kindergarteners

We used deidentified administrative data from Manitoba, Canada (population of ∼1.3 million). Data were accessed from the Manitoba Population Research Data Repository, housed at the Manitoba Centre for Health Policy (MCHP). The repository includes 99% of Manitobans and contains individual-level data on all contacts with the health care system.³¹  Data are linkable across domains (ie, health, social, education) and within families, allowing for robust epidemiological research. Linkages are completed by using scrambled deidentified personal health information numbers. These data have been widely used for health services research.^32–40 

Medication use was obtained by using data from the Drug Program Information Network, which contains information on all prescription medications dispensed in Manitoba. Health data were accessed by using hospital discharge abstracts and physician claims for services provided to give counts and diagnostic information for outpatient visits. Hospital discharge abstracts are coded by using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) (before April 1, 2004)⁴¹  and International Classification of Diseases, 10th Revision, with Canadian Enhancement (ICD-10-CA) (after April 1, 2004),⁴²  and physician claims are coded by using the ICD-9-CM.

Kindergarten assessment data were obtained from the provincial department of education.

This study included all mother-child dyads for mothers with a live hospital birth between April 1, 1996, and March 31, 2014 (with follow-up until 2015), in Manitoba. Dyads were excluded for the following factors: there was no continuous health coverage of mothers from 90 days before conception to birth; there was no linkage between the mother and child, or the child was linked to >1 mother; the infant was stillborn or a multiple; the infant did not have a recorded gestational age; the mother did not have at least 1 diagnosis of a mood or anxiety disorder from 90 days before conception, had only 1 filled SSRI or SNRI prescription, had a prescription for an antidepressant other than SSRIs or SNRIs during pregnancy, or was exposed to antipsychotics, benzodiazepines, or opioids during pregnancy. From eligible mother-child dyads, those with children with no recorded EDI were removed (Fig 1: cohort formation). The final cohort was subdivided for each EDI outcome on the basis of the availability of complete EDI records.

Mothers were considered to have a prenatal mood or anxiety disorder if in the 90 days before conception they had one of the following: 2 or more physician visits with a diagnosis for anxiety disorders; 1 or more physician visits with a diagnosis for depressive disorder, affective psychoses, or adjustment reaction; 1 or more hospitalizations with a diagnosis for anxiety disorders, anxiety states, phobic disorders, or obsessive-compulsive disorders; or 1 or more hospitalizations with a diagnosis for depressive disorder, affective psychoses, neurotic depression, or adjustment reaction. Specific ICD-9-CM and ICD-10-CA codes are listed in Supplemental Table 4.

Dyads that include a mother with at least 2 prescriptions filled for SSRIs or SNRIs from the time of conception to birth were included in the prenatal SSRI or SNRI exposed group. The unexposed group included dyads with mothers who had no recorded prescriptions for antidepressants during pregnancy.

We used the EDI to measure the impact of in utero antidepressant medications on developmental outcomes among children whose mother had a diagnosis of mood or anxiety disorder. The EDI is a 104-component kindergarten teacher-completed questionnaire that measures developmental health in 5 domains: physical health and well-being, language and cognitive development, social competence, emotional maturity, and communication skills and general knowledge.⁴³  Children are considered developmentally vulnerable in a domain if they score in the bottom 10th percentile on the basis of national norms. Children with completed assessments accessible in the repository were included in the outcome analysis. The EDI is administered every 2 years to all students enrolled in the public school system in the second half of kindergarten (mean age of 5.7 years). The first province-wide administration of the EDI was in the 2005–2006 school year. The validity and reliability of the EDI has been reported widely, and it is used globally as a population-level indicator of developmental health.^44–48 

Women diagnosed with a mood or anxiety disorder who continue their medications during pregnancy may be different in health characteristics and disease severity than women who discontinue antidepressants during pregnancy. These differences may contribute to variation in outcomes seen in children versus the effect of the medications in utero. To make the 2 study groups more comparable, we used HDPS adjustments to balance differences between groups. In studies, authors have found that adjustments using HDPS have resulted in effect estimates closer to randomized control trials when compared to standard covariate adjustments.^49,50  For each of the educational outcomes, we ran HDPS adjustments using a multistep algorithm that identifies potential confounders from the database by selecting variables correlated to both the exposure and the outcome, prioritizing these covariates by prevalence and potential for bias, and integrating them into a logistic model from which propensity scores are derived.

Covariates were drawn from data 365 days before conception, which included (1) medical service tariff codes, (2) physician diagnostic codes, (3) hospital procedure codes, (4) hospital diagnostic codes, and (5) prescription medication claims. Using information from these data dimensions improves control for confounding by using numerous proxy covariates that describe the health status of women in this study, including comorbid conditions, concurrent medication use, and disease severity.

Inverse probability of treatment weights (IPTW) was applied to the data to adjust for the top 500 covariates identified in the HDPS estimation risk estimates. Using IPTW will provide a true average treatment effect of the use of SSRIs or SNRIs among depressed women. Using another method, such as matching, would produce an estimate of the treatment effect among the treated, which does not address the potential benefit or harm of expanded use of SSRIs or SNRIs in this population. The cohort was trimmed to include only those dyads with a propensity score >0.2 and less the propensity score of the 80th percentile among the treated group, to establish comparability between exposed and unexposed groups.⁵¹  Specifically, to ensure that women from the low-propensity strata, which are primarily composed of patients for whom most physicians would regard treatment with antidepressants as inappropriate, are removed from the analyses. We calculated standardized differences for the covariates used in the HDPS model to investigate if the adjustment reduced the difference between groups. In Fig 2, we illustrate the standardized differences before and after inverse probability of treatment weighting between women unexposed and exposed to antidepressants during pregnancy and the example outcome of EDI not ready in 2 more domains. Standardized differences <0.10 were not considered clinically meaningful.

Logistic regression analyses were used for all outcome models. Prematurity, small for gestational age, and inadequate prenatal care were included as covariates in the outcome model.

Our cohort included 266 479 women with a live birth between 1996 and 2014. Among these women, mothers with only 1 SSRI or SNRI prescription during pregnancy, those who were treated with an antidepressant other than an SSRI or SNRI (n = 1247), and those who did not have 1 full year of health coverage before conception date (n = 499) were excluded. Among our study cohort, 12 072 women had a diagnosis of a mood or anxiety disorder before conception.

The exposed group, mothers who had 2 or more SSRI or SNRI prescription dispensations during pregnancy, included 2055 women; the unexposed group, those who did not have an SSRI or SNRI prescription during pregnancy included 10 017 women (see Fig 1). Among this cohort, 3050 children had an EDI record.

Mothers and children in the exposure groups had different levels of baseline characteristics (see Table 1). By using HDPS, each characteristic was weighted for both groups, and standardized differences were adjusted to <0.10 with a few exceptions (see Table 2, Supplemental Tables 5 through 10). The variables included in the HDPS differed for each model because they were selected on the basis of correlations to the exposure and each different educational outcome. Therefore, mothers received different weights for each outcome, resulting in a slightly different sample size for each of the outcomes presented. After trimming of each sample, exposed and unexposed dyads did not differ significantly on observed health and social covariates (see Supplemental Tables 5 through 10 for weighted and unweighted standardized differences of the top covariates in each outcome model).

Compared with children in the unexposed group, children whose mothers were dispensed at least 2 prescriptions for an SSRI or SNRI during pregnancy were at an increased risk for having developmental vulnerability in 2 or more domains measured by EDI (adjusted odds ratio = 1.43; 95% confidence interval 1.08–1.90; Table 3). Exposure to at least 2 SSRIs or SNRIs in utero was also significantly associated with being vulnerable in the domain of language and cognitive development (adjusted odds ratio = 1.40; 95% confidence interval 1.03–1.91; Table 3). Prenatal use of 2 or more SSRIs or SNRIs were not significantly associated with developmental vulnerability in children in the following domains: physical well-being, communication and/or general knowledge, social competence, and emotional maturity, as well as 1 or more domains (Table 3).

We found 21.43% of children exposed in utero to SSRIs or SNRIs were at increased risk of being developmentally vulnerable in 2 or more domains (versus 16.16% of unexposed children), which is an absolute risk difference of 5.27%. This means for every 19 women who took an antidepressant during pregnancy in our study, 1 more child was at risk for vulnerability in 2 or more domains on the EDI. Data collection across Canada reveals that 25% or more kindergarten children are vulnerable in at least 1 area of development.⁵²  The percentage of children who were vulnerable in 2 or more domains is a more sensitive measure of developmental vulnerability. We also found that ∼13% of children in the exposed group were vulnerable in the language and cognitive domain compared with 12% of unexposed children (absolute risk difference of 0.54%). This means, on average, 185 women would have to take an SSRI or SNRI during pregnancy for 1 more child to be vulnerable in language and cognition.

SSRI exposure on associations found in this study has biological plausibility because animal studies support that atypical serotonergic signaling resulting from prenatal SSRI exposure may alter fetal brain development.^19,53,54  Moreover, authors of a recent human study using MRI found prenatal SSRI exposure has a significant association with fetal brain development.⁵⁵ 

Our findings are consistent with Johnson et al³⁰  (2016) who conducted an observational longitudinal study in a US clinic setting and found a significant relationship between language skills and cognitive functions with prenatal SSRI exposure. However, our findings are inconsistent with results from adjusted analyses from Kragholm et al²⁹  (2018) which found no association between in utero exposure to SSRIs and delayed elementary school entry and special school needs in children.

Our results add to an emerging body of literature that reveals an association of negative academic outcomes in early childhood and language difficulties for children whose mothers have a clinical indication of a mood or anxiety disorder during pregnancy who have been treated with antidepressants. Additional research using population-based samples and robust methodology are required to triangulate findings before definitive clinical recommendations can be made. Results of our study are most generalizable to women with a diagnosis of a mood or anxiety disorder before pregnancy because this is the population in which the clinical decision to continue medications during pregnancy is most relevant. Women must weigh the risks of untreated depression with their physicians before making a decision to stop taking antidepressants during pregnancy.

Our findings point to the need for early intervention and supports for children of mothers who have depression and/or anxiety disorders during pregnancy. Developmental vulnerability as measured by the EDI is considered an important measure that is associated with later positive educational outcomes,^56,57  and EDI results are a strong predictor of future academic success.⁵⁸  Brinkman et al⁵⁹  report that children with scores in the bottom 10th percentile of the EDI later test just as poorly on national standardized tests of literacy and numeracy in grades 3, 5, and 7, with the strongest correlations found within the language and cognition domain. Our findings point to the increased vulnerability in the area of language and cognition among children exposed to in utero maternal depression, highlighting the risk these children have for negative educational outcomes in the future.

This study has several strengths. First, the use of the Manitoba Population Research Data Repository provides a population-based cohort, minimizes attrition and detection bias, and allows inclusion of numerous covariates over many years. These longitudinal data are ideal for investigating outcomes during early childhood. Moreover, the use of standardized Canadian education data and a recognized global measure of educational attainment strengthens the generalizability of results. We used sophisticated design and analysis to address a frequent critique of investigations examining outcomes of children exposed to in utero antidepressant medication, namely, the ability to determine if results are confounded by the underlying maternal depression that resulted in the exposure to medication. Our use of a restricted study population of all women diagnosed with a mood or anxiety disorder decreases this confounding by indication and takes account of the difference in postnatal interactions of a mother who may be depressed versus a mother who is not depressed. The use of HDPS adjustments to balance baseline covariates adjusts for disease severity and important maternal comorbidity. The HDPS algorithm accounts for all physician and hospital visits (number of visits to physician or hospital, including number of previous suicide attempts), diagnoses (all diagnoses of mood or anxiety disorders, all diagnosis of other mental disorders, or comorbid physical disorders), and procedures, as well as all prescription medications (all medications with and without psychiatric indication) dispensed to our study population. This powerful algorithm adjusts for all measured confounders present in these data; adjustments using HDPS result in effect estimates closer to randomized control trial findings when compared to standard covariate adjustments. Moreover, HDPS has also been shown to adjust for hidden confounders,⁶⁰  supporting that this method can adjust for some unmeasured confounders as well, strengthening the robustness of our analyses.

Despite the many strengths, there are important limitations. First, there is the potential for misclassification of exposure if prescriptions were filled but not used, creating a bias toward the null. To decrease this error, we included women who had 2 or more dispensations throughout their pregnancy. Furthermore, despite using sophisticated methods to control for confounding, we cannot account for all unmeasured or residual confounding, including disease severity. It is also important to note that although the EDI is administered across the entire public school system in Manitoba, children enrolled in some schools operated by indigenous communities and private schools are not included. Furthermore, the EDI is a community-level indicator and may lack the specific diagnostic ability to identify individual neurocognitive deficits in individual children but can identify population-level trends. Also, there may be bias in the assessment of educational outcomes by teachers; however, it is impossible to discern if bias would systematically affect our exposed versus unexposed group because teachers were not aware of exposure status. Hence, any error resulting from this bias would be random. Finally, given the multiple tests performed, chance finding cannot be ruled out. Replication of these findings in other large population-based cohorts should be conducted to help guide clinical decision-making. Studies looking at the variation in results of different types of antidepressants, dose-response effect, and gestational timing of medication on educational outcomes are warranted.

Children of mothers diagnosed with a mood or anxiety disorder who used SSRIs or SNRIs during pregnancy were at risk for developmental vulnerability and for language and cognitive difficulties. Early interventions should be provided to children exposed to maternal depression during pregnancy in kindergarten to help ameliorate later educational challenges.

We are indebted to Healthy Child Manitoba, Manitoba Education and Training, and Manitoba Families for the provision of data. We acknowledge the MCHP for use of data contained in the Manitoba Population Research Data Repository under project 2015/16-35 registered with the Manitoba Health Information Privacy Committee. The results and conclusions are those of the authors, and no official endorsement by the MCHP, Manitoba Health, Seniors and Active Living, or other data providers is intended or should be inferred.

 
• EDI

  Early Development Instrument

 
• HDPS

  high-dimensional propensity score

 
• ICD-9-CM

  International Classification of Diseases, Ninth Revision, Clinical Modification

 
• ICD-10-CA

  International Classification of Diseases, 10th Revision, with Canadian Enhancement

 
• IPTW

  inverse probability of treatment weights

 
• MCHP

  Manitoba Centre for Health Policy

 
• SNRI

  selective serotonin norepinephrine inhibitor

 
• SSRI

  selective serotonin reuptake inhibitor