Prenatal Antidepressant Exposure and Child Motor Development: A Meta-analysis

To address these issues, a systematic review of the literature was performed by using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. After this, a meta-analysis of findings was undertaken. To ensure a comprehensive search of the literature, a computerized search of multiple electronic databases was conducted. These databases included PsychINFO, Embase, Medline, PubMed, and Scopus. Each database was searched from its start date to July 24, 2017. Keywords and their combinations used were as follows: “antidepressants or selective serotonin reuptake inhibitors or SSRI/SRI/SNRI,” “pregnant/pregnancy,” “infant/child,” and “(motor/psychomotor/neuromotor) development” (Supplemental Information). Reference lists, including those of relevant review articles and meta-analyses, were searched for additional studies not already identified in the initial database search.

Researchers’ studies were included in the systematic review if they (1) reported primary data, (2) were published in the English language, (3) were studies involving humans, (4) used and described an accepted measure of motor performance, and (5) reported on infant or child motor development outcomes after any antidepressant exposure in pregnancy. Studies were excluded if they were individual case studies, conference abstracts, or proceedings without a related publication; reviews or meta-analyses of the literature; or were solely focused on motor outcomes in the neonatal period.

For the purposes of the subsequent meta-analysis, additional inclusion criteria were as follows: (1) included a control group of women who were not exposed to antidepressant medication during pregnancy and (2) provided sufficient raw data to perform effect size (ES) calculations.

It became apparent from our systematic review of the literature that only a small number of studies included a comparison group of infants of women with depression who did not receive antidepressant medication during pregnancy either as the sole control group or as an additional comparison group along with healthy controls. Therefore, in the main meta-analysis, we only included studies with data available for a healthy control group of women who were not exposed to antidepressant medication to ensure uniformity with the majority of available studies.

Titles and abstracts were examined for all studies identified by using the keyword search. When the inclusion criteria were met, the full-text article was retrieved. A second researcher reviewed those articles that were identified for inclusion as well as those for which ambiguity existed about their inclusion or exclusion from the systematic review. Information extracted from all the studies (when available) included the authors, year of publication, study design, sample size, demographic and clinical characteristics of participants, details of antidepressant exposure (type, dose, and timing of exposure during pregnancy), details pertaining to the timing, method of and age at motor assessment, and primary and secondary outcomes.

For the meta-analysis, we further extracted from the published studies the raw data to calculate the ES for each study. When repeated measures of motor assessment occurred in the same study population, and/or when subdomains of the motor assessment were reported, we preferentially extracted summary measures of motor performance at the latest available time point of assessment. When fine and gross motor data were reported separately within the same study, in the absence of the availability of a summary measure, the fine motor domain was used. In the absence of the necessary data being available in the published study, we contacted authors requesting the appropriate raw data.

To enable variability in study results to be critically evaluated, the methodological quality of the included nonrandomized studies was assessed by using the Newcastle-Ottawa Quality Assessment Scale (NOQAS).³⁷ Each study was allocated a star rating in multiple categories: (1) “selection” of cases or cohort, (2) “comparability” of cases and controls (or cohorts), and (3) “exposure” (for case control studies) or “outcome” (for cohort studies). Total scores were assigned to each study for a maximum possible rating of 9 stars. Quality ratings were performed before conducting the analyses. Irrespective of the NOQAS score, we included all studies that met inclusion criteria in the main analysis.

Data were analyzed by using Stata (Stata Corp, College Station, TX) and examined for both fixed and random effect models. The metan package and metareg were used in all analyses. We divided studies into subgroups. Those in which researchers used a standardized measure of motor development, such as the Bayley Scales of Infant Development (BSID), were compared with studies in which researchers drew on population data and made use of parent self-report or clinician report of motor abilities. This distinction is referred to as “data type” in the meta-regression reported below. Publication bias was assessed through the use of a funnel plot.

We identified 320 articles using the described literature search methodology with an additional 9 published studies obtained by review of reference lists and consultation with experts in the field. After duplicate removal, 160 unique articles were identified, of which 126 were excluded after extensive title and abstract review. In total, 34 full-text articles were assessed for eligibility. Ten articles were subsequently excluded (Fig 1). Of the excluded articles, 3 were excluded because researchers either did not use an accepted measure of motor performance^38,39 or did not clearly report on the procedures for the assessment of motor development.⁴⁰ Researchers in 4 of the excluded studies reported on other outcomes of interest,^41,–44 those in 1 reported only neonatal outcomes,⁴⁵ and those in 2 provided insufficient data,^18,46 leaving 24 articles for inclusion in the qualitative synthesis (Table 1).

For the meta-analysis, an additional 6 studies were excluded. Of these, 3 were excluded because they did not contain a control group without antidepressant exposure,^25,53,65 1 did not have a healthy control group for comparison,²⁰ and researchers in 1 used the same cohort of children that was later published as a follow-up study.⁵⁴ We excluded the initial article in keeping with our prescribed methodology to use only the results from the latest available time point of assessment. The final study was excluded after consultation with all the authors because there was no summary measure available that was suitable for inclusion in the meta-analysis.⁶⁴ 

Overall, we pooled results from 18 studies. The ES for poorer motor outcomes after in utero antidepressant exposure was 0.22 (95% confidence interval [CI] = 0.07 to 0.37; Fig 2). On further subanalysis determined a priori, researchers in 7 studies reported outcomes of motor assessment with categorical data. The pooled ES from these studies was significant (ES = 0.40; 95% CI = 0.18 to 0.62). The 11 studies in which researchers report continuous data resulted in a nonsignificant ES of 0.08 (95% CI = −0.11 to 0.26).

Heterogeneity measures were significant, both for the total pooled analysis (I² = 56.6; P = .002) and the analysis of the categorically reported data of motor assessment outcomes (I² = 67.4; P = .005; Fig 2). The studies reporting motor assessment outcomes with continuous rather than categorical data for motor development demonstrated an I² value of 34.2 (P = .125). Heterogeneity between these 2 subgroups was also significant (P = .018). Meta-regression analyses of potential predictors of heterogeneity, including the NOQAS, year of publication, and type of data obtained from motor assessment (ie, categorical or continuous), were performed. Only the type of data was a significant predictor of heterogeneity (P = .033; Table 2).

A funnel plot was produced to examine publication bias. This revealed no indication of publication bias and is presented in Fig 3.

Our objective in this systematic review and meta-analysis was to investigate the possible relationship between antidepressant exposure in utero and subsequent motor development outcomes in infants and children. The total pooled result of our meta-analysis reveals a small effect (0.22) in the direction of an increased risk of poorer motor scores and outcomes in children who are exposed to antidepressant medications during pregnancy. However, the overall study-to-study variation is moderate (I² = 56.6%); therefore, these initial findings require further qualification.

With the exception of de Vries et al,⁵⁸ researchers in the 6 other studies reported categorical data for motor outcomes and used indirect observation of assessment measures by parents or clinicians as reported through structured interviews or through database or chart review^{21,49,52,61,–63}. Because of the nature of these studies, blinding of assessor was mostly not appropriate or possible. In contrast, researchers in the 11 studies in which continuous data were reported primarily used standardized measures of motor assessment.^{19,24,26,48,50,51,55,–57,59,60} All were determined by direct observations from a limited number of trained assessors as part of the research team, the majority of whom were blinded to exposure status. When these 2 groups are analyzed separately, only the studies in which researchers report categorical data from clinician and/or parent ratings continue to exhibit a negative effect on motor outcomes (ES = 0.40; 95% CI = 0.18 to 0.62). Those reporting continuous data from standardized testing do not (ES = 0.08; 95% CI = −0.11 to 0.26). Furthermore, measures of heterogeneity are substantially less in the latter group (I² = 67.4% versus I² = 34.2%), with our subsequent meta-regression tests confirming that only the type of measure (ie, standardized testing versus clinician and/or parent rating) is a significant predictor of heterogeneity.

The methodological limitations of the individually included studies help to explain the moderate-to-large degree of heterogeneity observed in our analysis. In the included studies alone, there are 13 different methods used to quantify motor performance. Only Galbally et al²⁶ used a standardized and specific neuropsychological measure of motor development, the Movement Assessment Battery for Children, which has been established to have high test-retest reliability and moderate construct validity as a motor measure.⁶⁶ In comparison, the BSID was most commonly used among the included the studies and is widely considered the reference standard for the measurement of infant and toddler development, but its motor subscale has been less clearly validated against other measures of motor assessment.⁶⁷ Other methods used to assess motor outcomes included screening measures with motor components but not specific subscales (eg, BOEL [blicken orienterar efter ljudet] test), clinical assessment (eg, general practitioner or physician), maternal reports, and combinations of the aforementioned. Aside from the difficulty this creates in comparing results among studies, some of the clear limitations of these measures include their lack of validity as a measure of motor performance and lack of blinding of the assessors to knowledge about in utero pregnancy exposure to antidepressant medications.

The timing and duration of the antidepressant exposure during pregnancy also differs greatly across studies. It is possible that the cumulative duration of exposure or the timing of exposure in combination with the fetus’s stage of central nervous system development may explain some of the differences across studies, but there is insufficient evidence from the existing research to draw such conclusions. Like most studies in which researchers investigate the potential for adverse outcomes secondary to medication use, limitations pertaining to the establishment of medication compliance and associated confirmation of exposure are present. Methods such as self-report are susceptible to recall bias, and pharmacy dispensation records do not guarantee that a medication has actually been taken.

Confounding by indication and severity are particularly problematic in the original studies. Researchers in only 5 of the 24 studies included in the systematic review attempted to address this by including a second control group of women who had depression but were unmedicated during pregnancy.^{20,52,55,60,63} Researchers in 1 additional study designated a comparison group of prepregnancy SSRI users without continued use during pregnancy.⁶¹ The incorporation of such control groups within well-designed studies is important for interpretation purposes. Given that depression in the mother has been associated with a multitude of adverse outcomes in the child, including infant cognitive and motor development issues,¹⁰ developmental delay,⁷ lower adolescent IQ, and increased likelihood of emotional and disruptive behavior disorders,⁸ untangling the effects of depression versus those of antidepressant exposure is a necessity. Furthermore, for those studies in which researchers rely on maternal report of child motor problems, the impact of maternal depression on ratings of the child requires consideration in design and interpretation of studies.

Limitations in the design of our meta-analysis include the use of data for fine motor outcomes over gross motor outcomes when only the total scores for each were reported in the published studies. Only 3^24,59,61 of the included studies in the meta-analysis are impacted by this. Further post hoc analyses could be performed to ensure that these alternative data do not change the overall result of our study. If anything, we would expect to have underestimated the effect of in utero antidepressant exposure on the impairment of motor development. Hanley et al⁵⁹ identified statistically significant poorer gross but not fine motor scores. Researchers in the other 2 studies found similar nonsignificant results for both fine and gross motor assessments. The use of these gross motor scores would therefore be unlikely to influence our findings. In addition, as with any review of the literature, it is possible that a number of studies may have been missed, including those published in non–English-speaking journals or in databases not included in the electronic search.

Although there were identified studies in which researchers found differences in motor development in exposed children, the clinical relevance of these findings has been questioned. Despite the existence of statistical differences or small-to-moderate ESs, in at least 5 of the studies included in our systematic review and meta-analysis, the researchers note that raw scores still mostly fell within normal developmental ranges or differences were not discernable on clinical examination.^{25,52,54,60,61} Therefore, even with the results of our meta-analysis, the clinical significance of our finding remains somewhat unclear. It is possible that our findings predominantly reflect differences in mean scores within the normal developmental range rather than clinically significant impairments in functioning.

An interesting subset of studies that have been identified through our systematic review are those in which researchers have followed up on the development of children who exhibited symptoms of neonatal abstinence syndrome (NAS) at birth after antidepressant exposure in pregnancy. A number of the symptoms characterizing this syndrome are of a motor nature, including tremors, jitteriness, and poor muscle tone. There is considerable disagreement about whether these symptoms represent an acute withdrawal syndrome or are the result of structural changes that have occurred to the fetal brain secondary to the in utero antidepressant exposure. It follows then that neonates who suffer from this syndrome might be at greater risk of impaired motor development in childhood. At present, researchers in 4 studies included in our systematic review have examined motor outcomes in these children, with varying results.^26,50,53,65 Two of these studies did not meet inclusion criteria for our meta-analysis,^53,65 and thus quantifying these results is difficult. Future work in which researchers investigate the long-term motor as well as other developmental outcomes in children exhibiting NAS at birth would be clinically useful. If such a relationship exists, this would provide clarity about which children are at risk and allow for appropriate monitoring of and follow-up in this population.

Researchers in most of the included studies performed outcome assessments at a young age. Given that researchers in only 3 studies^26,63,64 investigated motor development in children solely after 3 years of age, this remains a significant gap in the literature. It is likely that at the moment, any abnormalities of motor development occurring in later years related to in utero antidepressant exposure would be overlooked. Furthermore, the assessment of motor capacity in young infants is imprecise, and measures such as BSID have limited predictive validity for later motor outcomes. Of course, research to date has been limited by the relatively recent increase in the prescription of SSRIs during pregnancy^2,3 and the subsequent availability of these populations for long-term follow-up of developmental outcomes in children. With a number of large population-based cohorts now being established,^52,61,62 researchers will be better positioned to draw conclusions about longer-term developmental outcomes going forward. Until such results are available, the results of our meta-analysis reveal that general practitioners and pediatricians involved in the care of children who are exposed to antidepressant medication during pregnancy might routinely ask about progress toward the achievement of motor milestones with appropriate follow-up if parents report concerns.

In this systematic review and meta-analysis, we provide a timely synthesis of the available literature on the relationship between antidepressant exposure in utero and subsequent motor development in infants and children. Our finding of a small association between antidepressant exposure in pregnancy and poorer motor outcomes in children is limited by the vast differences in methodological design in the included studies. This clearly highlights the need for future researchers to design larger, methodologically rigorous studies that include a pregnant, unmedicated control group with depression (when ethically and clinically appropriate) in addition to a standard, unmedicated control group without depression; in which they accurately and prospectively record the timing and duration of antidepressant exposure in pregnancy; and in which they manage offspring longitudinally into older childhood or adolescence using robustly administered neuropsychological measures of motor development. In view of increasing prescription rates for antidepressants during pregnancy, the lack of such studies, particularly when trends toward poorer motor outcomes observed in this meta-analysis exist, is concerning. Currently, these findings are unlikely to warrant alterations to the current antidepressant prescribing guidelines during pregnancy. However, there may be some benefit in the monitoring of at-risk children. Further research will better enable clinicians and patients to make informed, evidence-based decisions on this matter.

BSID

Bayley Scales of Infant Development

CI

confidence interval

ES

effect size

NAS

neonatal abstinence syndrome

NOQAS

Newcastle-Ottawa Quality Assessment Scale

SSRI

selective serotonin reuptake inhibitor

We thank the library staff at Fiona Stanley Hospital and Graylands Hospital for their assistance.