Association of Maternal Use of Benzodiazepines and Z-Hypnotics During Pregnancy With Motor and Communication Skills and Attention-Deficit/Hyperactivity Disorder Symptoms in Preschoolers

Key Points Question Is the association of prenatal benzodiazepine/z-hypnotic exposure with child developmental risks different according to timing of exposure, duration, or coexposure to opioids or antidepressants? Findings Among 41 146 pregnancy-child dyads in this cohort study, a moderate association between benzodiazepine/z-hypnotic exposure in late pregnancy and greater gross motor and communication deficits in children born to women with depressive/anxiety disorders were observed, but not to the extent that the impairment was of clinical relevance. There was no evidence for duration or coexposure associations on all outcomes. Meaning These findings show no clinically relevant detrimental risk of prenatal benzodiazepine/z-hypnotic exposure on motor, communication, and attention-deficit/hyperactivity disorder outcomes in preschoolers.

inattention and two questions entailed symptoms of hyperactivity -impulsivity. The five response options range from "never=1" to "very often=5". We dichotomized the unweighted response options (≤ 2=0, > 3=1) and summed them across the six questions, as indicated in Kessler et al. 6 We then categorized the final ASRS score into three clinical categories (no ADHD symptoms=score 0-1, mild symptoms=score 2-3, moderate to severe symptoms=score 4-6).
Parenting practices were measured via six of the 42-item Alabama Parenting Questionnaire (APQ) 7 in MoBa Q7, i.e. when the child was 5-year old. These selected APQ items measured the "Positive Involvement with children" dimension, which is relevant to the etiology and treatment of child conduct problems. 8 Mothers were asked to indicate on a scale from never=1 to always=5, how often certain situations happened at home (e.g., "You praise your child if he/she behaves well", "You ask your child about his/her day in childcare"). The item scores were summed, and standardized. Higher z-scores indicate more positive involvement with children.
Fathers were presented with a list of specific disorders and could indicate whether they currently had, or have had in the past, any psychiatric illness, i.e. manic depressive or other long-term mental illnesses, and sleeping problems.

Data analysis
Missing values on the sufficient set of covariates ranged from 0.4-0.7% for maternal and paternal education, to 1.5-1.9% for BMI and alcohol habits, and to 2.1-2.7% for LTH of MD, smoking status and yearly income. For the prenatal SCL-5/8 missing values were 2.8% and 4.6%, and for perinatal history of adverse events they were 4.9%. Overall, 16.5% of the pregnancies had incomplete data in at least one of the confounders. There was no specific pattern of missingness, and we explored the distribution of key variables in relation to missingness, according to BZD/zhypnotic exposure status ( Supplementary Figures 3-5). Under the assumption that data were missing at random, we imputed missing values via multiple imputation with chained equation (ten replications) in each maternal disorder stratum. The imputation procedure included exposure and outcome variables, and auxiliary variables (e.g., maternal age and illnesses, parity, comedication, risk factors for the outcomes). Imputed data were used in all analyses. [9][10][11] In the duration and co-exposure analysis, the propensity score was generated using a modified set of sufficient confounders relative to that used for the main timing analysis: the average standardized score for depressive/anxiety symptoms throughout pregnancy; co-medication with antidepressants, opioids, acetaminophen, or sedating antihistamines, at any time during pregnancy. When exploring co-exposure effects, the co-medication under study was omitted as covariate in the propensity score estimation.

Sensitivity and sub-analyses
We examined the robustness of the main results in the depressive/anxiety disorder stratum in a set of sensitivity analyses, specifically by i) restriction to term pregnancies or offspring with no congenital anomaly; ii) specification of alternative models including additional postnatal, paternal and maternal correlates (Supplementary Table 1). We also tested an interaction term between exposure and offspring sex in the final weighed models. When examining child communication, we excluded children with a medically confirmed hearing impairment, as measured by maternal report.
A complete-case approach was undertaken for the timing, duration, and co-exposure analyses, across the three maternal disorder strata. In the latter analysis, only pregnancy-child dyads with complete information on the sufficient set of confounders were included.
We conducted probabilistic bias analyses to correct for non-differential exposure misclassification, unmeasured confounding, and random error simultaneously (20000 simulations) 12,13 using bias parameters stemming from existing studies, 14,15 across the three maternal disorder strata. To address non-differential misclassification of exposure, we specified a trapezoidal distributions for sensitivity (0.35, 0.50, 0.75, 1.0) and specificity (0.97, 0.98, 0.99, 1.0). 14 To address unmeasured confounding by maternal traits of neuroticism or psychoticism, we assumed the prevalence of these traits to be four time more prevalent in BZD/z-hypnotic exposed (0.20) than unexposed (0.05). We allowed the risk posed by maternal personality traits on child motor development and ADHD to vary between 1.5 and 3.0. 15 For this analysis, we dichotomized our outcome measures (cutoff > 1.5 sd). A combined correction for exposure misclassification and unmeasured confounding was also carried out. 13

Sensitivity analysis results
In the depressive/anxiety disorder stratum, restriction to term pregnancies slightly inflated the point estimate for BZD/z-hypnotic late pregnancy exposure on greater gross motor deficits (βw: 0.78; 95% CI: 0.32 to 1.25), whereas restriction to children with no congenital anomaly did not materially change the main results. When excluding children with impaired hearing at 5 years of age, the association with communication remained, although the CI became wider (βw =0.40, 95% CI: -0.04, 0.84).
The alternative model specifications accounting for maternal life-time adverse events, childrelated postnatal factors, or maternal mental health from delivery to five years postpartum, did not materially change the main results. The association measure between late BZD/z-hypnotic exposure and greater gross motor deficits was reduced (βw 0.52-0.58) when respectively accounting for paternal disease or maternal depressive and anxiety symptoms at five years postpartum.
There was an interaction between BZD/z-hypnotics in late pregnancy and child's gender in relation to gross motor deficits (β for interaction=-0.84, 95% CI: -1.40 to -0.27; p=0.004). Only boys exposed in late pregnancy to BZD/z-hypnotics had greater gross (βw: 0.91; 95% CI: 0.47 to 1.35) motor deficits than unexposed. Such association was not evident among girls (βw: 0.22; 95% CI: -0.19 to 0.63). There was no suggestion of sex-specific differences on the other developmental outcomes, or in the other maternal disorder strata.
The alternative model specification including positive parenting practice by child age five years, produced a stronger relationship between BZD/z-hypnotic late pregnancy exposure and fine motor deficits (βw: 0.86; 95% CI: 0.15 to 1.57), relative to unexposed. Here, the association measures relating to gross motor and communication skill deficits did not materially change.
Upon accounting for maternal and paternal ADHD traits, BZD/z-hypnotic exposure in late pregnancy exerted a protective effect on child greater ADHD traits (βw: -0.44; 95% CI: -0.89 to 0.01), relative to unexposed. These two latter model specifications were conducted on reduced sample sizes (65% and 35% of the analytical sample, respectively) since parenting practice and parental ADHD traits were only measured in some of the versions of the MoBa questionnaire.
In the complete-case analyses, BZD/z-hypnotic exposure during late gestation seemed to be associated with all child outcomes, including greater ADHD traits (βw=0.34, 95% CI: -0.03, 0.70) and fine motor deficits (βw=0.56, 95% CI: 0.08, 1.05) in the depressive/anxiety disorder stratum. The protective effect of BZD/z-hypnotic in midpregnancy on child communication skills in the sleeping problems stratum was not evident in the complete-case analysis. The protective effect of BZD/z-hypnotic on child fine motor in the pain-related disorder stratum was consistent in the two missing data approaches.
When re-estimating the association measures with a 99% CI, the results of the timing analysis did not materially deviate from the main findings with a 95% CI across the three maternal disorder strata. Children born to women with depressive/anxiety disorders, who took benzodiazepines/z-hypnotics in late pregnancy, had greater motor skill (βw=0.63, 99% CI: 0.06, In the duration analysis the lower bound of the 99% CI of the association between prenatal exposure to benzodiazepines/z-hypnotics in multiple 4-week intervals and greater fine motor skills, moved substantially away from the null (βw=0.37, 99% CI: -0.19, 0.92), relative to the main analysis. The protective effect of prenatal co-exposure to benzodiazepines/z-hypnotics and opioids on child gross motor skills quantified in the main analysis, was no longer evident (βw=-0.57, 99% CI: -1.50, 0.37).
The association measures between BZD/z-hypnotic exposure during pregnancy and child developmental outcomes were shifted further away from the null after correcting for nondifferential exposure misclassification and random error. Across the maternal disorder strata, failure to account for exposure misclassification would have underestimated the BZD/z-hypnotic effects of about 12-27% (depression/anxiety disorders), 12-90% (sleeping problems) and 34-69% (pain-related disorders). The bias due to exposure misclassification was lowest in relation to child ADHD traits (12%). When correcting for unmeasured confounding by maternal personality traits and random error, the association measures decreased in magnitude (18% bias) across all strata and for all child outcomes. The combined correction for exposure misclassification,    Abbreviations: BDZ=benzodiazepines; ASQ=Ages and Stages Questionnaire; CPRS-R= Conners Parent Rating Scale-Revised; ADHD= attention-deficit/hyperactivity disorder. a Reference: unexposed pregnancies in the corresponding time window. b Weighted estimates with stabilized inverse probability of treatment weighting (constructed at each time point using baseline covariates, time-varying and time-fixed confounding factors, and BZD/z-hypnotic history treatment) and censoring weighting (loss to follow-up at 5 years postpartum). c Weighted estimates with stabilized inverse probability of treatment weighting (constructed as described in b) and censoring weighting during gestation (loss to follow-up at gestational week 30) and postnatally (loss to follow-up at six months and 5 years postpartum).