Maternal Spontaneous Abortion and the Risk of Attention-deficit/Hyperactivity Disorder in Offspring

A Population-based Cohort Study

Hui Wang; Fei Li; Maohua Miao; Yongfu Yu; Honglei Ji; Hui Liu; Rong Huang; Carsten Obel; Jun Zhang; Jiong Li


Hum Reprod. 2020;35(5):1211-1221. 

In This Article

Materials and Methods

Design and Population

We carried out a nationwide cohort study using data from the Danish national registers (a detailed description of registers is provided in Supplementary Table SI) (Timmermans, 2010; Lynge et al., 2011; Mors et al., 2011; Wallach Kildemoes et al., 2011; Schmidt et al., 2015). In Denmark, all live births have a unique personal identification number that permits the accurate linkage of individual-level data. We identified all singleton live births from 1 January 1995 to 31 December 2012 (n = 1 129 030) from the Danish Medical Birth Registry. We excluded 24 146 children who had missing or extreme gestational age (<154 or >315 days), 4660 stillbirths, 3121 children with chromosomal abnormalities (International Classification of Disease [ICD]-10 codes Q90–99) identified from the Danish National Patient Register (DNPR), 523 children without links to their fathers and 14 523 children who died or emigrated within the first 3 years. We further excluded 18 548 children whose mothers were born earlier than 1 January 1960 and did not have valid information on SA (Lynge et al., 2011; Schmidt et al., 2015). We also excluded 842 children whose mothers had a recorded date of abortion during the gestational period of a successful pregnancy. In the final analyses, we included a total of 1 062 667 children (as shown in Figure 1). We followed each child from 3 years of age until the date of the first diagnosis of ADHD, emigration, death or end of follow-up (31 December 2016), whichever came first. The study was approved by the Danish Data Protection Agency (2015-57-0002).

Figure 1.

Flowchart showing the identification of the eligible participants and analysis sample.

Data on Maternal History of Spontaneous Abortion

Information on maternal SA was obtained from the DNPR, using ICD codes (ICD-8 codes during 1977–1993: 6438, 6439, 6346 and 6451; ICD-10 codes since 1994: O03 and O021) (Lohse et al., 2010). We categorized the children into three groups: (i) unexposed children, (ii) children of mothers who had one single SA (1 SA) before the childbirth and (iii) children of mothers who had at least two SAs (≥2 SAs) before the childbirth.

Identification of ADHD Individuals

Children were classified as ADHD individuals if they either had a hospital diagnosis of ADHD or redeemed ADHD medication prescription for at least twice after the age of 3 years. Information on hospital ADHD diagnosis was obtained from the DNPR based on ICD-10 codes (F90.0-F98.8) (Schmidt et al., 2015). Data on ADHD specific drug use were extracted from the National Prescription Register. The Anatomical Therapeutic Chemical codes for ADHD-specific medication were N06BA04 (methylphenidate) and N06BA09 (atomoxetine). Children who had redeemed N06BA07 (modafinil) were included as ADHD individuals only if they had previously redeemed a prescription for either N06BA04 or N06BA07 (Wallach Kildemoes et al., 2011; Pottegård et al., 2012). Two or more medications prescribed on the same date were counted as one prescription. The date of the first ADHD diagnosis or medication was defined as the onset of ADHD. Previous studies have suggested a high validity of using both the hospital register and the prescription register to identify ADHD in children (Christensen et al., 2019; Sun et al., 2019), and the probability of misclassification of ADHD was relatively low and the inter-rater agreement was high (96%) (Mohr-Jensen et al., 2016).

Ethical Approval

The study was approved by the Danish Data Protection Agency (No. 2013-41-2569). According to Danish legislation, no informed consent is required for a registry-based study using anonymized data.


Based on previous research (Klemetti et al., 2012; Ahrens et al., 2016), the following factors were considered as potential confounders: sex of the child (male, female), preterm birth (yes [gestational age at birth <37 weeks], no), low birth weight (yes [birth weight < 2500 g], no), low Apgar score at 5 min (yes [Apgar score < 7], no), small for gestational age (SGA) (yes [birth weight below the 10th percentile for infants of the same gestational age and sex], no), calendar period of birth (1995–1998, 1999–2002, 2003–2006, 2007–2009, 2010–2012), maternal age at birth (≤25, 26–30, 31–35, ≥36), paternal age at birth (≤25, 26–30, 31–35, ≥36), maternal smoking during pregnancy (yes, no), maternal country of origin (Denmark, other countries), maternal education level (0–9, 10–14, ≥15), maternal cohabitation status (yes, no), maternal history of diabetes (yes, no), maternal hypothyroidism (yes, no), maternal psychiatric disorder before the childbirth (yes, no) and paternal psychiatric disorder before the childbirth (yes, no). The information for maternal social status and origin of country was obtained from the Danish Integrated Database for Longitudinal Labor Market Research (Timmermans, 2010). Information for maternal diabetes, hypothyroidism and parental psychiatric disorders was retrieved from the DNPR and Danish Psychiatric Central Research Register (Lynge et al., 2011; Mors et al., 2011).

Statistical Analyses

We used Cox proportional hazards regression model to estimate the hazard ratio (HR) with 95% confidence intervals (CIs) for the association of maternal history of SA with the risk of ADHD in offspring. The primary analysis was to compare the rate of ADHD in unexposed children with the rates in children of mothers with one SA or with at least two SAs before the childbirth. To control for the correlations of sequential births of the same mother, the robust sandwich estimator for correction of standard errors was used. We investigated the interaction between maternal history of SA and birth order based on the statistical significance of interaction terms in the Cox proportional hazards model. As an increased number of SA may indicate more severe conditions; we thus tested for a trend between maternal history of SAs and ADHD risk in offspring. This assumed an equidistant stepwise function for the level of maternal history of SA (continuous variable coded: no SA = 0, 1 SA = 1, ≥2 SAs = 2). Additionally, we tested whether the associations varied according to sex of the child. We performed four models for adjusting potential confounders. Only sex and birth year were included in Model 1. Model 2 was additionally adjusted for parental age at birth, parity, maternal education level, maternal origin, maternal cohabitation and maternal smoking during pregnancy (Galéra et al., 2011; Hvolgaard Mikkelsen et al., 2016). Model 3 was further adjusted for parental psychiatric disorders before the childbirth, maternal diabetes and hypothyroidism status (Päkkilä et al., 2014; Xiang et al., 2018). In Model 4 we additionally adjusted for preterm birth, low birth weight, low Apgar score at 5 min and SGA.

The positive predictive value of SA is high (97.4%) in the National Patient Register, while sensitivity and specificity have not been reported (Lohse et al., 2010). We tried to account for misclassification of maternal SA by using the probabilistic sensitivity analysis (Kristensen and Irgens, 2000; Orsini et al., 2008). Probabilistic sensitivity analysis was used to provide an external adjustment of the observed odds ratio (OR) upon specification of hypothetical values for maternal SA (Orsini et al., 2008). The probabilistic sensitivity analysis through Monte Carlo simulations involved two iterated steps: (i) draw a random sample from the specified probability density functions of maternal SA and (ii) back-calculate the bias-adjusted OR from maternal SA, which were repeated several times to obtain a bias-adjusted OR (Fox et al., 2005).

We performed mediation analyses to examine whether adverse birth outcomes (low birth weight, preterm birth, low Apgar score and SGA) mediated the association between maternal SA and ADHD in offspring by calculating direct and indirect effects (via the mediator) in the Stata module PARAMED (Emsley and Liu, 2013). The proportion of mediation was calculated as log (natural indirect relationship)/log (total relationship).

We estimated the population attributable fraction by using the adjusted hazard ratio of ADHD in the group whose mothers had a history of SA versus the unexposed group and the prevalence of SA before the childbirth in the entire population (P pop) according to the formula population attributable fraction = P pop × (HR − 1)/(P pop × (HR − 1) + 1) (Rockhill et al., 1998). To evaluate the robustness of the results to potential unmeasured confounding, we calculated the E-value for the overall estimate using the publicly available online E-value calculator for hazard ratios with an outcome prevalence of <15% ( The E-value is a measure that represents the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome to fully explain the association (VanderWeele and Ding, 2017).

Sensitivity Analyses

To test the robustness of our results, we did several sensitivity analyses. First, to remove the effect of induced abortion, we repeated our analyses by excluding the mothers who had induced abortion before the childbirth (n = 19 976). Induced abortion was identified as follows: ICD-8 codes, 640, 641 and 642; ICD-10 codes, O04, O05 and O06 (Lohse et al., 2010). Second, to examine whether the associations were modified by parental psychiatric disorder before the childbirth, we performed analyses stratified by parental psychiatric disorders (yes, no). Third, to ascertain potential mediating effects of neonatal outcomes, we performed the analyses stratified by preterm birth, low birth weight, lower Apgar score at 5 min and SGA. Fourth, spontaneous abortions include missed abortions (ICD-8 codes: 6346 and 6451; ICD-10 code: O021) and other SAs (ICD-8 codes: 6438 and 6439; ICD-10 code: O03) (Lohse et al., 2010), which may represent different aetiology. To illustrate whether the associations differed by these two subtypes, we repeated the analyses individually. Fifth, concerning that the fertility treatment could also affect the risk of ADHD in offspring (Bay et al., 2013), we performed the analysis excluding children born to women with fertility problems (ICD-8 code: 628; ICD-10 code: N97) (Svahn et al., 2015). Sixth, we investigated whether maternal SA after the childbirth was associated with ADHD risk in offspring to examine the role of genetic susceptibility and stable family environment over time. Seventh, we restricted analyses to offspring born before 2010 to exclude children who might not be able to receive a diagnosis at very young ages. We also performed the analyses by including all children born from 1995 to 2016. Lastly, we used the multiple imputation procedure by chained equations to impute 10 replications to handle missing values of these covariates, including birth weight, Apgar score at 5 min, maternal smoking status during pregnancy, maternal education level and maternal cohabitation.

All statistical analyses were performed using Stata, version 15.1 (StataCorp).