Behavioral Problems at Age 11 Years After Prenatal and Postnatal Exposure to Acetaminophen

Parent-Reported and Self-Reported Outcomes

Kosuke Inoue; Beate Ritz; Andreas Ernst; Wan-Ling Tseng; Yuying Yuan; Qi Meng; Cecilia Høst Ramlau-Hansen; Katrine Strandberg-Larsen; Onyebuchi A. Arah; Carsten Obel; Jiong Li; Jørn Olsen; Zeyan Liew


Am J Epidemiol. 2021;190(6):1009-1020. 

In This Article


Study Participants

The DNBC was established in Denmark during 1996–2002, when 100,418 pregnant women enrolled in the cohort at their first general-practitioner antenatal visit (during weeks 6–12), and the mothers and children have been followed since.[33] For analyses of prenatal acetaminophen use, we restricted the cohort to live-born children whose mothers answered the study enrollment form and the 3 subsequent telephone interviews (scheduled at approximately the 12th and 30th gestational weeks and at 6 months after birth), all of which collected information on prenatal acetaminophen use (n = 64,322). Among them, 40,934 had SDQ outcome scores reported by both the mother and the child when the index child was 11 years old. For postnatal acetaminophen exposure analyses, we additionally restricted the cohort to mothers who had answered the interview conducted at 18 months after birth with information on the infant's acetaminophen treatment (n = 27,742). Details in the study population selection are described in Web Figure 1 (available at All participants provided written informed consent at the time of inclusion in the DNBC. The research protocol for this study was approved by the DNBC steering committee (project no.: 2018–13), Danish data inspectorate (journal no.: 2016–051-000001, serial no.: 1297), and the institutional review boards at the University of California, Los Angeles (16–001849), and Yale University (2000024089).

Exposures to Acetaminophen

Information about maternal acetaminophen use during pregnancy was ascertained from the study enrollment form and 3 computer-assisted telephone interviews. At the first contact, women answered questions regarding any supplement and medication use covering the period from 4 weeks before pregnancy to the gestational week of reporting. In the subsequent telephone interviews, women were specifically asked to report whether they had taken any painkillers during pregnancy and provided with a list of 44 common medications, including acetaminophen as a single or combination drug. Women were asked to indicate the gestational week of intake for each medication, and we used the weekly intake information to calculate trimester-specific and cumulative weeks of use. Information regarding acetaminophen exposure during infancy was ascertained through the computer-assisted telephone interviews at about 6 and 18 months postpartum. Mothers were asked to report whether their children had experienced any of 16 types of conditions or diseases and the specific pharmaceutical treatment for these conditions (Web Table 1).

Parental and Self-reports of Children's Behavioral Problems at Age 11 Years

Children's behaviors were assessed based on the standardized SDQ, which is a 25-item screening tool that assesses behavioral problems and mental health status of children and adolescents between the ages of 4 and 18 years.[34] When the DNBC children turned 11 years of age, both parents and children were invited to complete the SDQ. There are 5 SDQ subscales (emotional symptoms, conduct problems, hyperactivity/inattention, peer problems, and prosocial behavior), all consisting of 5 items. Based on the recommendations for scoring the SDQ (, we calculated a total difficulties score (range, 0–40) by summing the first 4 subscales, ranging from 0–10 each, with higher scores indicating more negative behaviors and problems. We then dichotomized each subscale according to the recommended cutoff to indicate atypical behaviors for the parent-reported and child-reported SDQ.[34] We also created an "internalizing" subscale, which combined the emotional symptoms and peer problems subscales, and an "externalizing" subscale that combined the conduct and hyperactivity/inattention subscales.[35,36] There are no recommended cutoff points for internalizing or externalizing composite scores; thus, the top 95th percentile of each distribution was defined a priori as the cutoff. A subset of parents also answered 6 questions (each with a possible response value of 0, 1, or 2) about their own behavioral problems during childhood when the index child turned 7 years of age,[7] which allowed us to generate a parental behavioral problems score (range 0–12).

Statistical Analysis

We calculated Pearson correlation coefficients (r), Cronbach's α, and κ coefficients between parent-reported and child-reported SDQ. We employed modified Poisson regression models to estimate risk ratio and 95% confidence interval for acetaminophen exposure and binary classifications for overall behavioral problems, internalizing behaviors, externalizing behaviors, and each of the 5 subscales (emotional symptoms, conduct problems, hyperactivity/inattention, peer relationship, and prosocial behavior) reported either by parents or children. To compare relative risk estimates between parent-reported and child-reported SDQ, we computed P values for heterogeneity using a generalized linear mixed model including a product term for acetaminophen exposure and the rater of SDQ (i.e., parent or child). To account for potential differences in neurobehavioral development in boys and girls,[37] we also conducted analyses stratified by child's sex to evaluate effect measure modification. We tested for heterogeneity using a product term for acetaminophen exposure and child's sex. For the prenatal period, we further analyzed trimester-specific exposure (used only in the first (1–12 weeks), second (13–24 weeks), or third trimester (25th week to delivery), in any 2 trimesters, or in all trimesters) and cumulative weeks of exposure (1–5, 6–10, or >10 weeks). We also evaluated the linear exposure-response by fitting the number of weeks of acetaminophen use in pregnancy as a continuous variable.

Potential confounders were selected a priori considering factors that might affect child neurobehavioral development and might also be associated with acetaminophen exposure. In all models, we included mother's age at childbirth (continuous), parity (1, 2, >2), socio-occupational status (low, medium, high), maternal prepregnancy body mass index (calculated as weight (kg)/height (m)2; <18.5, 18.5–24.9, 25.0–29.9, ≥30.0), and birth year (continuous). In the model for acetaminophen exposure during pregnancy, we additionally adjusted for maternal smoking during pregnancy (never and ≤9 or >9 cigarettes per day), maternal alcohol intake during pregnancy, mother's self-reported psychiatric illnesses before and during pregnancy, indications for maternal acetaminophen use (including diseases of muscles or joints during pregnancy, episodes of fever during pregnancy, and inflammation/infections during pregnancy), and prenatal use of nonsteroidal antiinflammatory drugs such as aspirin and ibuprofen. In models estimating the effect of postnatal exposure to acetaminophen, we additionally adjusted for acetaminophen use during pregnancy, the major indications for acetaminophen use during infancy (including fever, infection or inflammation of the eye, ear or throat, and respiratory tract illnesses), being born preterm (defined as gestational age less than 37 weeks), and birth weight in grams (<2,500, 2,500–4,500, >4,500). Ten simulated complete data sets were generated via imputation assuming multivariate normal distribution for about 8% of participants who had at least 1 missing covariate value.[38]

We also conducted several sensitivity analyses. First, we reclassified the outcome defined as children who met the SDQ cutoff in both the parental and the child assessments.[39] Second, we additionally controlled for parental childhood behavioral problem scores to account for familial and genetic risks.[40] Third, given the potential confounding by mother's breastfeeding for postnatal exposure,[41] we also adjusted for duration of breastfeeding (none or, in months, <3, 3–6, >6) in the analyses for infancy use of acetaminophen. Fourth, we employed negative binomial regression models to estimate adjusted relative ratio for increasing SDQ score as count data according to prenatal or postnatal acetaminophen exposure. Finally, we evaluated whether the results were sensitive to the SDQ cutoff point used to define behavioral difficulties by varying the cutoff from −2 to +2 of the scores employed in the main analyses.

In the main analyses, we used the inverse probability of selection weight (IPSW) technique to account for possible selection bias due to nonparticipation at the 11-year follow-up.[42] Stabilized IPSW and 95% confidence intervals estimated with robust variance estimators were incorporated in all regression analyses. In sensitivity analyses, we compared results with and without IPSW. All statistical analyses were performed using SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina).