Association Between Meconium Acetaminophen and Childhood Neurocognitive Development in GESTE, a Canadian Cohort Study

Abstract

Acetaminophen is the only over-the-counter pain reliever that is not contraindicated during pregnancy, but recent studies have questioned whether acetaminophen is safe for the fetus, particularly the developing brain. This prospective birth cohort study probed the previously observed association between in utero exposure to acetaminophen and neurodevelopment by using concentrations of acetaminophen measured in meconium, which more objectively captures exposure of the fetus than maternal report. Exposure, measured by liquid chromatography coupled with tandem mass spectrometry, was categorized into nondetection, low detection, and high detection levels. At age 6–8 years, children completed a set of subtests from the Wechsler Intelligence Scale for Children, 4th edition. Additionally, this study examined potential effect modification by child sex on the association between acetaminophen exposure and neurodevelopment. In fully adjusted models, in utero exposure to acetaminophen was not statistically significantly associated with decreased scores on any of the examined subtests in all children combined (n = 118). The effect of in utero acetaminophen exposure on the Coding subtest was marginally significantly different among boys and girls, with girls performing significantly better on the task with higher levels of acetaminophen compared with girls with undetectable levels of exposure (β_{girls, low} = 2.83 [0.97, 4.70], β_{girls, high} = 1.95 [−0.03, 3.93], β_{boys, low} = .02 [−1.78, 1.81], β_{boys, high} = −.39 [−2.09, 1.31], p_interaction = .06). Effect modification by child sex was not observed on other subtests. These results do not support prior reports of adverse neurodevelopmental effects of in utero exposure to acetaminophen.

Access to pain management during pregnancy is important for women’s care, but options are limited as many over the counter (OTC) pain relievers, including nonsteroidal anti-inflammatory drugs (NSAIDs) and high doses of aspirin, are contraindicated in pregnant women due to risk of miscarriage, birth defects, and premature closure of the ductus arteriosis (Källén and Otterblad, 2003; Li et al., 2003; Rumack et al., 1981). Of the common OTC pain relievers, only acetaminophen is considered safe for use during pregnancy (Scialli et al., 2010). However, recent studies have questioned whether in utero exposure to acetaminophen, which is known to cross the placenta (Levy et al., 1975), can have long-term adverse impacts on child neurodevelopment and behavior (Avella-Garcia et al., 2016; Brandlistuen et al., 2013; Liew et al., 2014, 2016a; Stergiakouli et al., 2016; Thompson et al., 2014; Vlenterie et al., 2016; Ystrom et al., 2017).

These epidemiological studies have been met with concerns about potential unmeasured confounding and study design (Beale, 2017; Damkier et al., 2017; Little et al., 2017; Olsen and Liew, 2016; Saunders and Habgood, 2017). To date, studies have only used maternal self-report of acetaminophen intake. When collected retrospectively, this results in recall bias due to higher levels of reporting and greater reporting precision among parents of children with impaired neurocognition. Even when this metric is collected prospectively to reduce recall bias, it may still be measured with error and—more important—may not accurately capture the dose of acetaminophen that reaches the fetus. Further, the lack of an objective, clinical measurement of neurodevelopment in these studies may have caused overestimation of the adverse effects of acetaminophen—parents of children with some symptoms may overreport those traits—leading to unnecessary concern about the safety of acetaminophen.

Meconium accumulates throughout the second and third trimesters and, therefore, provides a unique opportunity to examine the effects of in utero exposures on the developing fetus (Ostrea et al., 2002). Chemicals detected in meconium reflect those directly deposited from the fetal bile, as well as those passed by the fetus into the amniotic fluid and ingested by the fetus before being deposited (Kwong and Ryan, 1997). Therefore, meconium may contain parent compounds as well as drug metabolites (Berton et al., 2014; Bielawski et al., 2005). To date, meconium has been used to measure illicit substances, particularly long-term maternal drug use within the last 20 weeks or more of pregnancy (Lozano et al., 2007). However, meconium is an underutilized tissue for determining in utero exposure to licit pharmaceutical compounds.

This study aims to address some of the limitations of previous studies by utilizing a commonly used objective evaluation of neurocognitive development, the Wechsler Intelligence Scale for Children, 4th edition (WISC-IV) (Wechsler, 2003), and by measuring concentrations of acetaminophen in meconium, an unbiased quantitative measure of exposure to the fetus.

MATERIALS AND METHODS

Study population.

This study was conducted in the GESTation and the Environment (GESTE) cohort in Sherbrooke, Quebec, Canada. Women were recruited between 2007 and 2009 during the first trimester of pregnancy and at delivery (n = 800). Women who reported using illicit drugs (n = 8) were not retained in the study. At age 6–8 years, 363 children completed a series of neurocognitive tests at the study site, 195 of whom had stored meconium. GESTE families are largely white, French-Canadian and of middle- to upper-class socioeconomic position. Children with collected meconium, neurocognitive, and psychomotor assessments at age 6–8 years, and with available data on all covariates were included in this analysis. We excluded children whose mothers reported smoking or drinking alcohol during pregnancy (n = 14). After excluding individuals with missing covariate data, the final sample size was 118 children. Parents provided informed consent for their children. All study protocols were approved by the Institutional Review Boards of the University of Sherbrooke, Harvard T.H. Chan School of Public Health, and Columbia University.

Exposure assessment.

Meconium was collected from infants after delivery in the hospital and stored at −80°C. If the child passed the meconium prior to delivery it was not collected and the child was not eligible for inclusion in this analysis. Acetaminophen was extracted from < 120 mg meconium by a solid-liquid extraction in ethyl acetate followed by a purification with a dispersive solid phase extraction in acetonitrile (Haroune et al., 2015). The samples were then analyzed by ultraperformance liquid chromatography (Aquity, Waters Corporation, Milford, Massachusetts) equipped with a HSS-T3 column (100 mm x 2.1 mm, 1.8 µm with a 0.2-µm fritted prefilter from Waters Corporation, Milford, Massachusetts) coupled with a mass spectrometer (XEVO TQ, Waters Corporation, Milford, Massachusetts) equipped with an electrospray ionization (ESI) source in positive mode in multiple-reaction-monitoring (MRM) (UPLC-MS/MS). Two product ions (transitions) were used for quantification (the most abundant, m/z = 10) and qualification (the second most abundant, m/z = 93). A 5-µl injection volume was used, and the data obtained were processed using MassLynx 4.1 (Waters Corporation). Concentrations were normalized to the starting mass of meconium. The method was validated with a limit of detection (LOD) at 2 ng/g, limit of quantification (LOQ) at 5 ng/g, a recovery of 104%, and a repeatability of ±15%. Among subjects with detectable levels of acetaminophen in meconium, the median concentration was 59.9 ng/g. In the primary analysis, exposure was categorized into 3 levels to capture the potential nonlinearity of the effect: below the analytical limit of detection (≤ LOD, n = 55), low (> LOD, ≤ 59.9 ng/g, n = 32), and high (> 59.9 ng/g, n = 31). In a sensitivity analysis with continuous exposures, values below the LOQ were replaced by the LOD/$\sqrt{2}$ (Baccarelli et al., 2005).

WISC-IV testing.

Children were invited to complete subtests from the WISC-IV between ages 6–8 years (median 6.5) (Wechsler, 2003). Subtests included Block Design, Coding, Digit Span (forward and reverse), Information, and Vocabulary. The Block Design subtest requires the subject to copy a printed pattern with bicolored blocks and measures spatial perception and reasoning. All GESTE children were under age 8 years and thus were administered the picture-based Coding subtest, which requires the children to draw a specific mark on a series of pictures based on a key and measures fine motor control. During the Digit Span subtest, children are asked to repeat a series of digits of increasing length, first in the same order the administrator says them and then in reverse. Three scores are reported, the forward score, which measures auditory short-term memory, the reverse score, which measures auditory working memory, and a combined score. The Information subtest requires children to define nouns and cultural concepts and measures long-term memory and general knowledge. In the Vocabulary subtest, children are either shown a picture of an object and asked to name it or read a word and asked to define it. This test measures word knowledge and retrieval. Each WISC-IV subtest has an age-scaled score with a mean of 10 and SD of 3.

Statistical analysis.

Data on confounders and predictors of neurocognitive and psychomotor function were obtained from questionnaires (during pregnancy, at delivery, and at the follow-up visit) and maternal medical records from pregnancy. Confounders and other covariates were selected a priori based on previous studies and included maternal characteristics (maternal age at delivery, maternal prepregnancy body mass index, parity, and maternal Raven Matrix score at the follow-up visit as a measure of maternal intelligence; Raven and Raven 2003), child characteristics (sex, gestational age, birth weight, 5-min Apgar score, and age at follow up visit), and socioeconomic characteristics (maternal education and family income dichotomized at the population median). Because the Raven Matrix score is likely not affected by age in early adulthood, raw Raven Matrix scores were used.

In our primary analysis, we fit generalized linear models with each WISC-IV subtest scaled score as the outcome and categorical acetaminophen exposure. Base models adjusted for confounders and predictors of the outcome. Fully adjusted models included confounders that may operate as mediators of the relationship between exposure and outcome including birth weight, gestational age, and 5-min Apgar score. Finally, because toxicological effects on neurodevelopment are often sex-specific, we explored whether acetaminophen has sex-specific effects on neurodevelopment by including an interaction term for child sex (Varshney and Nalvarte, 2017). We also explored effect modification by maternal Raven score, but, due to the small sample size, our results were inconclusive and results are not presented. In sensitivity analyses, exposure was treated continuously to test the linear effect of increasing exposure and dichotomized on whether acetaminophen was detected in the meconium or not. To assess whether our analyses were influenced by selection bias, we employed inverse probability weighting, which did not change results appreciably. All analyses were conducted in SAS 9.4.

RESULTS

We detected acetaminophen in 53% of the study population’s meconium samples. Children who had detectable levels of acetaminophen in their meconium were similar to children who did not, regardless of level of detection (low or high) (Table 1). Slightly more than half of the children were male (53%), and most were born at term with normal birth weight and 5-min Apgar scores. Although children were primarily born at term, children with low exposure had marginally longer gestational age than children with high exposure (39.8 compared with 39.2 weeks, p = .06). The average age at follow up was 6.5 years (± 0.52 SD). Mothers of the children were also similar, although mothers whose child’s meconium had high levels of acetaminophen had slightly higher Raven scores (p = .08). Most women had normal BMIs and just 37% were nulliparous.

Table 1.

Characteristics of the GESTE Population Stratified by Acetaminophen Detection Category (Mean ± SD, Except Where Noted)

		Acetaminophen Detection in Meconium
	Study Population (n = 118)	Not Detected (n = 55)	Low Exposure (n = 32)	High Exposure (n = 31)	p value*
Maternal Characteristics
Maternal Age at Delivery, years	29.5 ± 4.18	29.51 ± 4.39	29.4 ± 4.07	29.6 ± 4.03	.99
Maternal Prepregnancy BMI, kg/m2	24.4 ± 5.15	24.60 ± 5.04	24.3 ± 5.61	25.5 ± 4.93	.39
Parity (n (%))
Nulliparous	44 (37.3)	21 (38.2)	12 (37.5)	11 (35.5)	.97
Maternal Raven Score	54.08 ± 4.65	53.18 ± 5.12	54.5 ± 4.03	55.2 ± 4.15	.08
Family Income, Canadian dollars (n (%))
≤ 63, 000/year	63 (53.4)	32 (58.2)	17 (53.1)	14 (45.2)	.52
Maternal Education (n (%))
Associate’s degree or less	69 (58.5)	33 (60)	19 (59.4)	17 (54.8)	.89
Child Characteristics
Gestational Age, weeks	39 ± 1.33	39.49 ± 1.51	39.8 ± 1.18	39.2 ± 1.06	.06
Sex (n (%))
Male	63 (53.4)	28 (50.9)	17 (53.1)	18 (58.1)	.82
Birthweight, g	3358 ± 466	3363 ± 458	3374 ± 529	3334 ± 424	.99
5-min Apgar score	9.31 ± 0.65	9.25 ± 0.64	9.47 ± 0.51	9.23 ± 0.76	.33
Age at Neurological Test, years	6.55 ± 0.52	6.55 ± 0.49	6.56 ± 0.64	6.55 ± 0.46	.84

^*p values for continuous variables result from Kruskal-Wallis tests, p values for binary variables result from chi-squared tests.

In fully adjusted regression models, the concentration of acetaminophen in meconium was only significantly associated with changes in the Block Design subtest (Figure 1 and Supplementary Table 1). High exposure to acetaminophen was associated with a 1.05-point decrease on the Block Design subtest compared with children with no detectable acetaminophen in their meconium, but this effect was also only marginally significant (95% CI −2.31, 0.22), whereas low exposure to acetaminophen was associated with a 1.22-point increase on this test (95% CI −0.03, 2.48). Across the subtests, estimates for low and high exposure to acetaminophen were generally positive, with some indication of a linear dose response on the Coding and Reverse Digit Span subtests. However, these estimates were not significant. In a sensitivity analysis using a continuous exposure metric, a 1-µg/g meconium increase in acetaminophen was associated with a 0.01-point decrease on the Block Design subtest and a 0.01-point increase on the Information subtest (Supplementary Table 2). An additional sensitivity analysis that dichotomized exposure on acetaminophen detection in meconium found no significant associations, though estimates were in the same direction as the primary analysis (Supplementary Table 3).

Figure 1.

Effect of acetaminophen on Wechsler Intelligence Scale for Children subtest scores. Effect estimates and 95% confidence intervals for increasing exposure to acetaminophen on Wechsler Intelligence Scale for Children (WISC-IV) subtests. All models adjust for child sex, maternal prepregnancy Body Mass Index (BMI), maternal age at delivery, family income, maternal education, parity, child age at neurological test, maternal Raven score (intelligence), gestational age, birthweight, and 5-min Apgar score.

To explore potential effect modification by child sex, an interaction term between acetaminophen exposure level and child sex was included in the model. The effect of acetaminophen on the Coding subtest score was marginally significantly different among boys and girls, with the exposure having no significant association among boys and girls with higher exposure performing better (Figure 2, β_{girls, low} = 2.83, β_{girls, high} = 1.95, β_{boys, low} = .02, β_{boys, high} = −.39, p_interaction = .06). Effect modification was not observed on any other subtest. In a sensitivity analysis using continuous acetaminophen exposure, significant effect modification by child sex was observed only on the Information subtest (Supplementary Table 2, β_boys = .02, β_girls = .00 per 1 µg/g increase in acetaminophen exposure in meconium). Effect modification was not observed on any other subtest. However, in the dichotomous exposure analysis, girls whose meconium contained detectable levels of acetaminophen scored an average of 2.45 points better on the Coding subtest than girls who did not have detectable levels of acetaminophen in their meconium (Supplementary Table 3). This effect was not observed among boys.

Figure 2.

Effect modification by child sex on the association between acetaminophen and Wechsler Intelligence Scale for Children subtest scores. Evidence of modification of the effect of acetaminophen on Wechsler Intelligence Scale for Children (WISC-IV) subtest scores by child sex. All models adjust for child sex, maternal prepregnancy Body Mass Index (BMI), maternal age at delivery, family income, maternal education, parity, child age at neurological test, maternal Raven score (intelligence), gestational age, birthweight, and 5-min Apgar score. Circles represent point estimates for subjects with no detectable acetaminophen in meconium (control). Triangles represent point estimates for subjects with low exposure to acetaminophen relative to sex-specific controls. Squares represent point estimates for subjects with high exposure to acetaminophen relative to sex specific controls. Whiskers are 95% confidence intervals.

DISCUSSION

In this study, acetaminophen concentrations in meconium had no consistent association with WISC-IV subtest scores in midchildhood. Although inclusion of acetaminophen in our model was significant on the Block Design subtest, low exposure were associated with an increase in score whereas high exposure were associated with lower scores. This indicates that there is no cumulative decrement on this subtest attributable to acetaminophen exposure. Further, the significance of this association was driven by the positive association between low exposure and Block Design score. There was a positive association between acetaminophen levels in meconium and higher scores on the Coding subtest of the WISC-IV among girls, which was also observed in a supplementary analysis using a binary exposure metric. The effect size (2.8 points comparing girls with high exposure to those with no exposure) is similar to the SD on WISC-IV subtests (3 points). Although the binary exposure analysis uses a different metric than previous studies, which primarily compare the children of mothers who report acetaminophen use to those whose mothers report no acetaminophen use, the percent of meconium samples with detectable levels of acetaminophen is comparable to the percent of women who reported using acetaminophen in those studies. Additionally, in this population, subjects whose mothers had recorded administration of acetaminophen at delivery had significantly higher concentrations of acetaminophen in their meconium (data not shown).

Although in one sensitivity analysis using continuous exposure there was an observed decrease on the Block Design subtest, the effect size is minimal (0.01 point per 1 µg/g), particularly considering the relatively large difference in exposure relative to the median exposure level in the population (5.99*10⁻²µg/g). The observed increase on the Information subtest in this sensitivity analysis may be explained by the sensitivity of this test to residual confounding by parental education and intelligence because it is partially a measure of general knowledge. Further, it is important to note in this observational study that the observed associations are not necessarily causal.

Although the results of this study seem discordant with previous studies (Avella-Garcia et al., 2016; Brandlistuen et al., 2013; Liew et al., 2014, 2016a; Stergiakouli et al., 2016; Thompson et al., 2014; Vlenterie et al., 2016; Ystrom et al., 2017), they are not necessarily in conflict. In addition to the improved exposure assessment in our study, previous studies have primarily used instruments that measure child behavior, motor development, and symptoms of specific behavioral disorders including Attention-Deficit/Hyperactivity Disorder (ADHD) and Autism Spectrum Disorder (ASD). This study used subtests from the WISC-IV battery, which is an objective, validated metric that is commonly used to measure components of child intelligence. Because behavior and intelligence are different neuropsychological constructs, our results cannot be directly compared with other studies. Importantly, many of the instruments previously used relied on parental report of behavior, which may be inaccurate or biased (Bennett et al., 2012; Durbin and Wilson, 2012).

Additionally, few animal studies have been undertaken to elucidate the mechanism of action of acetaminophen on the developing brain. A mouse model has indicated the potential importance of increased brain-derived neurotrophic factor (BDNF) following acetaminophen exposure (Viberg et al., 2014). However, this study did not explore whether BDNF mediated the association between exposure to acetaminophen during developmentally sensitive windows and later-life cognition. Adult rat models have demonstrated the potential for acetaminophen to interfere with neurotransmission and the balance of amino acids in brain regions relevant to behavior and intelligence (Blecharz-Klin et al., 2013, 2014). However, the administered doses greatly exceed those observed in our study. Other hypothesized mechanisms of action include oxidative stress, endocrine disruption, or direct neurotoxicity (Avella-Garcia et al., 2016; Brandlistuen et al., 2013; Konkel, 2018; Liew et al., 2016a; Stergiakouli et al., 2016). Because of inconsistent epidemiological findings and without an identified mechanism, causality of any association observed in other studies cannot be established.

It is also important to place the potential harmful effects of in utero exposure to acetaminophen on neurodevelopment observed in other studies in the context of the observed effects of other pain relievers and antipyretic medications that may be substituted by pregnant women. Currently, pregnant women are advised against nonsteroidal anti-inflammatory drugs (NSAIDs) such as ibuprofen and naproxen because studies have associated first-trimester use with increased risk for miscarriage (Nakhai-Pour et al., 2011) and cardiac malformations (Ericson and Källén, 2001) and third-trimester use with premature closure of the ductus arteriosus (Koren et al., 2006), renal impairment (Bennett et al., 1996), and other adverse health effects in childhood (Nezvalova-Henriksen et al., 2013, 2016). Further, use during the peri-conceptual period is associated with reduced female fecundity (McInerney et al., 2017). Similarly, most pregnant women are advised against using high doses of aspirin, or using aspirin for extended periods (James et al., 2008; Rumack et al., 1981). However, it is advised that some women at high risk for pre-eclampsia take low doses of aspirin to lower that risk (Wright et al., 2017). Because of the concerns regarding NSAIDs and high-dose aspirin, it is critically important that the risks and benefits of treating pain and fever during pregnancy with acetaminophen are thoroughly studied and understood before any usage recommendations are made to pregnant women.

Our findings are reassuring with respect to the neurodevelopmental outcomes we studied, in the population we studied, but should be interpreted within the context of our observational study design. Although measuring concentrations of acetaminophen in meconium provides a robust, quantitative estimate of exposure, the method only measures the parent compound and not its metabolites. Acetaminophen is designed to metabolize quickly, and on an average is excreted within 6 h of dosing (Prescott, 1980). Thus, we may have underestimated fetal exposure to acetaminophen or its hepatotoxic metabolite, N-acetyl-p-benzoquinone imine (NAPQI). However, NAPQI is not known to cross the placenta, and thus any exposure would be from fetal metabolism of acetaminophen (Wilkes et al., 2005). The enzymes involved in acetaminophen metabolism, especially through NAPQI production and detoxification (CYP2E1, CYP1A2, CYP3A4, and GST) have varying activity levels during fetal development, which may result in differential exposure measurement depending on the timing of exposure, which may be related to acetaminophen neurotoxicity (Hines, 2008). Further work that measures individual enzymatic levels and activity in fetal organs might also shed light on differential susceptibility to acetaminophen toxicity. However, the percent of children with detectable levels of acetaminophen in their meconium (∼50%) is similar to the percent of women who report taking acetaminophen at some point during pregnancy in the United States, a similar population (∼65%) (Werler et al., 2005), indicating that the meconium captures maternal acetaminophen use once it begins to form in the second trimester and throughout the third trimester. Therefore, our study captures third-trimester exposure, which previous studies have indicated as a potentially sensitive window for the effect of acetaminophen exposure on neurodevelopment (Stergiakouli et al., 2016). Although meconium captures a wide window of exposure, it is not possible to determine precisely when during that window a fetus was exposed to the tested substance (Farst et al., 2011). Another limitation is the relatively small sample size. However, the homogeneous population in addition to precise measures of exposure and outcome reduce the likelihood that our null findings are the result of residual confounding by socioeconomic factors and race or misclassification of the exposure and outcome. Lastly, we are not able to control for indication for acetaminophen use. Although previous studies have found that confounding by indication does not significantly alter effect estimates (Brandlistuen et al., 2013; Liew et al., 2016a), or that the effect was further attenuated when accounting for indication (Liew et al., 2016b), it is possible that there is residual confounding or effect modification by the underlying reason for acetaminophen use.

Despite limitations, this is the first study to use a quantitative, objective measurement of acetaminophen exposure during pregnancy. As such, it is the first to approach a dose-response relationship between acetaminophen and neurodevelopment. Additionally, the use of WISC-IV subtests to capture the outcome provides an objective assessment of neurodevelopment. Whereas many previous studies have used parental report of their child’s abilities, our measurement is less biased and thus improves estimation of the effect estimate. Further, our a priori selection of covariates was based on previous studies, but also considered the relationships between these covariates, acetaminophen, and neurodevelopment using a causal directed acyclic graph. Lastly, the genetic and socioeconomic homogeneity of the GESTE cohort limit confounding by these factors.

In conclusion, we did not find evidence of neurodevelopmental harm from prenatal exposure to acetaminophen measured in meconium. Given the limited medical options for treatment of fever and pain, acetaminophen will likely continue to be used during pregnancy. Although our study provides some reassurance, its observational nature cannot definitively determine that women can safely consume acetaminophen during pregnancy and ongoing surveillance of long-term outcomes of fetal exposure remains warranted.

SUPPLEMENTARY DATA

Supplementary data are available at Toxicological Sciences online.

FUNDING

This work was supported by the National Institute of Environmental Health Sciences (R21ES024841, R01ES027845, P30ES009089, K23ES022242) and the Canadian Institutes of Health Research (MOP-84551). The funding sources had no involvement in study design, collection, analysis, data interpretation, manuscript preparation, or the decision to submit the article for publication.