Associations between paracetamol (acetaminophen) intake between 18 and 32 weeks gestation and neurocognitive outcomes in the child: A longitudinal cohort study

Abstract Background The majority of epidemiological studies concerning possible adverse effects of paracetamol (acetaminophen) in pregnancy have been focussed on childhood asthma. Initial results of a robust association have been confirmed in several studies. Recently, a few cohort studies have looked at particular neurocognitive outcomes, and several have implicated hyperactivity. Objectives In order to confirm these findings, further information and results are required. Here, we assess whether paracetamol intake between 18 and 32 weeks gestation is associated with childhood behavioural and cognitive outcomes using a large population. Methods Data collected by the Avon Longitudinal Study of Parents and Children (ALSPAC) at 32 weeks gestation and referring to the period from 18 to 32 weeks, identified 43.9% of women having taken paracetamol. We used an exposome analysis first to determine the background factors associated with pregnant women taking the drug, and then allowed for those factors to assess associations with child outcomes (measured using regression analyses). Results We identified 15 variables independently associated with taking paracetamol in this time period, which were used as potential confounders. Of the 135 neurocognitive variables considered, adjusting for the likelihood of false discovery, we identified 56 outcomes for adjusted analyses. Adjustment identified 12 showing independent associations with paracetamol use at P < .05, four of which were at P < .0001 (all related to child behaviours reported by the mother at 42 and 47 months; eg conduct problems: adjusted mean score + 0.22 (95% confidence interval 0.10, 0.33)). There were few associations with behavioural or neurocognitive outcomes after age 7‐8 years, whether reported by the mother or the teacher. Conclusions If paracetamol use in mid‐to‐late pregnancy has an adverse effect on child neurocognitive outcome, it appears to mainly relate to the pre‐school period. It is important that these results be tested using other datasets or methodologies before assuming that they are causal.


| BACKG ROU N D
The thalidomide tragedy involved an apparently benign drug being taken for a condition that is common in early pregnancy (nausea and vomiting); its teratogenic effect was discovered because many of the offspring were born with striking and unusual malformations. 1 Ever since, there has been a concern that relatively common adverse effects would not be easily identified, especially if they were not apparent immediately after the child was born. Few such effects have been discovered, but few drugs taken in pregnancy have been investigated for outcomes that would take many years to be recognised. This is particularly (but not exclusively) true of over-the-counter medications, currently the most common of which is paracetamol.
In general, despite evidence that paracetamol crosses the placenta 2 and is a known endocrine disruptor 3 and a COX inhibitor, 4 the medical profession is relatively relaxed about the use of this analgesic in pregnancy; it is usually claimed to be the analgesic of choice. It is true that it has not been implicated in major malformations, 5 but there has been consistent evidence that women who take paracetamol in pregnancy increase the risk of their child developing asthma. 6 In addition, there is now increasing evidence that prenatal exposure to the drug increases the risk of childhood hyperactivity and diagnosed ADHD. 7,8 There is some evidence of an increased risk of autism 9 but one study found that this was conditional on the autistic child being hyperactive. 10 In addition, other neurocognitive outcomes have been suggested and these include: various behaviour problems; 11 poorer attention and executive function; 12 diagnoses of Oppositional Defiant Disorder (ODD) 13 and cerebral palsy. 14 In contrast, there are no reported associations in the offspring of psychotic symptoms 15 or miscarriage, 16 and contradictory findings with IQ. 17,18 Because it is so common, and no links with congenital malformations have been made, paracetamol is considered the analgesic of choice for pregnant women. 19 It is commonly taken in Europe, with estimates of fetal exposure from over 40% of pregnancies in Spain, 9 55% in Denmark 7 to 65% in France. 20 When assessing effects of prenatal exposures to the child with a neurocognitive or behavioural problem, maternal recall of events in a pregnancy occurring years earlier, is likely to result in failure to record many common exposures, particularly if she has had more than one pregnancy.
Consequently, the most efficient and effective method is to use data collected longitudinally, starting in pregnancy. This we do here by analysing information collected by the Avon Longitudinal Study of Parents and Children (ALSPAC). First, we determined the factors that were independently associated with the intake of paracetamol using an exposome technique, 21 and then taking these factors into account, we determined independent associations with cognitive and behavioural outcomes of the offspring.

| ME THODS
Pregnant women who were resident in Avon, UK with expected dates of delivery between 1st April 1991 and 31st December 1992 were invited to take part in the Avon Longitudinal Study of Parents and Children (ALSPAC). [22][23][24] The initial number of pregnancies enrolled was 14 541 (for these at least one questionnaire had been returned). Of these initial pregnancies, 14 150 reached 32 weeks (391 having miscarried, been terminated or delivered before this gestation). Questionnaires sent out at this stage included information on paracetamol intake, and were returned by 12 418 (88%) of It is important that these results be tested using other datasets or methodologies before assuming that they are causal.

K E Y W O R D S
ALSPAC, child cognitive outcomes, childhood behaviour, exposome, paracetamol, prenatal medication

Study Question
Does paracetamol consumption at 18-32 weeks of gestation affect childhood neurocognition or behaviour?

What's already known
Previous epidemiological studies concerning possible adverse effects of paracetamol (acetaminophen) consumption in pregnancy have suggested associations with cognition, and childhood behaviour, particularly involving hyperactivity and/or attention problems.

What this study adds
Using a large prospective cohort of 14,062 children, linked to information on child cognitive and behavioural outcomes from 6 months to 17 years, we have shown that paracetamol consumption between 18 and 32 weeks gestation was associated with adverse trends in pre-school child behaviour, but the associations were no longer present by the end of primary school (age 10-11 years). Boys appeared to be more susceptible than girls to possible behavioural effects of the drug. these women. The study website contains details of all the data that are available through a fully searchable data dictionary and variable search tool: http://www.brist ol.ac.uk/alspa c/resea rcher s/our-data/.
Ethical approval for the study was obtained from the ALSPAC Ethics

| Statistical analysis
The analyses were undertaken in two strands: (a) a determination of those variables to be included as potential confounders, and (b) an assessment of the relationship between paracetamol and outcome, taking the potential confounders into account; further analyses repeated the analyses for boys and girls separately. 2. Altogether 135 unadjusted relationships between paracetamol intake 18-32 weeks and the cognitive and behaviour outcomes recorded by ALSPAC were considered. Of these, 41% (n = 56) associated at P < .001 were identified for adjusted analysis (Table 1).
In instances, where more than one similar outcome were available a decision was made to use the outcome with the highest numbers of children involved.
In general, the outcomes were measured on a continuous scale and thus multivariable linear regression was employed using the raw scores. Since there were no differences between boys and girls in regard to the likelihood that they had been exposed to paracetamol (50.5% and 49.5%, respectively), sex of the child was not used as a confounder. However, in order to determine whether associations differed by sex, stratified analyses were undertaken to determine whether there were apparent interactions-if so, interaction terms would be included in the analyses.

| Missing data
We have not modelled using missing data techniques as the data are unlikely to be missing at random. The proportion of missing data varies for each variable, but can be readily ascertained from the ALSPAC study data dictionary (www.brist ol.ac.uk/alspa c/resea rcher s/our-data); it varies from 0% for the sex of the child to 62% for the IQ measure at age 15.

| Identification of potential confounders
As noted earlier, 43.9% of the women answering the questionnaire administered at 32 weeks gestation reported taking paracetamol since 18 weeks. In order to identify potential confounders, we considered all features of the mother that were relevant to the period 18-32 weeks pregnancy or earlier in pregnancy or during the mother's previous life. Of the factors considered, 33 were associated at P < .0001 with paracetamol intake, and a further two variables were considered as likely associations even though not associated at this level (mother had rheumatism and social class of partner). The 35 variables were divided into four groups (Table S1)  were offered together, with the result that pelvic inflammatory disease, hypertension and a history of severe depression were dropped.
In parallel, the 11 surviving variables from (c) and (d) were offered together-caffeine consumption, residing in public housing and crowding index were dropped. The 20 surviving variables were then offered together, and five variables were eliminated (anxiety and depression at 32 weeks; no. of cigarettes smoked at 32 weeks; whether the mother worked during pregnancy, and her age at first pregnancy).
The 15 independent predictors of taking paracetamol between 18 and 32 weeks are shown in Table 2. Not unexpectedly, the factor with the largest odds ratio was the report of headache(s) in the 18-32 week period. This was independent of having a history of migraine. Protective factors were close adherence to eating a 'healthy' dietary pattern and of taking no alcohol during this time period.
Interestingly, none of the traditional measures of socio-economic status were retained in the model.
The 15 variables were treated as possible confounders, since we did not consider it warranted to carry out similar modelling procedures for each of the 56 outcomes and thence identify factors predictive of both paracetamol intake and each outcome. The confounders identified in this way would necessarily include some of the factors predictive of paracetamol intake.

| Adjusted analyses for cognition variables
Of the 18 measures of cognitive function considered, the eleven relating to IQ were considered further; almost all subtests and overall measures showed strong inverse unadjusted associations (Table S3).
However, on allowing for the 15 potential confounders, although all measures continued to show inverse associations, the effects were largely attenuated: only one showed a significant trend-the 'Freedom from distractibility' subset of IQ measured at age 8 (mean difference (MD) −0.35 (95% CI −0.70, −0.00)) ( Table 3). Stratification by sex of the child showed no differences of note (Tables 4, 5 and S3a,b).

| Adjusted analyses for temperament variables
There were 11 measures of the child's temperament showing a significant unadjusted association with maternal paracetamol intake.

| Adjusted analyses for hyperactive behaviour
Measures of attention and excessive activity used the Strengths and Difficulties Questionnaire (SDQ) measures 28 at various ages from 3 to 11 years including both maternal and teacher reports, as well as from the adapted Development and Well-being Assessment (DAWBA) series of questions answered by mother and teacher at age 7-8. 29 Of the seven measures of hyperactivity using the SDQ type questions, only two survived adjustment-hyperactivity identified in her child by the mother at ages 42 and 47 months (Table 3).
There were no teacher associations, although the measure in school year 3 was of borderline-adjusted association (MD 0.16 (95% CI −0.01, +0.33)). When the boys and girls were considered separately (Table S5a,b), there were similar significant adjusted associations for hyperactivity at ages 42 and 47 months, and for girls, the teacher report at seven was also significant (Tables 4 and 5).
The DAWBA scales separated hyperactive and attentive be- There were no such associations for the teachers' report for pupils in school year 6 (ages 10-11) (Table S6). When boys were considered separately, all the associations were attenuated, but among the girls,

| Conduct/troublesome behaviour
A scale for identification of conduct problems was included in the SDQ and in two scales included in the DAWBA as asked of the mother and the teacher. There were adjusted associations evident for the children aged 42, 47 and 81 months (Table 3). When the two sexes were analysed separately, it can be seen that for girls all the associations were attenuated (  (Table 5). There were no significant differences between the associations for the two sexes (Tables 4, 5 and S7a,b).

| Emotional behaviour
As with the other scales based on the SDQ or Rutter, seven ages were assessed-five by mothers and two by teachers. All were attenuated on adjustment (Tables S8a,b).

| Total behavioural difficulties
Summation of the hyperactivity, conduct, peer and emotional difficulty scales of the SDQ was used to create a score of total behavioural difficulties. Although all seven measures were associated with maternal paracetamol intake in pregnancy, the associations were attenuated for all except the two early measures (Table 3).
Stratification by sex showed associations on adjustment with the three earliest measures for boys, and the 42-month measure for girls (Tables 4 and 5 and S9a,b). In all these instances, the child whose mother had taken paracetamol had more behavioural problems.

| Principal findings
In this paper, we have investigated 135 continuous outcomes related to cognition, temperament and behaviour in children born to women who had taken paracetamol at some stage between 18 and 32 weeks of pregnancy compared with those who had not. In all, 56 (41%) were associated at the 0.1% level before adjustment. In order to determine appropriate confounders, we used a hypothesis-free exposome technique to identify 15 features of the mothers that independently predicted whether she had taken paracetamol. We then used multiple regression to assess the level of association between paracetamol intake and each of the 56 outcomes after adjustment for the 15 potential confounders; after adjustment 12 of the 56 remained significantly associated. The 12 variables were almost entirely features of hyperactive or attentive behaviour-being less adaptable at 6 months, having poorer persistence at 24 months, having elevated scores on the hyperactive behaviour scales at 42 and 47 months (maternal reports) and teacher reports of poor attention in school year 3 (age 7-8 years). Hands-on testing to determine IQ at age 8 identified a subcategory 'freedom from distractibility' that has an association with attention. 34 The other factors associating with paracetamol were conduct problems at ages 42 and 47 months, and total behaviour difficulties (which include hyperactivity) at these ages. This is not the first time that fetal paracetamol exposure has been linked to hyperactivity, but it is probably the first time that the potential confounders have been identified using an exposome technique; this technique was hypothesis-free, and thus likely to identify some factors that had not previously been considered to be confounders. One other study has used the ALSPAC data to investigate associations between maternal paracetamol intake in pregnancy (both pre-18 weeks and between 18 and 32 weeks) in regard to hyperactivity and other behaviour measures (using the SDQ) at 81 months. They showed significant associations, particularly with taking the drug in the period 18-32 weeks and children with the worst levels of hyperactivity, conduct disorder, emotionality and total behaviour difficulties. Their strategy included taking a number of confounders into account, but also assessing whether paternal consumption of paracetamol, or intake by the mother 5 years later showed similar associations. 11 Our analysis also using these outcomes at 81 months showed that there was an association with conduct disorder, none with emotional symptoms and associations of borderline significance only with hyperactivity and total behavioural difficulties after allowing for the 15 potential confounders (Tables 3 and S4-S9). These differences could have been because of the different choices of factors adjusted for, or a genuine difference between the dichotomy of identifying the worst 10% on the different scales.
Results from other longitudinal studies concerning neurocognitive outcomes have been reviewed by Bauer and colleagues in 2018. 35 They identified nine studies from five cohorts, the Norwegian MoBA and the Danish DNBC being the largest and most informative. There were no two studies that measured the exposure, the confounders or the outcomes in the same way. Nevertheless, the conclusion of the review was that 'associations were strongest for hyperactivity and attention-related outcomes'. This supports our findings.
In order to determine the possible associations of an environmental exposure with an outcome X, a major task is to ensure that the analyses have appropriately adjusted for factors that are associated with the exposure. Here, we have used an exposome technique to model all available ALSPAC data concerning the study mother prior to 32 weeks gestation to identify the environmental features that need to be taken into account when analysing consequences of  We repeated the analyses separately for the two sexes and found that the associations with hyperactive and conduct behaviour were more likely to be found among the boys where there were 14 significant adjusted associations (Table 5), in contrast with the girls with only six significant adjusted associations (

| Strengths of the study
A strength of this study lies in the fact that it concerns a geographically defined population of pregnant women, not selected in any way other than by the place of residence (Avon) and the expected date of delivery (April 1991 to December 1992). Nevertheless, although about 80% of the eligible population took part, those who did not enrol were more likely to be young, to smoke and be from the more deprived social groups. As we have shown, however, these factors were not linked to the likelihood of the woman taking paracetamol in the period 18-32 weeks gestation, and consequently may not be responsible for bias.
Another advantage lies in the fact that the study collected information from a variety of sources including hands-on assessments (eg IQ at 8; the TEACh measures of attention), structured questionnaires completed by the mother (temperament and behaviour), and the teacher (behaviour). Thus, if there were particular biases with the source of data this should be revealed; in fact, most of the positive findings were from maternal report, and the independent hands-on test results were not related to maternal paracetamol intake. This may, however, be a function of the cognitive test rather than revealing a defect in the design.

| Limitations of the data
There are a number of limitations. (a) Although we took time to investigate possible confounders using an exposome technique, the results are necessarily confined to the information collected. There may be many other features that should have been adjusted for (residual confounding), and which may change the conclusions of the paper. (b) Relevant to this is the fact that we used a very statistically stringent approach to identifying the confounders from the data collected by

| Interpretation
Unlike most of the other studies (see review by Bauer et al 35 ), we have used the continuous trait scales rather than a dichotomy identifying pathology. This gives the study more statistical power but, on the other hand, may hide important associations.

| Conclusions
We have shown that paracetamol taken in the period 18-32 weeks is associated with aspects of child attention and hyperactivity until 7 years of age, but there is little sign of adverse associations at later ages, with the exception of their sons who are more likely to demonstrate conduct problems up until 9 years of age. Given the increase in these behaviours it will be important to assess whether they are accompanied by difficulties in scholastic achievements, or whether any adverse effects survive puberty.

ACK N OWLED G EM ENTS
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses.

CO N FLI C T O F I NTE R E S T
The authors confirm they have no conflicts of interest.