Use of paracetamol, ibuprofen or aspirin in pregnancy and risk of cerebral palsy in the child

Abstract

Background

It has been debated whether mild analgesics, mainly paracetamol, adversely affect aspects of neurodevelopment. We examined whether mother’s use of paracetamol, aspirin or ibuprofen in pregnancy is associated with increased risk of cerebral palsy (CP) in the child.

Method

We included 185 617 mother-child pairs from the Danish National Birth Cohort and the Norwegian Mother and Child Cohort Study. We created harmonized definitions of analgesic use in pregnancy, as well as indications for analgesic use and other potential confounders. Children with CP were identified in nationwide registers. We estimated the average causal effect of analgesics on risk of CP using marginal structural models with stabilized inverse probability weights.

Results

Paracetamol use was reported in 49% of all pregnancies, aspirin in 3% and ibuprofen in 4%. Prenatal exposure to paracetamol ever in pregnancy was associated with increased risk of overall CP [adjusted odds ratio (aOR) 1.3, 95% confidence interval (CI): 1.0–1.7] and unilateral spastic CP (aOR 1.5, 95% CI: 1.0–2.2). The association appeared to be driven by an increased risk of unilateral spastic CP in children exposed in second trimester (aOR 1.6, 95% CI: 1.0–2.5). Children ever prenatally exposed to aspirin in pregnancy had an elevated risk of bilateral spastic CP (aOR 2.4, 95% CI: 1.1–5.3) compared with unexposed.

Conclusion

We observed an increased risk of spastic CP in children prenatally exposed to paracetamol and aspirin. Although we controlled for several important indications for analgesic use, we cannot exclude the possibility of confounding by underlying diseases.

Prenatal exposure, acetaminophen, aspirin, ibuprofen, cerebral palsy, cohort, MoBa, DNBC

Key Messages

• We used harmonized data from two large Nordic birth cohorts to investigate whether prenatal exposure to mild analgesics increases risk of cerebral palsy (CP).

• Children prenatally exposed to paracetamol had increased risk of unilateral spastic CP.

• Children prenatally exposed to aspirin had increased risk of bilateral spastic CP.

• Infection and fever did not appear to fully explain these findings, although we cannot exclude the possibility of residual or unmeasured confounding.

• It remains controversial whether prenatal exposure to paracetamol or aspirin affects neurodevelopment. The safety of these drugs should be further evaluated. Meanwhile, caution regarding use of these analgesics in pregnancy should be given.

Introduction

Pregnant women frequently use mild analgesics bought as over-the-counter drugs, especially paracetamol (acetaminophen), which in Scandinavian countries is used in about half of all pregnancies.^1,2 Use of paracetamol, aspirin (acetylsalicylic acid) and ibuprofen is generally perceived to be safe during pregnancy. Recent studies have suggested an association between prenatal exposure to paracetamol and autism,^3,4 attention-deficit/hyperactivity–like problems,^2,4,5 impaired cognition⁶ and poor gross motor function.² Whereas these results are controversial, they have raised concern about possible effects of paracetamol on other neurodevelopmental outcomes.

Cerebral palsy (CP) is the most common motor disability of childhood, affecting approximately 2 per 1000 children.^7,8 Maternal use of analgesics during pregnancy was investigated in relation to CP in the Extremely Low Gestational Age Newborns (ELGAN) cohort study.⁹ No association was found with paracetamol, although maternal use of aspirin and other nonsteroidal anti-inflammatory drugs (NSAIDs) was associated with increased risk of bilateral spastic CP. A Chinese case-control study also found an increased risk of overall CP with prenatal exposure to NSAIDs.¹⁰ The existing evidence of a link between analgesics and CP is based on either high-risk children, which may limit the generalizability, or on data on analgesic use collected after the diagnosis of CP, which carry the possibility of recall bias. Furthermore, underlying diseases, which are indications for the use of analgesics, may confound the observed associations. We investigated whether use of paracetamol, aspirin or ibuprofen in pregnancy is associated with increased risk of CP, including its unilateral and bilateral spastic subtypes that are assumed to have partly distinct aetiologies, using data from a large cohort study in which most data had been collected prospectively during pregnancy.

Methods

Study population

We used data from the MOthers and BAbies in Norway and Denmark (MOBAND) cohort,¹¹ comprising pooled data from the Danish National Birth Cohort (DNBC)¹² and the Norwegian Mother and Child Cohort Study (MoBa).¹³ Pregnant women were recruited to DNBC in 1996–2002 around gestational week (gw) 6–12, and to MoBa in 1999–2008 around gw 17–18.

Information on exposures during pregnancy was collected in DNBC in the enrolment form and with computer-assisted telephone interviews at gw 16 and 31. In MoBa, these data were collected by self-administered questionnaires at gw 17 and 30. Exposures in gw 30 and later were covered by data collections 6 months after delivery, in both sub-cohorts. The pooled MOBAND cohort included 212 926 pregnancies. We excluded multiple pregnancies (n = 4062), pregnancies without information from the earliest data collection (n = 20 461) and pregnancies ending in fetal (n = 2257) or infant deaths (n = 529). This led to a study population of 185 617 mother-child pairs (DNBC n = 86 465, MoBa n = 99 152) (Supplementary Figure 1, available as Supplementary data at IJE online).

Exposure to analgesics

The questions about analgesic use differed in DNBC and MoBa, and have been described in detail elsewhere.^1,2 Briefly, the women in DNBC were asked about drug use in an open-ended question on the enrolment form, and about analgesic use at the second and third interview. The second interview also included questions on drugs taken to treat fever, inflammation, infection and muscle and joint diseases. All drugs were reported on a week-by-week basis. In MoBa, each questionnaire contained a list of specific diseases. For each disease, the women were asked to specify when they had experienced the disease and the drugs used to treat the disease. In additional questions, women had the opportunity to report drugs not already stated in relation to specific diseases. Timing of drug use was reported in gw intervals (0–4, 5–8, 9–12 etc.).

We created harmonized definitions of exposure to paracetamol (Anatomical Therapeutic Chemical (ATC) Classification System: N02AA59, N02BE01, N02BE51, N02BE71), aspirin (ATC: N02BA01, N02BA51, N02BA71, B01AC06), and ibuprofen (ATC: M01AE01, M02AA13, M01AE51). We assessed ‘ever use’ of each drug in pregnancy, and also the use of paracetamol by trimester. Women were classified as ‘ever users’ if they reported use in any of the data collections, and considered as ‘never users’ if they indicated no use in all data collections. Women who did not complete all data collections and did not report use in the data collections they participated in were assigned missing values for exposure ever in pregnancy. When we defined trimester-specific exposure, we assigned missing values to the 13% of women in DNBC who were unable to recall when they used analgesics. Conversely, the structure of the MoBa questionnaires made it impossible to determine whether no reporting of analgesics was explained by no use, or by women being unable to recall use. Thus, all women from MoBa who did not report any analgesic use were categorized as unexposed.

Cerebral palsy

The Danish National Cerebral Palsy Registry includes information on all children with a diagnosis of pre- and perinatal acquired CP verified by a neuro-paediatrician from 1 to 6 years of age.¹⁴ In Norway, there is a similar Cerebral Palsy Registry⁸ with a coverage rate of 76%. The remaining children with CP were identified through record linkage with the Patient Registry of Norway; only children with CP verified by a neuro-paediatrician were included.¹⁵ We assessed overall CP and the two major subtypes; unilateral spastic CP and bilateral spastic CP, which presumably may have different aetiologies.

Statistical analyses

We used marginal structural models with stabilized inverse probability weights as this allowed us to estimate the average causal effect of analgesics on risk of CP and also to separate the effect of analgesic use in the different trimesters.¹⁶ The weights were calculated as Pr[A]/Pr[A|L], where Pr denotes probability, A exposure status and L possible confounders. Selection of potential confounders we adjusted for was guided by use of directed acyclic graphs.¹⁷ These variables included in vitro fertilization (IVF) treatment (no, yes) and child’s birth year derived from the Danish IVF Registry and the Medical Birth Registries of Denmark and Norway, and self-reported measures of maternal occupational status (employed, unemployed, student, receiving social benefits or pension), pre-pregnancy body mass index (<18.5, 18.5–24.9, 25–29.9, ≥30), number of cigarettes smoked per day (0, 1–9, ≥10), use of the other mild analgesics than the index exposure (no, yes), respiratory infection (no, yes), fever (no, yes) and urinary tract infection (no, yes). In the analysis for trimester-specific exposure, we calculated weights using the product of the probabilities estimated separately for the exposure at each trimester. The probabilities of exposure in a particular trimester were based on information about use of other mild analgesics than the index exposure, respiratory infection and fever in previous and current trimesters. The remaining confounders used in these predictions were not available by trimester. Estimates were computed as odds ratios (OR) for CP in the marginal structural models (logistic regression models). We used robust variance estimators to calculate 95% confidence intervals (95% CIs) to take the weights and dependency between siblings into account. Missing information on both exposure and covariates was imputed using Substantive Model Compatible FCS Multiple Imputation (Supplementary Method 1, available as Supplementary data at IJE online).

We addressed potential confounding by indication in several ways. We directly adjusted for respiratory infection, fever and urinary tract infection. We also compared children prenatally exposed to paracetamol with unexposed children who were (or were not) exposed to respiratory infection, fever or urinary tract infection. We modelled the joint effect of paracetamol and these indications, using children exposed to neither as reference. Analyses additionally adjusted for maternal rheumatoid arthritis (no, yes) were also performed.

We investigated the effect modification of paracetamol use and child’s sex to explore sex differences. We also stratified by cohort affiliation to assess differences between DNBC and MoBa. Analyses were repeated restricting to children with no missing data, and using conventional outcome regression, and we furthermore performed analyses adjusted for maternal occupational status alone in order to explore to what extent socioeconomic status confounds the association. Finally, we assessed the association of maternal use of analgesics in first trimester with fetal and infant death (dying within first year of life), to assess competing risks of death.¹⁸

Results

Use of analgesics during pregnancy

Paracetamol was used in 49% of pregnancies, aspirin in 3% and ibuprofen in 4%. Analgesics were most often used in the first and second trimester, and use of an analgesic in a specific trimester was often followed by use of the same analgesic in subsequent trimesters, and with co-use of other analgesics. More women in DNBC used paracetamol and aspirin, and more women in MoBa used ibuprofen. Use of any type of analgesic was associated with being out of the workforce, smoking, urinary tract infection and fever. For further details see Table 1.

Table 1

Analgesic use and baseline characteristics of the study population

		Paracetamol	Aspirin	Ibuprofen
Exposure status ever in pregnancy		n (%)	n (%)	n (%)
No		71 345 (38.4)	138 763 (74.8)	137 138 (73.9)
Yes		91 015 (49.0)	5746 (3.1)	7358 (4.0)
Missing		23 257 (12.5)	41 108 (22.2)	41 121 (22.2)
Characteristics	n (%)	Pct. exposed in MI data	Pct. exposed in MI data	Pct. exposed in MI data
Entire study population	185 617 (100)	56.5	4.4	5.5
Maternal age
≤24 years	19 318 (10.4)	59.3	3.8	6.8
25–29 years	66 037 (35.6)	56.9	4.2	5.3
30–34 years	70 268 (37.9)	56.1	4.5	5.4
≥35 years	29 994 (16.2)	54.7	5.3	5.5
Missing	0 (−)
Maternal occupational status
Employed	144 997 (78.1)	55.6	4.3	5.3
Unemployed	13 785 (7.4)	61.1	5.4	6.0
Student	22 083 (11.9)	58.0	4.8	6.4
Receiving benefits or pension	2692 (1.5)	70.2	6.3	9.3
Missing	2060 (1.1)
BMI
<18.5	6702 (3.6)	54.3	5.6	5.3
18.5–24.9	120 415 (64.9)	54.4	4.5	5.1
≥25	54 341 (29.3)	64.5	4.1	6.5
Missing	4159 (2.2)
IVF treatment
No	181 261 (97.7)	56.6	4.4	5.6
Yes	4356 (2.4)	52.4	4.7	2.9
Missing	0 (−)
Cigarettes smoked per day
0	162 280 (87.4)	55.0	4.1	5.2
1–9	14 078 (7.6)	64.4	6.0	7.4
≥10	8366 (4.5)	72.3	8.2	8.6
Missing	893 (0.5)
Respiratory infectiona
No	61 994 (33.4)	49.8	4.6	4.0
Yes	78 412 (42.2)	62.6	4.3	6.9
Missing	45 211 (24.4)
Urinary tract infection
No	128 276 (69.1)	54.2	4.0	5.1
Yes	21 158 (11.4)	66.2	7.0	8.1
Missing	36 183 (19.5)
Fevera
No	131 497 (70.8)	53.0	4.2	5.4
Yes	32 078 (17.3)	73.3	5.7	6.4
Missing	22 042 (11.9)
Use of other analgesicsb
No	–	54.6	3.3	2.5
Yes	–	74.5	5.3	7.8

Pct, percent; MI, multiple imputation of missing values.

^aTrimester-specific measures for respiratory infection and fever, respectively, are combined because the patterns of analgesic use according to respiratory infection and fever were similar in the different trimesters.

^bThe definition of ‘use of other analgesics’ depends on which analgesic is being assessed, e.g. when assessing paracetamol, the measure is defined as use of aspirin and/or ibuprofen.

Prenatal exposure to analgesics and risk of cerebral palsy

There were 357 children with CP, of whom 38% had unilateral and 47% had bilateral spastic subtypes. Compared with unexposed, children ever exposed to paracetamol in pregnancy had higher risk of overall CP [adjusted odds ratio (aOR) 1.3, 95% CI: 1.0–1.7] and unilateral spastic CP (aOR 1.5, 95% CI: 1.0–2.2). The increased risk of unilateral spastic CP in children exposed to paracetamol was driven by use in the second trimester (aOR 1.6, 95% CI: 1.0–2.5) (Table 2). From the estimated aOR, a risk difference of 3.5 per 10 000 live births was obtained, corresponding to number needed to harm = 2825. There was no suggestion of sex-specific differences (Supplementary Table 1, available as Supplementary data at IJE online). The risk of overall CP and unilateral spastic CP increased with the number of trimesters exposed to paracetamol (P-value for trend 0.03 and 0.01, respectively). However, trends were less evident after adjustment for potential confounders (P-value for trend 0.2 and 0.1) (Supplementary Table 2, available as Supplementary data at IJE online). Children ever exposed to aspirin in pregnancy had increased risk of bilateral spastic CP (aOR 2.4, 95% CI: 1.1–5.3) compared with unexposed. From the estimated aOR, a risk difference of 12.1 per 10 000 live births was obtained, which corresponds to number needed to harm = 830. The crude estimate for ibuprofen and bilateral spastic CP suggested an association, but the association disappeared after adjustment (Table 2).

Table 2

Estimated average causal effect of prenatal exposure to analgesics on risk of cerebral palsy

Prenatal exposure to analgesics		Cerebral palsy (CP)
		All CP (n = 357)			Unilateral spastic CP (n = 135)			Bilateral spastic CP (n = 169)
		n cases/ non-casesa	ORb (95% CI)	ORc (95% CI)	n cases/ non-casesa	ORb (95% CI)	ORc (95% CI)	n cases/ non-casesa	ORb (95% CI)	ORc (95% CI)
Paracetamol
Ever in pregnancy	No	107/71238	1 (ref.)	1 (ref.)	41/71304	1 (ref.)	1 (ref.)	50/71295	1 (ref.)	1 (ref.)
	Yes	185/90830	1.45 (1.15–1.83)	1.34 (1.04–1.71)	72/90943	1.47 (1.01–2.13)	1.47 (0.98–2.20)	83/90932	1.37 (0.96–1.95)	1.32 (0.92–1.90)
1st trimester	No	220/128723	1 (ref.)	1 (ref.)	81/128862	1 (ref.)	1 (ref.)	108/128835	1 (ref.)	1 (ref.)
	Yes	108/48392	1.33 (1.06–1.67)	1.13 (0.86–1.49)	44/48456	1.48 (1.02–2.13)	1.03 (0.66–1.61)	47/48453	1.16 (0.82–1.64)	1.07 (0.71–1.60)
2nd trimester	No	158/103677	1 (ref.)	1 (ref.)	59/103776	1 (ref.)	1 (ref.)	74/103761	1 (ref.)	1 (ref.)
	Yes	101/50920	1.37 (1.08–1.74)	1.38 (1.04–1.84)	45/50976	1.63 (1.13–2.35)	1.61 (1.04–2.50)	40/50981	1.21 (0.84–1.72)	1.35 (0.87–2.10)
3rd trimester	No	208/122404	1 (ref.)	1 (ref.)	80/122532	1 (ref.)	1 (ref.)	96/122516	1 (ref.)	1 (ref.)
	Yes	43/28628	0.90 (0.66–1.22)	0.71 (0.49–1.02)	22/28649	1.20 (0.76–1.90)	0.91 (0.53–1.57)	16/28655	0.70 (0.42–1.14)	0.47 (0.26–0.85)
Aspirin
Ever in pregnancy	No	220/138543	1 (ref.)	1 (ref.)	93/138670	1 (ref.)	1 (ref.)	92/138671	1 (ref.)	1 (ref.)
	Yes	14/5732	1.95 (1.15–3.31)	1.79 (0.91–3.53)	2/5744	NE	NE	9/5737	2.94 (1.62–5.34)	2.43 (1.11–5.33)
Ibuprofen
Ever in pregnancy	No	219/136919	1 (ref.)	1 (ref.)	93/137045	1 (ref.)	1 (ref.)	91/137047	1 (ref.)	1 (ref.)
	Yes	15/7343	1.61 (0.93–2.80)	1.04 (0.57–1.89)	5/7353	1.12 (0.45–2.83)	0.70 (0.25–1.93)	8/7350	2.22 (1.00–4.93)	1.29 (0.53–3.19)

Trimesters are defined as: ≤12th, 13^th–28th, ≥29th week of gestation.

NE, no estimate.

^aNumber of cases/non-cases in complete-case data.

^bCrude estimates.

^cMarginal structural model with stabilized inverse probability weighs calculated from birth year, maternal occupational status, body mass-index, IVF treatment, smoking, use of other mild analgesics, respiratory infection, fever and urinary tract infection in pregnancy. For trimester-specific exposures, information on use of other analgesics, respiratory infection and fever in previous or same trimesters as index exposures were used.

Confounding by indication

The risk of unilateral spastic CP following exposure to paracetamol was higher in children who also had been exposed to infection or fever in pregnancy than in those not exposed to infection or fever in pregnancy (aOR 1.8, 95% CI: 1.0-3.1 vs aOR 1.2, 95% CI: 0.6-2.3), although this may have been due to chance (test for interaction P-value 0.4). Prenatal exposure to infection or fever without exposure to paracetamol was not associated with increased risk of unilateral spastic CP (aOR 0.8, 95% CI: 0.4–1.5) (Table 3). Findings were similar when we assessed each indication for analgesic use separately (Supplementary Table 3, available as Supplementary data at IJE online). Moreover, results from analyses additionally adjusted for rheumatoid arthritis did not differ from our main findings (Supplementary Table 4, available as Supplementary data at IJE online).

Table 3

Estimated average causal effect of prenatal exposure to paracetamol on risk of unilateral spasticity according to indications for analgesic use

Exposure ever in pregnancy		Unilateral spastic cerebral palsy (CP) (n = 135)
		n cases/non-casesa	ORb (95% CI)	ORb (95% CI)
Paracetamol	Indicationsc
No	No	23/34331	1 (ref.)	1 (ref.)
Yes	No	27/35118	1.20 (0.64–2.25)	1.20 (0.64–2.25)
No	Yes	18/36973	0.78 (0.40–1.51)	1 (ref.)
Yes	Yes	45/55825	1.39 (0.85–2.28)	1.78 (1.03–3.10)
Test for interaction				P = 0.36

^aNumber of cases/non-cases in complete-case data.

^bMarginal structural model with stabilized inverse probability weighs calculated from birth year, maternal occupational status, body mass-index, IVF treatment, smoking and use of other mild analgesics.

^cIndications are defined as the presence of respiratory infection, fever or urinary tract infection in pregnancy.

Assessment of other potential sources of bias

As expected, when stratifying by cohort affiliation, the estimates for the association between paracetamol and unilateral spastic CP were less precise. The estimates for exposure in the first and second trimesters, but not in the third trimester, appeared larger in DNBC than in MoBa. Adjustment led to reduced effect sizes of exposure in the first and third trimester in both sub-cohorts. The estimate for exposure ever in pregnancy was higher in DNBC than in MoBa, but this could be due to chance (test for interaction P-value 0.7) (Supplementary Table 5, available as Supplementary data at IJE online). We were unable to examine aspirin according to the risk of bilateral spastic CP stratified by cohort affiliation because there were only two exposed cases in MoBa. Results from analyses restricted to observations with no missing values and results from outcome regressions did not differ markedly from our main results (Supplementary Tables 6 and 7, available as Supplementary data at IJE online), and estimates adjusted for maternal occupational status alone were quite similar to the crude estimates (Supplementary Table 8, available as Supplementary data at IJE online). Use of any of the analgesics was associated with fetal or infant death, but adjustment for possible confounding factors almost removed the associations except for aspirin, which was reduced to OR = 2.1 (95% CI: 1.6–2.8) (Supplementary Table 9, available as Supplementary data at IJE online).

Discussion

About half of the mothers in MOBAND used paracetamol in pregnancy. Children exposed to paracetamol in second trimester had higher risk of unilateral spastic CP compared with unexposed children. Aspirin was not as commonly used as paracetamol. Nonetheless, our findings suggest that children prenatally exposed to aspirin have elevated risk of bilateral spastic CP compared with unexposed children. Analyses suggest that the increased risks found are not fully explained by confounding by infection or fever.

Possible mechanisms

The increased risk of unilateral spastic CP in children exposed to paracetamol in the second trimester is biologically plausible since white matter injuries predominantly occur after gw 24,¹⁹ Several studies have suggested an adverse effect of prenatal paracetamol exposure on neurodevelopment, though the findings remain controversial.^2–6,20 Possible mechanisms whereby paracetamol may affect brain development include neurotoxicity induced by oxidative stress^21,22 or low sulphating capacity,^23,24 interaction with maternal hormones that regulate normal brain development^25,26 and stimulation of endocannabinoid receptors required for normal axonal growth and fasciculation.^27,28 Both paracetamol and aspirin are found to disrupt fetal testicular function and thereby the production of androgens.^1,29 This could potentially affect risk of CP, since androgens are known to influence neurodevelopment.³⁰ We would then expect to see a higher risk of CP in boys than in girls, which we did not for paracetamol, whereas the low number of CP cases exposed to aspirin prevented us from examining potential sex differences. Aspirin has anti-thrombotic properties, but haemorrhage is unlikely to explain the association of aspirin with increased risk of bilateral spastic CP, since vascular events mainly are associated with unilateral spastic CP.³¹

Comparison with other studies

Previous MoBa studies have suggested that long-term use of paracetamol in pregnancy, i.e. 28 days or more, is associated with poor child neurodevelopment.^2,32 Information on number of days the women used analgesics was available only in MoBa. As a proxy for dose, we used number of trimesters with exposure to paracetamol. Despite the crudeness of the measure, and adjustment weakening the trend, there was still evidence of a dose-response correlation of paracetamol use and both overall CP and unilateral spastic CP. In contrast to our findings, prenatal exposure to paracetamol was not associated with CP in the ELGAN cohort study.⁹ A possible explanation for the null finding could be that the ELGAN cohort comprised 877 very preterm infants mainly with bilateral spastic CP (84 out of 100 CP cases), and unilateral and bilateral spastic CP to a large extent have different aetiologies.^33,34

In our study, aspirin was associated with increased risk of bilateral spastic CP. A weaker association between ibuprofen and bilateral spastic CP appeared to be caused by confounding. In the ELGAN cohort,⁹ prenatal exposure to aspirin and other NSAIDs were also associated with bilateral spastic CP (aOR: 3.0, 95% CI: 1.3–6.9 and 2.4, 95% CI: 1.0–5.8). A Chinese case-control study of 145 children with CP and 290 controls found an association between maternal use of NSAIDs in pregnancy and overall CP, though the estimate was highly imprecise.¹⁰ In contrast, the Etude Epidemiologique des Petites Ages Gestationnels (EPIPAGE) cohort study did not find an association between low-dose aspirin administration and overall CP in 656 preterm infants.³⁵ There is evidence that low-dose aspirin prescribed to high-risk pregnant women prevents preeclampsia, preterm birth, and intrauterine growth restriction,³⁶ all of which are risk factors for CP.³⁷ The discrepancy between findings from the EPIPAGE cohort and our study might be due to the fact that the EPIPAGE cohort was restricted to high-risk pregnancies.

Confounding by indication

Confounding by indication is an important concern in any study of medication effects. Urinary tract infection and fever have previously been associated with increased risk of CP in DNBC.^38,39 Our findings suggest that infection or fever is not associated with unilateral spastic CP among children unexposed to paracetamol. This indicates urinary tract infection and fever per se may not induce CP; at least in the milder spectrum of disease where women do not expose themselves to analgesics. Similarly, the effect of paracetamol on unilateral spastic CP does not appear to be fully attributable to confounding by infection or fever. However, we cannot exclude the possibility of residual confounding by illness severity, since women who are most ill may be more likely to use analgesics, which might explain the association between paracetamol and CP in children also exposed to infections or fever.

We were able to adjust for some but not all relevant indications for analgesic use. E.g., only MoBa had information on headache, migraine and triptan use, which are both strong predictors for paracetamol use (Supplementary Table 10, available as Supplementary data at IJE online) and suspected risk factors for perinatal stroke.⁴⁰ Other relevant indications for analgesic use for which we were unable to harmonize data included pelvic instability and musculoskeletal pain. Moreover, MoBa presumably provides better information on indications for analgesic use than DNBC, since use of analgesics was reported in relation to specific diseases. Nonetheless, adjustment led to same magnitude of change in estimates in the two sub-cohorts. We did not adjust the analyses for preeclampsia, preterm birth or small for gestational age, because these factors are likely to share pathological causes with CP. A growing literature has shown that adjustment for such variables can cause major distortions due to collider stratification bias.^41,42 Moreover, these factors may be mediators on the causal path from aspirin to CP. To include them in the analyses could distort the total effect we are interested in.⁴³

Strengths and limitations

We have combined two large birth cohorts with data on exposures prospectively collected during pregnancy, which minimizes potential recall bias. Over-the-counter drugs, infections, and fever are not recorded on regular basis in nationwide registries, but are available in the MOBAND cohort through interviews and questionnaires. This has provided a rare opportunity to investigate the impact of analgesics in a follow-up design, and to address confounding by indication. Another strength is that record linkages were performed using national population-based CP registers, with verified CP diagnoses and nearly-complete follow-up.

We estimated the effects of analgesic use on fetal or infant death and found that exposure to aspirin in early gestation was moderately associated with dying between conception and 1 year of age. Thus, there is a possibility of underestimation risk of CP, if children with brain damage compatible with CP are also more likely to die before CP can be diagnosed, but such negative bias is likely to be small.¹⁸

Non-differential misclassification of exposure may have occurred more often in MoBa because the structure of the questionnaires required us to code women unable to recall when they used analgesics as ‘unexposed’. This presumably would bias the estimates towards the null. In contrast, such women were assigned with missing values in DNBC data, and exposure status was predicted for these women by use of multiple imputation.

Our analyses were adjusted for ‘maternal occupational status’ as analgesic use in pregnancy is highly depended on socioeconomic status and there also seems to be a socioeconomic gradient in risk of CP.⁴⁴ Other socioeconomic variables, for which we were unable to harmonize data, might have been more suitable to limit residual confounding by socioeconomic position. Self-selection into the DNBC and MoBa has resulted in populations that on general are healthier and socially more advanced than the general populations of Denmark and Norway. This results in a population relatively homogenous in terms of socioeconomic position, which also reduces the possibility of residual confounding by socioeconomic status,¹¹ and it was reassuring to experience that ‘maternal occupational status’ only was a weak contributor to the weakening of estimates we saw with the full adjustment.

We used stabilized inverse probability weights in marginal structural models to estimate the average causal effect of analgesic use on risk of CP. This is a valid method to deal with time-dependent confounding, but controlling for indications for analgesic use will lead to different estimates than conventional outcome regression only when treatment in a trimester affects the likelihood of use of other medications, infection, or fever in subsequent trimesters.¹⁶ Nonetheless, the method allowed us to determine the separate effect of exposure to paracetamol in each trimester, which is not possible with conventional outcome regression.

Conclusion

This study suggests that children exposed to paracetamol in second trimester have increased risk of unilateral spastic CP. Similarly, children exposed to aspirin have increased risk of bilateral spastic CP. No elevated risk of CP was found with exposure to ibuprofen after adjustment. Although, our findings indicate that the increased risks of CP cannot be fully explained by confounding by infection or fever, we cannot eliminate the possibility of residual confounding due to severity of diseases. Moreover, unmeasured confounding is especially a concern for aspirin, since aspirin is used to treat pregnancy complications that are associated with risk of CP. Before causal inferences can be drawn, our findings should be replicated in other large studies based on prospective collected data with possibilities of addressing confounding by indication, and more insight from animal studies on the mechanism for the negative effect on brain development is required.

Supplementary Data

Supplementary data are available at IJE online.

Funding

This work was supported by The Faculty of Health and Medical Sciences at University of Copenhagen [grant no 620–0032/14–3000 to T.G.P]; Bevica Foundation and Ludvig and Sara Elsass Foundation [grant no. 5269 to T.G.P], and partly supported by the Intramural Program of the National Institute of Environmental Health Sciences, NIH, USA and by the NIH/NIEHS career development award [grant no. K99ES026729 to Z.L].

Acknowledgements

We thank the participating families in Denmark and Norway who took part in the DNBC and MoBa. The Norwegian Mother and Child Cohort Study are supported by the Norwegian Ministry of Health and Care Services and the Ministry of Education and Research, NIH/NIEHS (contract no N01-ES-75558), NIH/NINDS (grant no.1 UO1 NS 047537–01 and grant no.2 UO1 NS 047537–06A1). We would like to express our thanks to the MOBAND scientific advisory board: Nigel Paneth from Department of Epidemiology and Biostatistics at Michigan State University, Eve Blair from Telethon Kids Institute at the University of Western Australia and Karin B. Nelson from the National Institute of Neurological Disorders and Stroke at the National Institute of Health, U.S. for constructive comments to the manuscript.

Author Contributions

MOBAND was initiated in 2011 by senior investigator A.W. from the National Institute of Environmental Health Sciences in the United States and has been established by him in collaboration with the University of Copenhagen, Aarhus University, the Danish Cerebral Palsy Registry, the Norwegian Institute of Public Health, the University of Bergen, and the Cerebral Palsy Registry of Norway. Data harmonization was led by Dag Moster and conducted by Ingeborg Forthun and M.C.T. at the University of Bergen, and by K.S-L. and T.G.P. at the University of Copenhagen, under the supervision of A.W.

Conflict of interest: None declared.