Introduction
The developmental origins of mental illness are incompletely understood [
1]. Given that a substantial amount of brain development takes place in-utero, the prenatal period is a time of particular susceptibility to environmental influences [
2]. Numerous specific pregnancy complications such as hypertension, bacterial infection, anaemia, influenza and gestational diabetes mellitus have been associated with different diagnoses in offspring such as psychosis, schizophrenia, bipolar disorder and attention deficit hyperactivity disorder (ADHD) [
3‐
7] but with small effects. Yet these individual pregnancy complications rarely exist in isolation [
8,
9] and as a result, the strength of any associations identified may change if the total number of pregnancy complications (PCs) that occur in-utero is considered.
Studies that have considered total number of complications experienced in a pregnancy in relation to that child’s mental health have revealed mixed findings and tend to include birth complications in their cumulative measure i.e. include exposures that occurred both in-utero and during the birthing process [
10‐
16]. For example, Wagner et al.(2009)found that a higher number of obstetric/neonatal complications did not increase the odds of ADHD or conduct disorder [
15] at 7–8 years (n = 750 twins) but Milberger et al.(1997) found that the total number of complications was positively associated with an ADHD diagnosis among males aged 6–17 years (n = 260) [
16]. Fuchs et al. (2022) found that the total cumulative obstetrical complications predicted Total Problems Score on the Child Behaviour Checklist (CBCL) at 8-years in a very small sample (n = 54) [
10]. In a much larger sample (n ~ 7898), the number of adverse prenatal exposures was associated with the odds of clinically elevated CBCL scores at 9–10 years [
11]. The effect was only significant when two or more adverse prenatal exposures occurred, and a dose-response association existed between the number of prenatal exposures and the odds of clinically elevated CBCL scores. One potential explanation for the mixed findings is the different combination of pre- and perinatal risks that these studies consider. Some of the cumulative scores included birth complications (e.g. gestational age, mode of delivery, low birth weight) however birth complications could be effect modifiers or mediators of the association between complications in pregnancy and child outcomes. In fact, it has been demonstrated that pregnancy complications significantly differed by gestational age [
17]. Therefore, the focus of this study is on complications that occur in-utero
prior to birth.
As to whether the effects of exposure to pregnancy complications on mental health outcomes varies by sex or the socio-economic status of the child is also unclear. Laucht et al. (2000) found no evidence that the association between number of pregnancy complications and child mental health (measured by total-CBCL score at 8-years) differed by sex or adverse family circumstances [
18]. Wagner et al. (2009) reported a stronger association between cumulative perinatal score and ADHD/conduct disorder for females compared to males at 7/8-years [
15], and Nomura et al. (2012) reported that the effect of gestational diabetes on ADHD did differ by socioeconomic status at 4–6 years [
7]. Inflammation markers in pregnancy were found to be associated with internalising problems in females but not males at 9–11 years [
19] whilst bacterial infection in pregnancy (n = 15,421) was more strongly associated with psychosis in males over 40-year period [
4]. Studies examining these interaction effects are however extremely limited in the literature on cumulative pregnancy complications and outcomes in middle childhood (5–12 years).
Furthermore, examining the effects of pregnancy complications on one specific mental health diagnosis at one point in time has limited practical utility in paediatric populations, as children tend to move within, and across, different diagnostic mental health categories [
20‐
23]. A more general measure of observed emotional and behavioural problems may therefore be a better way to capture poor mental health in young childhood compared to specific binary disorders [
24,
25]. The p-factor representing the person’s general psychopathology vulnerability is an important indicator of current childhood psychopathology and a good predictor of later mental health [
23].
The current study explored the association between cumulative pregnancy complications and childhood mental health at 5 and 9-years in a large population-based sample of Irish children. We aimed to investigate: (1) whether a dose-response relationship existed between the total number of pregnancy complications experienced and later mental health problems; (2) whether cumulative pregnancy complications were associated with clinically significant mental health scores; and (3) whether any of these associations differed by sex or adverse social circumstances.
Data analysis
Analyses were performed in STATA (Stata, v.17). For descriptive statistics, a table of the distribution of the covariates across the pregnancy complication categories was created. All analyses were also stratified firstly by sex and subsequently by SR.
To examine the association between pregnancy complications and SDQ-total, we used Generalised linear mixed models (GLMMs) with a random effect of participant ID. This allowed us to account for multiple observations for some participants and an alternative distribution and link function accounted for the positive skew in SDQ-total (i.e. not normally distributed as most children scored low). This meant that participants could be included if they provided an SDQ-score at one time (age 5/9) or both.
We demonstrated, using goodness-of-fit statistics that a mixed-model with gamma-distribution and log-link function was the best fit to the data (Table
S.2a). Stages in model selection were: (1) intercept-only; (2) fixed-effects of pregnancy complications only; (3) random-effects only; (4) fixed and random effects; (5) sex; (6) SR; (7) full model including smoking in pregnancy. The full model demonstrated the best model fit.
Effects are modelled on ‘change in log mean’ but results are transformed and provided as exponentiated coefficients for easier interpretability. Post-hoc comparisons between all possible levels of exposures of interest were conducted to test all possible contrasts. The ‘Scheffe’ adjustment for multiple comparisons was used as it is the most conservative adjustment with widest confidence intervals (CIs) but at a cost of the lowest power. For a gamma distribution all response values must be > 0. Therefore the outcome was winsorized with n = 532 (6.3%) participants being assigned an SDQ-total of 1 rather than 0 (max. possible total = 40). Relevant descriptive statistics were calculated to ensure that this change did not significantly alter the mean, median, standard deviation or inter-quartile range (Table
S.2b).
To examine whether pregnancy complications were associated with clinically significant SDQ-totals, a GLMM specifying binomial distribution and its default link function was used. The stages involved and conclusions in model selection were the same as in aim 1(Table
S.3a). Results are reported as Odds Ratios. The predicted probabilities for being in the clinical range for different combinations of predictors were also calculated (Table
S.4).
To examine whether the effect of pregnancy complications on childhood mental health differed by sex or adverse social circumstance (SR) we added a two-way interaction term (complications*sex; complications*SR) to separate fully-adjusted GLMMs.
Discussion
This study is one of few prospective studies using national cohort data to investigate
cumulative pregnancy complications and child mental health outcomes. We found that children of mothers who experienced more complications had poorer mental health in middle childhood than children of mothers with no complications in a dose-response fashion. The majority of women (78.7%) experienced none or only one pregnancy complication. However, 21.3% of women experienced 2 or more complications, and their children had double to six-fold increased odds of experiencing clinically significant levels of mental health symptoms in middle childhood compared to children whose mothers experienced no complications. This is similar to the finding of Dooley et al. (2023) who found that cumulative number of complications was a predictor of attention problems in a U.S. cohort [
38] and a dose-response relationship was also identified between cumulative number of prenatal exposures and child psychopathology in the U.S. cohort [
11].
There are at least two mechanisms discussed in the literature that could explain the relationship between cumulative complications and child mental health, namely “allostatic load” [
39] and immune/inflammatory processes on the developing brain [
40,
41]. Although we cannot infer a causal relationship and it may not be possible to disentangle the myriad of causal possibilities underlying the association between pregnancy complications and child mental health, it could be considered as a possible prognostic factor for child mental health. Our findings highlight the potential of the cumulative number of complications experienced to be used to screen/identify children at-risk of poorer mental health outcomes who may benefit from early intervention.
This is of public health concern for two reasons. From an economic perspective, the costs of intervening for families at-risk preventatively has been shown to be lower and require less intense treatment that the cost of providing interventions targeting children who already present in the clinical range [
42]. Screening for pregnancy complications and SR at baby developmental appointments may help identify vulnerable individuals during a unique window of opportunity for early interventions, promoting both maternal and child well-being
prior to the onset of mental health difficulties. Secondly, although we know there is a relationship between social-risk and outcomes such as cognition in high-risk groups e.g. preterm birth [
43], our findings highlight a relationship between social-risk and
mental health outcomes at population-level, that is, for
all children.
In fact, both pregnancy complications and social risk exhibited independent but similar magnitudes of association with child mental health. O’ Callaghan et al. (1997) similarly found that maternal age, education, family income and marital relationship at the time of first visit in pregnancy were significantly associated with child behaviour scales at 5-years [
44]. Unlike our findings, their common obstetric and perinatal risk factors did not independently predict child behaviour problems in children at 5-years once adjusted for social disadvantage. However they did not consider them
cumulatively. Our finding that both cumulative complications and SR were both independent and significant risk factors point towards the possibility that low-cost, non-invasive, easily measurable factors in the perinatal period may be useful in terms of predicting and identifying children at-risk of poor mental health outcomes
before symptoms are evident.
Whilst we did not find evidence to support significant quantitative interaction for child sex, similar to other studies [
20], males exhibited a stronger dose-response effect compared to females. Sex-based differences in brain development and responses to insults in-utero are still not fully understood. Some studies have shown associations between placenta size and mental health outcomes in males only [
45] whilst others have shown associations between prenatal predictors and childhood depression/anxiety in females only [
46]. A viability-vulnerability trade-off has been suggested with males possibly being more vulnerable to in-utero in the shorter term but with females possibly adapting and feeling the consequences of this adaptation in the longer term [
47]. This may explain why in our study males were found to be more at-risk as our outcome was measured at 5 and 9-years of age and females have been shown to experience mental health symptoms later in development compared to males [
48‐
52].
It is important to note that although there was increased risk among children exposed to complications, the majority of children (ranging from 80 to 95%) of mothers with complications did not have mental health problems. Further research into determinants of such resilience might be important for designing interventions for children at-risk. Additionally, future research should consider cumulative exposures in pregnancy across biopsychosocial domains not just biological, psychological or social exposures independently. Furthermore, in line with more recent research [
53], the strength of pre-conception experiences as prognostic factors should also be considered as this would allow us to shift the window of intervention even earlier in the life-course.
Strengths & limitations
We used a standardised well-validated instrument (SDQ) as a measure of child’s mental health at two time-points in middle childhood. A 2021 study empirically tested parents potential reporting biases due to their own psychopathology in a large sample and found only minimal evidence for biases in maternal reports of child psychopathology [
54]. Repeated measures at two time-points increases reliability also. Mothers’ report of pregnancy complications has been shown to be reliable [
55,
56] and was recorded at 9-months post-natally, long before the child’s mental health symptoms were measured.
We measured SDQ-total at two ages but did not specifically look at the change of association (if any) between pregnancy complications and SDQ-total at each time-point. We did not explore whether it is a sub-facet of mental health that is related to pregnancy complications i.e. whether there is a specific effect between pregnancy complications and internalising or externalising problems.
Experiencing 3 complications skewed the overall dose-response relationship (Fig.
1). Whilst this could be attributable to unmeasured confounding it is also possible that sample sizes at higher levels of complications and the need to apply the ‘scheffe’ correction for all possible multiple comparisons may have led to power issues at higher levels of complications. The analysis needs to be replicated with a priori hypotheses to examine whether this was a power issue or a particular pattern of risk/resilience.
Whilst we did conduct a sensitivity analysis (S.8) including maternal depression score at baseline, we are cognisant that we did not have information available pertaining to psychiatric family history. Additionally, information about the use of alcohol/substance use in pregnancy is not available in the data-file and thus has not been included.
This is a longitudinal cohort where some people were lost to follow-up. While we weighted the analysis, there may be residual bias.
Conclusion
Research into the heritability of neuropsychiatric disorders suggests that while genetic and epigenetic factors play an important role, the manifestation of these disorders is likely to be multi-factorial, involving pre and/or postnatal insults [
57]. Our findings indicate that pregnancy complications, social risk and sex all independently contributed to mental health outcomes in childhood.
Our findings support that the total number of complications experienced during pregnancy associated with child mental health outcomes in middle childhood in a dose-response fashion even after adjustment for covariates. Our findings indicate that males who experienced 4 + pregnancy complications in the context of social risk were most at-risk of later psychopathology and therefore may need to be prioritised for infant mental health intervention. The combined presence of early life risks such as pregnancy complications, male sex and high social risk post-natally has important implications for both clinical practice and policy.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.