Background
Congenital heart diseases (CHDs) are the most common congenital anomaly, affecting around 1% of live births and 10% of stillbirths [
1,
2]. CHDs are a leading cause of childhood mortality and many CHD patients experience health problems that persist into adulthood [
3,
4]. The causes of CHDs are largely unknown, but the pregnancy environment (intrauterine factors) may play a role in the underlying pathophysiology [
5]. Identifying modifiable risk factors for CHDs is important for improving aetiological understanding and developing preventive interventions to reduce disease burden.
Several modifiable maternal characteristics have been found to be associated with increased risk of CHDs, including maternal pre/early pregnancy body mass index (BMI) [
6‐
8], smoking [
9] and alcohol [
10] consumption in pregnancy. The causal relevance of the results from meta-analyses is unclear, due to many studies not controlling for key confounders and for the risk of residual confounding. Previously, using parental negative exposure control analyses, we found that positive associations between maternal overweight and obesity with offspring CHDs may be being driven by confounding factors [
11]. This work found some evidence of an intrauterine effect of maternal smoking on offspring CHDs. For alcohol consumption, results were inconclusive due to limited data [
11]. Negative control analyses attempt to address the issue of residual confounding in observational studies [
6,
11,
12], but rely on assumptions that cannot be empirically verified, such as it being implausible that the exposure in the father (e.g. their smoking) could influence offspring CHD risk to a similar magnitude of any effect in mothers.
Mendelian randomisation (MR) uses genetic variants as instrumental variables (IVs) to test causal effects in observational data [
13]. The key assumptions for MR are (i) relevance assumption—the genetic instruments are robustly associated with the exposure, (ii) independence assumption—there is no confounding of the genetic instrument-outcome association, (iii) exclusion restriction criteria—the genetic variant is not related to the outcome other than via its association with the exposure [
14]. Genetic variants are less likely to be confounded by the socioeconomic and environmental factors that might bias causal estimates in conventional multivariable regression [
15], but may be biased by violation of their assumptions due to weak or irrelevant instruments, population stratification (causing confounding of the genetic instrument-outcome association) and a path from the genetic instrument to CHD not mediated by the exposure, for example via horizontal pleiotropy or foetal genotype [
16]. Triangulating results from negative control and MR analyses, whereby the key sources of bias differ can help improve the causal understanding of maternal risk factors on CHDs [
17]. Consistent results from both would increase confidence that the relationship is causal. The recent acquisition of genotype information on a large number of maternal-offspring dyads means that we now have relevant data to further test the potential effects of BMI, smoking and alcohol with a complementary method to those used previously. The objective of this study was therefore to explore associations between genetically predicted maternal BMI, smoking and alcohol on offspring CHD using Mendelian randomisation.
Discussion
Using MR across three birth cohorts, we found no strong evidence for an effect of genetically predicted maternal BMI, smoking or alcohol on risk of offspring CHD. However, for all three exposures, confidence intervals were wide and the pseudo R2 and AUC suggested potential weak instrument bias for alcohol and smoking. Weak instruments in this study would be expected to bias results toward the confounded association. Weak instruments and imprecise associations also limit clear interpretation of our sensitivity analyses to explore bias due to GRS influencing CHDs via other paths independently of the exposure of interest. We tried to identify all cohorts with maternal genetic data and offspring, CHD measures and to the best of our knowledge, this is the first MR study of these maternal exposures on offspring CHD risk. Despite the relatively large sample, our inconclusive findings highlight the importance of existing and new cohorts, many of which have genomic data, linking to health care records to obtain information on CHDs and other rare outcomes, for example through electronic health records.
This MR study complements our previous negative paternal control study [
11]. Our MR analyses of BMI are consistent with our previous negative control study, in suggesting that higher maternal BMI may not causally influence offspring CHD and that previous multivariable regression analyses [
6,
8] were likely confounded. We have not clearly replicated our previous result for smoking, which suggested an increased risk of offspring CHD in women who smoked in pregnancy. However, as noted above our imprecise MR results do not rule out an effect, and future larger MR studies are important. Due to the lack of information on alcohol consumption around the time of their partners pregnancy, previous analyses using a negative control design were inconclusive [
11]. Recent meta‐analyses found consistently modest increases in risk of offspring CHD in mothers reporting alcohol consumption in pregnancy, however, many of the included studies did not adjust for confounders [
10,
56], meaning that it is difficult to determine whether the association is a result of alcohol or other characteristics that are related to alcohol and offspring CHDs. In the present study, the results for alcohol were inconclusive, although notably confidence intervals of the pooled effect (pooled OR: 1.06 (0.98, 1.15)) did not rule out an association further emphasising the need for future larger studies.
There are several strengths of the current study. We attempted to identify all studies with relevant data knowing that MR is statistically inefficient, and CHD is a rare outcome. We explored potential bias due to other paths from the GRS to CHDs by examining associations of each GRS with the other two exposures and with other risk factors that might influence CHD and undertook multivariable MR where such associations were found. We also adjusted for offspring genotype in a subsample of the pooled data cohort, which is important in attempting to separate the influence of a path from GRS to CHDs via foetal genotype rather than solely from the mothers’ exposure [
16].
The key limitation of this study is that despite a relatively large sample size (
N = 65,510,
N = 562 CHD cases) the effect estimates were imprecise due to CHD being a rare condition. In knowing this, we explored a collaborative base of birth cohorts and searched the literature for any additional cohorts that might contribute but found none that were eligible for inclusion. Furthermore, the GRS for smoking and alcohol may have been biased by the weak instruments. These limitations importantly contribute to our main and sensitivity analysis results. We were not able to clearly differentiate between horizontal (i.e., where smoking is the main exposure of interest, and an association of the smoking GRS with BMI reflects an independent path) and vertical pleiotropy (i.e. where the relation of the smoking GRS with BMI is downstream of the GRS effect on smoking). We want to adjust for horizontal pleiotropy but not vertical pleiotropy, as the latter would be adjusting away part of the mechanism by which, for example smoking might influence CHD. Adjustment for offspring genotypes was only possible in a subsample of the main analysis, making results more imprecise and prone to selection bias. However, it is encouraging that our results do not notably differ in these analyses. MR results may be biased by population stratification confounding. We tried to mitigate against that by adjusting for ancestry PCs and exploring the consistency of the main results with results removing BiB. Largely these were consistent but even more imprecise. Whilst we included all participants, including those from non-European ancestries, both MoBa (the largest contributing study) and ALSPAC participants are mostly of White European origin, and the GWAS data used to construct the GRSs were restricted to European participants. Therefore, our results may not generalise to other populations. By examining gene associations, without estimating causal effects, as we have done here (i.e. not adopting a formal IV framework), the three IV assumptions that we discuss need to be considered. However, with this approach, the fourth assumption, which is often ignored even in MR studies that do attempt to estimate causal effects, is not required. The fourth assumption which often receives less attention states that the effect of the exposure on the outcome may differ for different people [
57,
58].
We were only able to explore associations of GRS with any CHD (analyses of subtypes would have been very imprecise) and therefore could have missed the potential effects of these exposures on specific CHD subtypes. Nevertheless, we believe there is value for prospective parents, clinicians and policy makers in knowing the effects of any CHD. MoBa cohort only had cases diagnosed antenatally or around the time of birth (first year of life) obtained from a single source (The Medical Birth Registry of Norway) which would increase the chances of outcome misclassification by assigning CHD cases which were diagnosed later in life as non-CHD cases. This misclassification is likely to be random with respect to the GRS (i.e., later age at offspring diagnosis could not influence mothers genotype) and would be expected to bias results towards the null, meaning we may have missed some associations. A previous study using MoBa data found a larger proportion of CHD cases than the present study (1.39% vs 0.83%) via ascertainment of linked health records [
59]. We were unable to access these data for the present study. Potential reasons for the difference in birth prevalence could be firstly that they had access to more detailed ICD-coded data, and secondly because the MoBa genotype project datasets that we used in the present study had different inclusion criteria and therefore the two study populations are not directly comparable.
Identifying modifiable causal risk factors for the development of CHD is important for developing preventive interventions to reduce the risk of CHDs. Improvements in surgery over the last two decades mean that most patients with CHD now live into adulthood. Nonetheless, prevention remains important. Many patients require repeat procedures through childhood and adolescence to accommodate their growth, which produces a burden on them, their family and society. Despite trying to identify all relevant studies our results are inconclusive. They highlight the need for more data with maternal genetic and offspring CHD data. We think this is possible over the coming years as running GWAS is relatively cheap, and most cohort studies increasingly have these data. The following could considerably increase the sample size for MR in this field and result in key advances in preventing CHDs: (i) Add data on CHDs through electronic record linkage to existing cohorts; this was done recently in ALSPAC nearly 30 years after the original pregnancies [
37]. Many of the cohorts that we considered for inclusion had genetic data but no information on CHD (or other congenital anomalies). (ii) Linkage to electronic records should be regularly updated at least until early adulthood so that cases that are diagnosed later in life are also captured [
37,
38]. (iii) Ensure new cohorts, particularly large birth/pregnancy cohorts or those with the potential to prospectively collect data during pregnancy (such as the planned
UK Our Future Health) gain consent to collect health data on CHDs (and other congenital anomalies). (iv) Continue to update the cohorts used in this study and update our results. For example, there are plans to continue running GWAS assays on mothers, fathers and offspring in MoBa who are currently not genotyped which will more than double the sample available in that study. (v) To the best of our knowledge, there are currently no publicly available GWAS summary statistics for CHD. To date, the largest GWAS for CHD in a European population included ~ 4000 cases [
60]. As these GWAS continue to grow, significant data sharing and collaboration will be required, which could then pave way for large-scale two-sample MR studies to explore maternal risk factors for CHDs.
Acknowledgements
We are grateful to all the families who took part in the ALSPAC (Avon Longitudinal Study of Parents and Children), 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. BiB (Born in Bradford) study is only possible because of the enthusiasm and commitment of the children and parents. We are grateful to all the participants, practitioners, and researchers who have made the BiB study happen. The Norwegian Mother, Father and Child Cohort Study is supported by the Norwegian Ministry of Health and Care Services and the Ministry of Education and Research. We are grateful to all the participating families in Norway who take part in this ongoing cohort study. We thank the Norwegian Institute of Public Health (NIPH) for generating high-quality genomic data. This research is part of the HARVEST collaboration, supported by the Research Council of Norway (#229624). We also thank the NORMENT Centre for providing genotype data, funded by the Research Council of Norway (#223273), Southeast Norway Health Authorities and Stiftelsen Kristian Gerhard Jebsen. We further thank the Center for Diabetes Research, the University of Bergen for providing genotype data and performing quality control and imputation of the data funded by the ERC AdG project SELECTionPREDISPOSED, Stiftelsen Kristian Gerhard Jebsen, Trond Mohn Foundation, the Research Council of Norway, the Novo Nordisk Foundation, the University of Bergen, and the Western Norway Health Authorities.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.