Study design and subjectspatients

A hospital-based case–control study was conducted in Shanghai Children’s Medical Center through June 2016 to December 2017. Sample size was estimated by the following calculations:^{Reference Jekel11}

$$\eqalign{ n = {\bigg\{ [{Z_{1 - {\alpha \over 2}}}\sqrt {2\overline P\left( {1 - \overline P} \right)} + {Z_\beta }\sqrt {{P_1}\left( {1 - {P_1}} \right) + {P_0}\left( {1 - {P_0}} \right)} ]/ }\cr \hskip 12 {({P_1} - {P_0})\bigg\} ^2} \cr \overline P = ({P_1} + {P_0})/2 \cr {P_1} = (OR \times {P_0})/(1 - {P_0} + OR \times {P_0})} $$

In formulas, P ₀ = proportion of events in the control group, OR = odds ratio of events, P ₁ = proportion of events in the case group, ${\overline P}$ = proportion of events in total, ${Z_{1 - {\alpha \over 2}}}$ = value of the standard normal distribution corresponding to 1 − ${a\over 2}$ , and ${Z_\beta }$ = value of the standard normal distribution corresponding to β.

In our preliminary study, 24.0% of pregnant women used cod liver oil supplements. Compared to those mother who did not use cod liver oil as supplements, odds ratio for the risk of offspring CHD was 0.432 in mothers who used cod liver oil supplements. Therefore, 0.24 and 0.432 were taken as the estimates of P₀ and odds ratio of events, respectively. In total, 214 cases versus 214 controls were calculated to be the minimum sample size to achieve appropriate statistical power (α = 0.05, β = 0.1). In this study, 262 children with severe CHD, along with 262 control children without any birth defects, were enrolled. The detailed information of the case–control study has been described elsewhere.^{Reference Zhao, Zhao and Xia12}

The CHD diagnoses were based on the codes of the International Classification of Diseases, the Tenth Revision, Clinical Modification. A team of experts, including paediatric cardiothoracic surgeons and fetal ultrasonologists, evaluated and ensured the accuracy of the final diagnosis. Severe CHD in this study included tetralogy of Fallot (n = 97, 37.0%), transposition of the great arteries (n = 32, 12.2%), functional single ventricle (n = 23, 8.8%), pulmonary atresia (n = 17, 6.5%), single atrium (n = 10, 3.8%), atrioventricular septal defect (n = 10, 3.8%), tricuspid atresia (n = 9, 3.4%), total anomalous pulmonary venous connection (n = 8, 3.1%), double-outlet right ventricle (n = 6, 2.3%), hypoplastic left heart syndrome (n = 6, 2.3%), congenital mitral stenosis (n = 6, 2.3%), persistent truncus arteriosus (n = 6, 2.3%), interrupted aortic arch (n = 6, 2.3%), and others (n = 26, 9.9%) with reference to the classification of previous studies.^{Reference Yeh, Chen and Lu13} All the sampled children with severe CHD survived after birth and were younger than 2 years old.

The children in the control group were recruited from the children admitted into the same hospital during the same period when the cases were recruited. Among the 262 controls, 132 came from paediatric respiratory medicine, 91 from paediatric general surgery, and 39 from paediatric gastroenterology. Similar to the cases, all children in the control group were younger than 2 years old.

Children with any of the following conditions were excluded from the study: (1) cases with other congenital deformities except CHD, such as Down syndrome, 22q11.2 microdeletion, Noonan syndrome, Williams syndrome, Marfan syndrome, and so on; (2) death of the mother; (3) mother diagnosed with mental disorders; and (4) inability to locate the mother for interview.

Measures

Information on socio-demographic characteristics and parental health-related behaviours was retrospectively collected through the Parental Behaviours and Environmental Exposure Questionnaire, a self-made questionnaire used in our previous study.^{Reference Zhao, Zhao and Xia12} For both cases and controls, only those mothers who signed the informed consent were invited to fill out the questionnaires whilst in hospital.

In this study, maternal folic acid and cod liver oil supplementation were evaluated by two items in the Parental Behaviours and Environmental Exposure Questionnaire. One was regarding maternal folic acid supplementation: “how often did you take folic acid supplements around the periconceptional period?”, and the other one was related to cod liver oil supplementation: “how often did you take cod liver oil capsules around the periconceptional period?”. Periconceptional period was defined as during 1 month before pregnancy and throughout the gestation. The responses to the two questions were rated on a three-point scale: “occasionally/never” for <1 day per week on an average, “frequently” for 1–3 days per week on an average, and “almost always” for ≥4 days per week on an average. To implement the propensity score matching analysis and explore the additive interaction between folic acid and cod liver oil, maternal folic acid supplementation was classified as two categories: regular folic acid supplementation if the responses were “frequently” or “almost always”, and none if the response was “occasionally/never”; and maternal cod liver oil supplementation was classified as two categories: regular cod liver oil supplementation if the responses were “frequently” or “almost always”, and none if the response was “occasionally/never”.

Potential confounding variables

Maternal ethnicity was categorised as Han ethnicity versus others; maternal age at delivery was grouped into two categories: <35 years old versus ≥35 years old; maternal educational level was grouped into three categories: middle school and below, high school, and college and above; marital status was categorised as married versus unmarried/divorced/widowed; residence was categorised as urban versus suburban/rural; maternal pre-pregnancy obesity was grouped into two categories: yes versus no (obesity was defined as body mass index ≥28;^{Reference Zhou14} body mass index was calculated as weight in kilograms divided by height in meters squared, based on pre-pregnancy height and weight), multiple birth was grouped as yes versus no; family history of CHD was grouped as yes versus no; infant gender was categorised as male versus female; maternal pre-pregnancy diabetes (defined as pre-gestational diabetes, yes versus no) and/or maternal pre-pregnancy hypertension (defined as pre-gestational hypertension, yes versus no); maternal smoking (defined as smoking before and/or during pregnancy, yes versus no); and maternal drinking (defined as drinking before and/or during pregnancy, yes versus no).

Statistical analysis

Statistical description was made by using percentages for categorical variables, and the chi-square test and Fisher’s exact test were employed to compare the difference between cases and controls.

To identify the relationship of maternal folic acid and cod liver oil supplementation with the risk of CHD, the odds ratio and 95% confidence intervals were calculated through univariable and multivariable logistic regression models. Adjustments were made in the multivariable logistic regression models. Model A was adjusted for demographic and obstetric characteristics. Model B, based on model A, was further adjusted for maternal health indicators and behaviours, including pre-pregnancy diabetes/hypertension and smoking/drinking. To reduce the interference with each other within those taking folic acid and cod liver oil, model C was further adjusted for folic acid or cod liver oil supplementation on the basis of model B. To examine the combined effect of the two supplementations, the relative excess risk was used to evaluate the additive interaction.

Furthermore, the present study adopted a propensity score method for matching, to reduce potential selection bias and to balance demographic differences and other confounding variables.^{Reference Austin15} A multivariable logistic regression model was developed to estimate the propensity score for mothers who used cod liver oil supplementation or not. All the potential confounding variables related to CHD, including maternal ethnicity, maternal age at delivery, maternal education, marital status, residence, maternal pre-pregnancy obesity, multiple births, infant gender, family history of CHD, pre-pregnancy diabetes, hypertension, smoking/drinking, and folic acid supplementation, were included in the logistic regression model for minimising the confounding effects and estimating the independent impact of cod liver oil. In this propensity-score-matched analysis, mothers who did not use cod liver oil supplementation were matched 1:1 to mothers who used cod liver oil supplementation. The propensity score matching was based on a greedy nearest neighbour-matching algorithm on propensity score with a calliper equalling to 0.1. Standardised mean difference was applied to examine the balance of covariate distribution between mothers who used, and those who did not use, cod liver oil supplementation.^{Reference Zhang, Kim and Lonjon16} A standardised mean difference smaller than 0.1 can be considered to be balanced between the two groups.^{Reference Zhang, Kim and Lonjon16} Meanwhile, a multivariable conditional logistic regression was applied in the matched data.

We conducted the propensity score and additive interaction analyses using R version 3.5.1 (The R Foundation for Statistical Computing). All other analyses were performed with the Statistical Package for the Social Sciences (SPSS Statistics v23.0; IBM, Chicago, Illinois, United States of America). Statistical significance level was set at p value <0.05.