This was a retrospective study and we used data from the China Birth Cohort Study (CBCS), which is a population-based birth defect surveillance system in China from November 2017 to December 2021 (24). The CBCS was conducted in 38 research sites in 17 provinces, cities, autonomous regions, and municipalities covering most areas of China. We obtained data access permissions for 9 hospitals among them. Data of each participant including birth outcome, demographic characteristics, lifestyle behaviors, environmental pollution, maternal disease history, and the current pregnancy information were required to be reported by trained obstetricians, sonographers, and pediatricians for the same CBCS system. To build a robust nomogram, participants from four hospitals in eastern China, including Beijing Obstetrics and Gynecology Hospital, were used as a development cohort. Participants from five hospitals in central and western China were used as an external validation cohort. All participants’ data of the above nine hospitals were collected. For the development cohort, 29,733 participants were initially recruited. Of these, 25 participants were excluded because of significant neurological or psychiatric illness of pregnant women (the accuracy of questionnaires could not be guaranteed), 494 multiple pregnancies were excluded since etiology of CHD may be different between singleton and multiple pregnancies (18), and a further 113 participants were excluded due to outliers or missing data in any variable of interest. Thus, 29,101 participants (28,798 normal + 303 with CHD) were retrieved for inclusion in the development cohort. At the same time, 15,477 participants (15,324 normal + 153 with CHD) were used as the external validation cohort using the same criteria as for the development cohort (Figure 1 and Supplementary Table 1). The sensitivity analysis that excluded participants with missing data yielded results similar to those presented (data not shown). This study was guided by the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) criteria (Supplementary Table 2) (25).

Our primary end point was the presence or absence of offspring CHD. CHD cases included livebirths, stillbirths, and late abortions who were diagnosed as CHD for more than 17 weeks of gestation. All cases were evaluated by a trained obstetrician, pediatrician, or cardiologist before hospital discharge or within 3 days after birth. Two echocardiography experts independently examined the echocardiographic images of each CHD case, and they were blinded to the outcome of each other. If there was still a dispute, computed tomography, cardiac catheterization, or autopsy were used to further determine a CHD case. All CHD cases were classified and recorded based on the International Classification of Diseases, Tenth Revision Modification (ICD-10-CM).

We identified potential predictors using a literature search. The following candidate predictors were selected from our standard and structured questionnaires: parental age, parental ethnicity (Han or Minority), parental first-trimester body mass index (BMI), household annual income (< 100,000 CNY or 100,000–400,000 CNY or > 400,000 CNY), parental education level (≤ 12 years or 13–16 years or ≥ 17 years), parental smoking (yes or no), maternal secondhand smoke exposure (yes or no), parental drinking (yes or no), maternal pre-pregnancy diabetes (yes or no), maternal fever (yes or no), maternal folic acid supplementation (yes or no), maternal multivitamin supplementation (yes or no), mode of conception (natural conceived or assisted reproductive technology), and environmental pollution (yes or no).

The height and weight of the participants were obtained through an accurate measurement. The standing height was measured to the nearest 0.1 cm using a stadiometer (26). The value of weight was accurately measured using an electronic scale (BW-150; UWE, Beijing, China) with participants wearing light clothes, no shoes, and empty pockets (27). The BMI was calculated as the weight in kilograms divided by the square of the height in meters. Parental smoking or drinking was defined as mothers or fathers who smoked at least one cigarette per day or drank alcohol once a week for over 6 months (28). Maternal secondhand smoke exposure was defined as non-smokers being exposed to another person’s tobacco smoke for at least 15 min daily for more than 1 day per week (29). Pre-pregnancy diabetes of pregnant women was defined as type 1 or type 2 diabetes diagnosed before pregnancy (15). Maternal fever was defined as body temperature exceeding 38°C or self-reported fever experience in the first trimester (16). Environmental pollution of this study was defined as the presence of smelly water ditches, garbage stations, or coal-fired factories within 100 m of the residential area.

Data were given as a median (interquartile range) or n (%). The independent t-test and Mann–Whitney U test were used to analyze differences in continuous variables between groups, while the chi-square test and the Fisher’s exact test were used for categorical variables. The univariable and multivariable analyses were used to select the dominant predictors. The multivariable logistic regression analysis was used to construct a risk prediction model for offspring CHD, based on the selected predictors of the development cohort. The variables in the final multivariable model were selected by the bidirectional stepwise regression with the Akaike information criterion. The results were described as the odds ratio (OR) and 95% confidence interval (CI). In addition, the outliers and collinearity of multivariable regression model were assessed. A nomogram consisting of the dominant predictors was created to translate the prediction model into a visual scoring system to calculate the estimated probability of offspring CHD. Each predicator was assigned a “points” on the nomogram, based on its predictive ability for offspring CHD. High total points based on the sum of the assigned points for each predicator in the nomogram were associated with a high CHD probability. To validate the performance of the nomogram, both internal and external validations were conducted. For both of them, the discriminatory ability of the nomogram was assessed using the receiver-operating-characteristic curve (AUC); if the AUC was closer to 1, the nomogram was seen as having a good discriminatory ability (30); the calibration of the nomogram was assessed using the calibration curves. To assess possible overfitting, the Delong test was implemented to compare the AUCs. Besides, a web-based nomogram was created to directly indicate the patient-specific probability of offspring CHD occurrence. All statistical tests were performed using the R statistical software version 4.1.1¹. The statistical significance was assumed at P < 0.05.

The characteristics of participants in the development and external validation cohorts are presented in Table 1. In the development and external validation cohorts, the median [interquartile range (IQR)] age of maternal age was 31.0 (29.0–34.0) and 31.0 (28.0–34.0) years, respectively. The paternal age was 33.0 (30.0–36.0) and 33.0 (30.0–37.0) years, respectively. The development cohort had a similar parental first-trimester BMI, parental ethnicity, household annual income, parental smoking, maternal secondhand smoke exposure, parental drinking, maternal pre-pregnancy diabetes, mode of conception, and environmental pollution status compared with the external validation cohort. A total of 456 CHD cases were included in this study, of which the three most types were ventricular septal defect (35.75%), multiple congenital heart disease (26.10%), and Tetralogy of Fallot (13.38%) (Supplementary Table 3).

In univariable analysis based on the data of the development cohort, the predictors associated with offspring CHD were maternal age (1.19, 95% CI: 1.15–1.22), paternal age (1.13, 95% CI: 1.11–1.15), household annual income (100,000–400,000 CNY, 0.29, 95% CI: 0.22–0.37; > 400,000 CNY, 0.16, 95% CI: 0.11–0.23), maternal education (13–16 years, 0.40, 95% CI: 0.31–0.52; ≥ 17 years, 0.38, 95% CI: 0.27–0.54), paternal education (13–16 years, 0.51, 95% CI: 0.40–0.65; ≥ 17 years, 0.38, 95% CI: 0.26–0.54), maternal secondhand smoke exposure (2.44, 95% CI: 1.93–3.09), paternal smoking (2.00, 95% CI: 1.59–2.51), maternal drinking (1.95, 95% CI: 1.27–3.00), paternal drinking (2.36, 95% CI: 1.88–2.96), maternal pre-pregnancy diabetes (3.38, 95% CI: 2.08–5.48), maternal fever (3.49, 95% CI: 2.71–4.50), maternal folic acid supplementation (0.16, 95% CI: 0.13–0.21), maternal multivitamin supplementation (0.31, 95% CI: 0.25–0.39), mode of conception (5.24, 95% CI: 4.07–6.75), and environmental pollution (2.73, 95% CI: 2.12–3.53). In contrast, parental ethnicity, parental first-trimester BMI, and maternal smoking did not show any association with offspring CHD (Table 2).

Multivariable logistic regression analysis led us to exclude paternal education, paternal smoking, and maternal drinking with P values higher than 0.05. Dominant risk factors for offspring CHD included increased maternal age (1.14, 95% CI: 1.10–1.19), increased paternal age (1.05, 95% CI: 1.02–1.09), maternal secondhand smoke exposure (2.89, 95% CI: 2.22–3.76), paternal drinking (1.41, 95% CI: 1.08–1.84), maternal pre-pregnancy diabetes (3.39, 95% CI: 1.95–5.87), maternal fever (3.35, 95% CI: 2.49–4.50), assisted reproductive technology (2.89, 95% CI: 2.13–3.94), and environmental pollution (1.61, 95% CI: 1.18–2.20). A higher household annual income (100,000–400,000 CNY, 0.47, 95% CI: 0.34–0.63; > 400,000 CNY, 0.23, 95% CI: 0.15–0.36), higher maternal education level (13–16 years, 0.68, 95% CI: 0.50–0.93; ≥ 17 years, 0.87, 95% CI: 0.55–1.37), maternal folic acid supplementation (0.21, 95% CI: 0.16–0.27), and maternal multivitamin supplementation (0.33, 95% CI: 0.26–0.42) were found to be protective factors for offspring CHD (Table 2).

Twelve dominant predictors, including maternal age, paternal age, household annual income, maternal education, maternal secondhand smoke exposure, paternal drinking, maternal pre-pregnancy diabetes, maternal fever, maternal folic acid supplementation, maternal multivitamin supplementation, mode of conception, and environmental pollution, were used to construct the prediction model and nomogram. There was no outlier with a model covariate of approximately ± 0.05 (Supplementary Figure 1). The variance inflation factor values of the twelve covariates ranged from 1.10 to 2.06, indicating no multicollinearity in the model (Supplementary Table 4). The predictive ability of each independent predictor of the model for CHD is reflected in the nomogram (Figure 2).

Internal validation of the nomogram based on 10-fold cross-validation method showed the AUC of 0.843, indicating the adequate overall risk discrimination power of the nomogram. The nomogram calibration curve showed good agreement between the nomogram-predicted probability of CHD and actual presence of CHD (Figure 3). As for external validation, the nomogram showed good discriminatory power, with AUCs of 0.849 and 0.837 in the development and the external validation cohorts, respectively (Figure 4). The Delong test revealed that the difference was not statistically significant between the AUCs of the nomogram (D = 0.627 and P = 0.531), indicating the absence of potential overfitting. The nomogram also showed good calibration in both the development and the external validation cohorts (Figure 5).

The web-based nomogram provided a user-friendly access, which is available at https://postdocxiaoman.shinyapps.io/DynNomCHDapp/ or the QR code (Supplementary Figure 2), and then the users could self-evaluate the risk of offspring CHD by simply selecting or filling the values of predictors on the user interface. For example, in case of a pregnant woman, aged 36 years and her husband aged 40 years, annual household income of over 400,000 (CNY), was educated for ≥ 17 years, exposed to secondhand smoke, alcoholic husband, maternal pre-pregnancy diabetes, maternal fever, maternal folic acid supplementation, maternal multivitamin supplementation, conceived through assisted reproductive technology, and not exposed to environmental pollution, corresponding to points of 64 + 36 + 0 + 5 + 30 + 12 + 38 + 33 + 0 + 0 + 30 + 0 = 248, the risk probability for offspring CHD was 0.226. The screenshots of the web-based nomogram are presented in Supplementary Figures 3–5.

In China, the current recommendation is to offer second-trimester ultrasound examination to screen for CHD and other malformations. An effective screening method for CHD in the first trimester is still lacking. Our study has demonstrated that the nomogram achieves good performance in the development and external validation cohorts. This implies that, if the nomogram is used, pregnant women with high-risk of offspring CHD could be detected in the early stage of pregnancy, and more preventive interventions could be provided to them as early as possible. Therefore, screening of pregnant women during routine first-trimester prenatal examination to identify those with high risk of offspring CHD should be encouraged. In addition, most of the predictors included in our nomogram are easily accessible and modifiable. Healthcare workers and expectant parents should be educated about the risk and protective factors of CHD, and the corresponding study should also be conducted to promote the awareness of influences from non-genetic factors in social communities.

This study has several strengths. First, our study has a large sample size based on a nationwide multicenter study, which enabled us to build a nomogram with a small expected relative bias. Second, detailed information of participants was recorded, allowing us to analyze potential predictors as many as possible, including demographic characteristics, lifestyle behaviors, environmental pollution, maternal disease history, and the current pregnancy information of the study population. Third, the nomogram established shows good performance, the web-based nomogram could be conveniently and effectively applied to the prediction of CHD, especially in developing countries that lack well-developed pregnancy monitoring systems. Fourth, we conducted extensive internal and external validations using population from different centers based on TRIPOD criteria, which increases the generalizability of the findings.

However, we acknowledge that several limitations cannot be neglected. First, although we included as many predictors as possible in this study, we cannot rule out the possibility of some residual effects related to other factors such as the adverse emotion of pregnant women during pregnancy (32). Second, participants with outliers or missing data in any variable of interest were excluded, which may have contributed to bias in our findings, but we believe that our final conclusion is robust since the total rate of outliers and missing data is only 0.35% (160 of 45,504). Third, maternal/paternal data such as drinking, smoking, and environmental pollution were from self-administered questionnaire, which may decrease the reliability of our findings. However, the questions are presented in simple language and have been piloted, and the forms completed by participants of the study and medical staff have shown high concordance. Fourth, our participants were predominantly from a single ethnicity and tertiary hospitals, and the nomogram requires prospective validation in other ethnicities and general community populations. Fifth, the future use of our nomogram may be limited by the annual increase in household income like in 10 years when people have much higher income, so updating periodically is needed. The nomogram may not be applicable to individuals with factors lying outside the range of the nomogram such as younger or older pregnant women.