#These authors contributed equally to the manuscript
Congenital heart disease (CHD) is a large, rapidly emerging global problem in child health; in the Global, regional, and national burden of CHD Study 2017, the global prevalence of CHD is estimated to be nearly 1.8‰ [
However, the findings presented in previous studies are usually based on logistic regression analysis and the premise that the factors are independent with each other. However, CHD is a multifactorial complex disease. Current research indicates that the interactions between factors and the synergistic effects of multiple factors on the risk of CHD are difficult to consider and accurately estimate.
Therefore, this study aims to establish an optimal Bayesian network (BN) model according to our large maternal-fetal dataset, detect factors that influence fetal CHD and quantify the extent of their correlations. This method mainly shows the causal relationships among variables by performing probabilistic reasoning and reflects the potential relationships among multiple factors [
In all, 16,086 pregnant women who underwent fetal echocardiography from June 2010 to June 2017 at our center and other participating centers were consecutively recruited to reduce the potential bias partially. All data are kept in our center’s maternal-fetal medicine database. Data from our center accounted for the main part of the dataset. Fetal ages ranged from 16 to 39 gestational weeks (GWs) and were calculated from the last menstrual period.
Enrollment criteria included (1) relatively complete data, and (2) fetal heart examination and diagnosis of fetal CHD meeting the specifications for the diagnosis and treatment of fetal cardiac disease released by AHA [
The data and fetal echocardiography images obtained from the participating institutions were analyzed in our center. They were exported from our maternal-fetal database together with our center’s data. We acquired the following information through questionnaires while patients were waiting to be examined: maternal factors included age, comorbidities (diabetes, upper respiratory infection during early pregnancy, anemia, connective tissue diseases and thyroid disease), medication exposure, history of induced labor or spontaneous abortion or CHD, consanguineous marriage, occupation and bad habits of subjects and spouses, radioactive substances exposure, and mental stress during the early pregnancy. The diagnostic cut-off for Diabetes was set at 5.3–10.0–8.6 mmol/L in 75 g OGTT [
N of Variables | Variables | Value | Group details |
---|---|---|---|
1 | Maternal age (years) | 1 | <20 |
1 | 2 | 20 ≤ age < 35 | |
1 | 3 | ≥35 | |
2 | Induced labor | 1 | Yes |
2 | 2 | No | |
3 | Diabetes mellitus | 1 | Yes |
3 | 2 | No | |
4 | Progesterone use | 1 | Yes |
4 | 2 | No | |
5 | Spontaneous abortion | 1 | Yes |
5 | 2 | No | |
6 | Conception method | 1 | Natural conception |
6 | 2 | IVF | |
7 | Upper respiratory infection | 1 | Yes |
7 | 2 | No | |
8 | NSAID use | 1 | Yes |
8 | 2 | No | |
9 | Anticonvulsant use | 1 | Yes |
9 | 2 | No | |
10 | Lithium use | 1 | Yes |
10 | 2 | No | |
11 | ACEI use | 1 | Yes |
11 | 2 | No | |
12 | Anemia | 1 | Yes |
12 | 2 | No | |
13 | Pregnant woman smoking | 1 | Yes |
13 | 2 | No | |
14 | Pregnant woman drinking | 1 | Yes |
14 | 2 | No | |
15 | Spouse smoking | 1 | Yes |
15 | 2 | No | |
16 | Spouse drinking | 1 | Yes |
16 | 2 | No | |
17 | Spouse’s medication use during preparation for pregnancy | 1 | Yes |
17 | 2 | No | |
18 | Exposure to radioactive substances | 1 | Yes |
18 | 2 | No | |
19 | Mental stress | 1 | Yes |
19 | 2 | No | |
20 | Intermarriage | 1 | Yes |
20 | 2 | No | |
21 | Family history of CHD | 1 | Yes |
21 | 2 | No | |
22 | Twin pregnancy | 1 | Single |
22 | 2 | Twin | |
23 | Fetal arrhythmia | 1 | Yes |
23 | 2 | No | |
24 | Fetal hydrops | 1 | Yes |
24 | 2 | No | |
25 | Thyroid disorder | 1 | Yes |
25 | 2 | No | |
26 | Gestational weeks | 1 | 16 ≤ GW < 28 |
26 | 2 | 28 ≤ GW < 40 |
The diagnosis of fetal CHD was based on fetal echocardiography using a Voluson E8-RAB4-8 machine equipped with a 2- to 8-MHz transducer (GE Healthcare, Little Chalfont, United Kingdom), an Aloka 10, UST-9130 equipped with a 3- to 6-MHz transducer (Aloka, Tokyo, Japan), or a Philips IU-22, C5-1 equipped with a 1- or 5-MHz transducer (Philips Healthcare, Bothell, WA, USA). The acquisition of fetal echocardiographic images was performed according to the guidelines and standards of the AHA 7 and the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) [
Fetal echocardiography was performed by experienced associate chief physicians and chief physicians, and diagnoses were then made based on grayscale and color images and pulse wave Doppler according to multiple section screening, including four-chamber, left and right ventricular outflow tract (LVOT and RVOT), three-vessel (3V), and three vessels and trachea (3VT) views as well as sagittal views of the superior and inferior vena cava, aortic arch, and ductal arch. All relevant physicians at the participating institutions were trained according to these guidelines. All images that were uploaded to our database from participating institutions were independently reviewed by two experienced physicians who confirmed or corrected the diagnoses.
Continuous variables with a Gaussian distribution were expressed as the mean ± standard deviation and were compared using a
BN was performed to detect potential relationships between factors and fetal CHD, which consisted of two components: a directed acyclic graph (DAG) that encoded the dependency structure of the network and a conditional probability table (CPT) for each node given its parent set. The learning process of the BN consisted of two parts: structural learning and parameter estimation. A BN can reconstruct a join probability distribution between variables. Let
We construct a BN using the Matlab software (https://matlab.mathworks.com) in two steps. Firstly, learn the structure of the BN, i.e., identifying the topological structure of the BN; Secondly, learn the parameters of the BN, i.e., by estimating the CPT of each node from the dataset after the structure of the BN is identified. In order to simplify the model, BN is based on discrete variables here. The values of most of the exposure variables are discrete. The variables with more values, such as the gestational weeks and the age of the pregnant woman, are classified into several discrete variables with value to facilitate the learning of BN.
We applied four algorithms to learn the structure of the BN, which is now available via the open -source software BNT Murphy (2004) [
When the network topology G is confirmed by these four algorithms, the parameters of CPT at each node are learned by the maximum likelihood estimation (MLE) given the complete dataset.
We chose the best model that had a better representation of the original data distribution. K-fold cross-validation was applied to evaluate the performance of the model, in which the dataset was divided into K independent subsets, and every time the K-1 subset was applied to train the BN model, another subset was used to test the model that was just trained; after K times, ROC curves were drawn based on the prediction score of the classifying class (
When BN is established, causal reasoning can be conducted by the junction tree algorithm [
We conducted two experiments. When GWs were ignored, the exposure group was set up, and we calculated the probability of the disease occurring given different combinations of observed factors based on causal reasoning under the condition of a node with no considering set to normal value. The probability that all nodes would be set to normal values as a result of the control group was determined, and the risk ratio (RR) could then be calculated by the predictive incidence probability in the exposure and control groups. When GWs were considered, the subjects were divided into two groups (Group A: 16≤,<28 weeks, and Group B: 28≤,<40 weeks).
To study whether there is a risk of disease caused by hazardous environmental factors, the experimental group and the control group are usually set up, and the risk ratio (Risk Ratio, RR) of the two groups is calculated, which represents the ratio of the incidence of the experimental group and the control group. The calculation process is shown in
In this article, the experimental group and the control group are distinguished by the values of environmental variables. In the Bayesian network model, the calculation of the risk ratio RR value is shown in
Among all the participants, 3,312 pregnant women had CHD fetuses, and the remainders were normal fetuses. The age range of most pregnant women (approximately 80%) was between 20 and 35 years old. The proportion of subjects in group A (GWs < 28 weeks) was 75.9%, and the proportion in group B (GW ≥ 28 weeks) was 24.1%. The percentages of patients with diabetes, anemia, upper respiratory infection, progesterone use and family history of CHD were higher in the normal group than in the CHD group, while the rates of spontaneous abortion, spouse smoking and drinking, and twin pregnancy were higher in the CHD group (all
Group | Fetal CHD group | Normal group | |
---|---|---|---|
N | 3312 | 12774 | – |
Age (n, %) | 0.048 | ||
<20 years old | 18 (0.54%) | 38 (0.30%) | |
20≤,<35 years old | 2746 (82.91%) | 10728 (84.00%) | |
≥35 years old | 548 (16.55%) | 2008 (15.72%) | |
Induced labor (n, %) | 0.309 | ||
Yes | 124 (3.74%) | 432 (3.38%) | |
No | 3188 (96.26%) | 12342 (96.62%) | |
Diabetes mellitus | 1.09 × 10–45 | ||
Yes | 145 (4.38%) | 1680 (13.15%) | |
No | 3167 (95.62%) | 11094 (86.85%) | |
Gestational weeks | 2.27 × 10–106 | ||
16≤, <28 | 2034 (61.41%) | 10178 (79.68%) | |
28≤, <40 | 1278 (38.59%) | 2596 (20.32%) | |
Progesterone use | 9.18 × 10–4 | ||
Yes | 542 (16.36%) | 2410 (18.87%) | |
No | 2770 (83.64%) | 10364 (81.13%) | |
Spontaneous abortion | 8.49 × 10–14 | ||
Yes | 918 (27.72%) | 2760 (21.61%) | |
No | 2394 (72.28%) | 10014 (78.39%) | |
Conception method | 0.694 | ||
Natural conception | 3216 (97.10%) | 12387 (96.97%) | |
IVF | 96 (2.90%) | 387 (3.03%) | |
Upper respiratory infection | 0.016 | ||
Yes | 641 (19.35%) | 2717 (21.27%) | |
No | 2671 (80.65%) | 10057 (78.73%) | |
NSAID use | 0.511 | ||
Yes | 22 (0.66%) | 99 (0.78%) | |
No | 3290 (99.34%) | 12675 (99.22%) | |
Anticonvulsant use | 0.611 | ||
Yes | 0 (0.00%) | 1 (0.01%) | |
No | 3312 (100.00%) | 12773 (99.99%) | |
Lithium use | 0.471 | ||
Yes | 0 (0.00%) | 2 (0.02%) | |
No | 3312 (100.00%) | 12772 (99.98%) | |
ACEI | 0.145 | ||
Yes | 3 (0.09%) | 4 (0.03%) | |
No | 3309 (99.91%) | 12770 (99.97%) | |
Anemia | 8.30 × 10–6 | ||
Yes | 242 (7.31%) | 1256 (9.83%) | |
No | 3070 (92.69%) | 11518 (90.17%) | |
Pregnant woman smoking | 0.080 | ||
Yes | 9 (0.27%) | 64 (0.50%) | |
No | 3303 (99.73%) | 12710 (99.50%) | |
Pregnant woman drinking | 0.579 | ||
Yes | 11 (0.33%) | 51 (0.40%) | |
No | 3301 (99.67%) | 12723 (99.60%) | |
Spouse smoking | 9.26 × 10–5 | ||
Yes | 428 (12.93%) | 1379 (10.8%) | |
No | 2884 (87.08%) | 11395 (89.20%) | |
Spouse drinking | 5.50 × 10–4 | ||
Yes | 218 (6.58%) | 654 (5.12%) | |
No | 3094 (93.42%) | 12120 (94.88%) | |
Spouse’s medication use during preparation for pregnancy | 0.865 | ||
Yes | 81 (2.45%) | 319 (2.50%) | |
No | 3231 (97.55%) | 12455 (97.50%) | |
Exposure to radioactive substances | 0.115 | ||
Yes | 26 (0.79%) | 140 (1.10%) | |
No | 3286 (99.21%) | 12634 (98.90%) | |
Mental stress | 0.271 | ||
Yes | 79 (2.39%) | 265 (2.07%) | |
No | 3233 (97.61%) | 12509 (97.93%) | |
Intermarriage | 0.283 | ||
Yes | 2 (0.06%) | 3 (0.02%) | |
No | 3310 (99.94%) | 12771 (99.98%) | |
Family history of CHD | 0.030 | ||
Yes | 24 (0.72%) | 148 (1.16%) | |
No | 3288 (99.28%) | 12626 (98.84%) | |
Twin pregnancy | 5.72 × 10–5 | ||
Single | 3235 (97.68%) | 12601 (98.65%) | |
Twins | 77 (2.32%) | 173 (1.35%) | |
Fetal arrhythmia | 0.950 | ||
Yes | 40 (1.21%) | 156 (1.22%) | |
No | 3272 (98.79%) | 12618 (98.78%) | |
Fetal hydrops | 0.124 | ||
Yes | 72 (2.17%) | 338 (2.65%) | |
No | 3240 (97.83%) | 12436 (97.35%) | |
Thyroid disorder | 0.467 | ||
Yes | 19 (0.57%) | 88 (0.69%) | |
No | 3293 (99.43%) | 12686 (99.31%) |
Type of fetal CHD | N (%) |
---|---|
Congenital anomaly of systemic, pulmonary, or umbilical veins or ductus venosus | 323 (9.75) |
Congenital anomaly of an atrium or atrial septum | 89 (2.69) |
Congenital anomaly of an atrioventricular connection | 174 (5.25) |
Congenital anomaly of a ventricle or the ventricular septum | 632 (19.08) |
Congenital anomaly of a ventriculo-arterial connection | 784 (23.67) |
Congenital anomaly of the great arteries, including the arterial duct | 598 (18.06) |
Fetal cardiomyopathy | 11 (0.33) |
Fetal cardiac tumors | 86 (2.60) |
Fetal arrhythmia | 222 (6.70) |
Others | 393 (11.87) |
Total | 3312 (100) |
Others included ventricular outpouchings, aneurysm of the atrial appendage and atrium, abnormal heart position, abnormal cardiothoracic or atrial-ventricular ratio, abnormal proportions of the aorta and pulmonary artery, heart failure, pericardial tumors, multiple intracardiac echogenic focus or multiple calcifications, bicuspid aortic valves.
The BN structure analysis showed that several factors were directly associated with fetal CHD; these factors included a history of spontaneous abortion, upper respiratory tract infection during early pregnancy, anemia, and mental stress as well as twin pregnancy and parental smoking. Based on the causal reasoning of the BN, we found that the risk of fetal CHD gradually increased with potentially synergistic exposure of ranging from a single factor to multiple factors. A single factor analysis demonstrated that the RRs of twin pregnancy, spontaneous abortion, or spouse smoking were 1.50, 1.38, and 1.11 (all
We further performed a sensitivity analysis in the participants grouped by GWs. We obtained a result consistent with the above-described result no matter we explored the population in Group A or Group B or used a single factor or multifactor exposure analysis. These findings demonstrate that the risk of a combination of these five factors occurring was as high as 2.88 (
Exposure factor group | Subject group | Variable 1 | Variable 2 | Variable 3 | Variable 4 | Variable 5 | Predictive CHD incidence probability in CHD group | Predictive CHD incidence probability in control group | Risk ratio (RR) | |
---|---|---|---|---|---|---|---|---|---|---|
Single factor exposure | Total population | Single or twin = 2 | / | / | / | / | 0.3244 | 0.2169 | 1.496 | 5.72 × 10–5 |
Abortion = 1 | / | / | / | / | 0.2999 | 0.2169 | 1.383 | 8.49 × 10–14 | ||
Spouse smoking = 1 | / | / | / | / | 0.2415 | 0.2169 | 1.113 | 5.50 × 10–4 | ||
16 ≤ gw < 28 | Single or twin = 2 | / | / | / | / | 0.2675 | 0.1740 | 1.537 | ||
Abortion = 1 | / | / | / | / | 0.2458 | 0.1740 | 1.412 | |||
Spouse smoking = 1 | / | / | / | / | 0.1949 | 0.1740 | 1.120 | |||
28 ≤ gw < 40 | Single or twin = 2 | / | / | / | / | 0.4959 | 0.3620 | 1.370 | 3.54 × 10–9 | |
Abortion = 1 | / | / | / | / | 0.4674 | 0.3620 | 1.291 | 9.55 × 10–43 | ||
Spouse smoking = 1 | / | / | / | / | 0.3947 | 0.3620 | 1.090 | 2.25 × 10–21 | ||
Synergistic effect of two factors | Total population | Abortion = 1 | Single or twin = 2 | / | / | / | 0.4262 | 0.2169 | 1.965 | 0.039 |
Spouse smoking = 1 | Single or twin = 2 | / | / | / | 0.3556 | 0.2169 | 1.640 | 0.025 | ||
16 ≤ gw < 28 | Abortion = 1 | Single or twin = 2 | / | / | / | 0.3610 | 0.1740 | 2.074 | 0.198 | |
Spouse smoking = 1 | Single or twin = 2 | / | / | / | 0.2956 | 0.1740 | 1.699 | 0.764 | ||
28 ≤ gw < 40 | Abortion = 1 | Single or twin = 2 | / | / | / | 0.6034 | 0.3620 | 1.667 | 0.041 | |
Spouse smoking = 1 | Single or twin = 2 | / | / | / | 0.5307 | 0.3620 | 1.466 | 0.585 | ||
Synergistic effect of three factors | Total population | upper respiratory infection = 1 | Anemia = 1 | Single or twin = 2 | / | / | 0.3386 | 0.2169 | 1.561 | 0.029 |
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | / | / | 0.3136 | 0.2169 | 1.446 | 0.023 | ||
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | / | / | 0.2534 | 0.2169 | 1.168 | 0.013 | ||
16 ≤ gw < 28 | upper respiratory infection = 1 | Anemia = 1 | Single or twin = 2 | / | / | 0.2803 | 0.1740 | 1.611 | 0.146 | |
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | / | / | 0.2578 | 0.1740 | 1.482 | 0.938 | ||
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | / | / | 0.2052 | 0.1740 | 1.179 | 0.236 | ||
28 ≤ gw < 40 | upper respiratory infection = 1 | Anemia = 1 | Single or twin = 2 | / | / | 0.5120 | 0.3620 | 1.414 | 0.049 | |
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | / | / | 0.4834 | 0.3620 | 1.335 | 4.16 × 10–5 | ||
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | / | / | 0.4102 | 0.3620 | 1.133 | 7.85 × 10–6 | ||
Synergistic effect of four factors | Total population | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | Single or twin = 2 | / | 0.4607 | 0.2169 | 2.124 | 2.59 × 10–7 |
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | / | 0.4325 | 0.2169 | 1.994 | 4.14 × 10–6 | ||
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Single or twin = 2 | / | 0.3704 | 0.2169 | 1.708 | 7.51 × 10–4 | ||
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Mental stress = 1 | / | 0.3615 | 0.2169 | 1.667 | 0.001 | ||
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | Single or twin = 2 | / | 0.4419 | 0.2169 | 2.037 | 1.69 × 10–6 | ||
16 ≤ gw < 28 | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | Single or twin = 2 | / | 0.3938 | 0.1740 | 2.263 | 1.09 × 10–6 | |
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Single or twin = 2 | / | 0.3092 | 0.1740 | 1.777 | 0.002 | ||
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Mental stress = 1 | / | 0.3010 | 0.1740 | 1.730 | 0.003 | ||
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | Single or twin = 2 | / | 0.3759 | 0.1740 | 2.160 | 6.13 × 10–6 | ||
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | / | 0.3669 | 0.1740 | 2.109 | 1.41 × 10–5 | ||
28 ≤ gw < 40 | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | Single or twin = 2 | / | 0.6364 | 0.3620 | 1.758 | 4.06 × 10–8 | |
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Single or twin = 2 | / | 0.5466 | 0.3620 | 1.510 | 2.09 × 10–4 | ||
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Mental stress = 1 | / | 0.5371 | 0.3620 | 1.483 | 4.32 × 10–4 | ||
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | Single or twin = 2 | / | 0.6187 | 0.3620 | 1.709 | 2.82 × 10–7 | ||
Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | / | 0.6096 | 0.3620 | 1.684 | 7.27 × 10–7 | ||
Synergistic effect of five factors | Total population | Abortion | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | Single or twin = 2 | 0.5692 | 0.2169 | 2.624 | 5.48 × 10–13 |
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Mental stress = 1 | Single or twin = 2 | 0.4954 | 0.2169 | 2.284 | 6.03 × 10–9 | ||
16 ≤ gw < 28 | Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | Single or twin = 2 | 0.5012 | 0.1740 | 2.880 | 4.54 × 10–12 | |
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Mental stress = 1 | Single or twin = 2 | 0.4275 | 0.1740 | 2.457 | 3.24 × 10–8 | ||
28 ≤ gw < 40 | Abortion = 1 | upper respiratory infection = 1 | Anemia = 1 | Mental stress = 1 | Single or twin = 2 | 0.7302 | 0.3620 | 2.017 | 1.41 × 10–13 | |
upper respiratory infection = 1 | Anemia = 1 | Spouse smoking = 1 | Mental stress = 1 | Single or twin = 2 | 0.6679 | 0.3620 | 1.845 | 9.31 × 10–10 |
This study focused on factors affecting fetal CHD in a large sample size consisting of 3,312 pregnant women with CHD fetuses among 16,086 subjects. Importantly, instead of traditional logistic regression analysis, we used BN analysis, a method based on probabilistic reasoning, to explore the interactions among specific factors and factors associated with an increased risk of fetal CHD. The results of this research have revealed that the factors directly associated with fetal CHD included history of spontaneous abortion, upper respiratory tract infection during early pregnancy, anemia, and mental stress as well as twin pregnancy and parental smoking. The synergistic exposure of more factors increased the risk of fetal CHD, leading to RR as high as 2.62 for the five-factor synergistic effect. In addition, a sensitivity analysis of groups divided by GWs demonstrated the risk showed the same trends as those described above from the single factor to synergistically multiple-factor exposure in both groups.
The factors associated with fetal CHD identified in this study have also been described in other similar studies. One of these factors is maternal illness, such as upper respiratory tract infection during early pregnancy. A meta-analysis of maternal viral infection and the risk of fetal CHD suggested that mothers who had history of viral infection in early pregnancy had significantly higher risk of having offspring with CHD (RR = 2.28), and this risk was more significant in mothers with rubella and cytomegalovirus. The effect of nonspecific maternal infection is difficult to definitively separate from the effects of medications used to treat the illness, including maternal fever and infection [
In addition to maternal infection, other maternal chronic diseases associated with a risk of CHD in offspring, including diabetes, hypertension, anemia, connective tissue disorders, epilepsy and mood disorders, have been reported to be significantly associated with a higher prevalence of any form of CHD in offspring. Moreover, the population-attributed risk for CHD was investigated, and the results showed that the highest population-attributable risk was noted for anemia (2.17%), followed by type 2 diabetes (1.45%) and hypertension (0.71%) [
The association between maternal diabetes and an increased risk of CHD has been clearly described in many studies [
The correlation between paternal smoking and congenital cardiovascular defects has been studied, but too little information is available to determine the associated risk. Of the many congenital defects observed in a nursery, there was a significantly higher incidence of cardiovascular system abnormalities in the tobacco-exposed group [
With regard for fetal factors, we found that there was a correlation between multiple gestations and fetal CHD. In 2016, Panagiotopoulou et al. conducted a study on CHD in twin pregnancy and showed that monochorionicity (OR 3.49, 95% CI 1.57–7.77) was a significant determinant of CHD that was independent of maternal age, parity, and the gender of the offspring [
The large sample size included in this study makes it the first study to assess factors related to fetal CHD using BN analysis, which can visually reflect potential relationships among multiple factors by constructing a DAG. There is no strict requirement for statistical assumptions. However, conclusions from previous studies have usually been drawn based on logistic regression models and the premise that the evaluated factors are independent with each other. Based on the current results, the interactions among factors and the synergistic effects of multiple factors on the risk of CHD are difficult to consider and accurately estimate. Therefore, considering the large sample dataset and the interactions observed between the factors included in this study, a BN is a very reasonable method for exploring factors that affect CHD, and these results are objective and in agreement with medical explanations. This method not only selected several factors that were clearly related to fetal CHD but also presented an accurate estimation of factors and the synergistic effect of multiple factors to compensate for the shortcomings of current research. Another strength of the study is the fact that we examine fetal CHD as opposed to only live births and thus would capture pregnancies who would go on to have intrauterine demise or termination that would not be captured in a neonatal/live birth registry. We acknowledge that although it based on a large population, the data were mainly obtained from self-reported questionnaires, and the accuracy of information collection is therefore a problem that needs to be considered. Moreover, this is a cross-sectional study that demonstrates only the correlations between these factors and fetal CHD but does not provide causal relationships. One additional limitation is that our center is a referral center for fetal heart disease. The fetuses referred to our center come from all over the country. Therefore, some of the patients coming to our center are pregnant women with known risk factors or with a fetus previously found to have CHD at a local hospital, and this may have led to selection bias in the population. In addition, fetal CHD was diagnosed by fetal echocardiography and we did not make postnatal verification for every case. But our findings can be credible, because fetal echocardiographic diagnoses were mostly consistent with autopsy findings in our center [
The structured learning and parameter estimation in the BN demonstrated that factors directly associated with CHD included a history of spontaneous abortion, upper respiratory tract infection during early pregnancy, anemia, and mental stress as well as the number of births and spouse smoking. Combinations containing higher numbers of these factors were associated with a higher risk of fetal CHD for the total population or the population grouped by GWs. All these findings suggest that improvements in the management of obstetric healthcare and the provision of prenatal counseling for women with these risk factors should be strengthened to decrease the incidence of CHD.
Congenital heart disease
Bayesian network
Risk ratio
American heart association
American Society of Echocardiography
International Society of Ultrasound in Obstetrics and Gynecology
Left and right ventricular outflow tract
Right ventricular outflow tract
Three-vessel
Three vessels and trachea
Interquartile range
Directed acyclic graph
Conditional probability table
Maximum likelihood estimation
Area under the curve
Gestational weeks
Maximum weight spanning tree
With two random initializations
K2 with MWST initialization
K2 with inverse MWST initialization
Starting from an empty structure
GS starting from a MWST-initialized structure
Greedy search in the space of equivalent classes