In a recent Nature article, researchers explored a new method to improve prenatal detection of isolated, non-syndromic cyanotic congenital heart disease (CCHD). The team used maternal salivary metabolomics combined with machine learning (ML) to uncover metabolic changes linked to CCHD. Leveraging artificial intelligence (AI) models, the team was able to achieve impressive diagnostic accuracy, addressing some of the limitations of traditional ultrasound screening.
Study: Precision fetal cardiology detects cyanotic congenital heart disease using maternal saliva metabolome and artificial intelligence. Image Credit: Susmit Das/Shutterstock.com
Why Focus on Congenital Heart Disease?
CHD is the most common severe congenital abnormality and remains a leading cause of childhood mortality. Among its various types, CCHD is particularly critical because it requires immediate medical intervention after birth. Yet, detection during pregnancy remains challenging, especially when no obvious risk factors are present.
Although metabolomics and ML have been used in CHD research before, most studies have focused on blood or urine samples. This study stands out by exploring maternal saliva as a diagnostic tool, offering a non-invasive, accessible alternative. Additionally, it examines lipid metabolism disruptions—an area that could deepen our understanding of fetal heart development and lead to more precise interventions.
Study Design and Methodology
The research team conducted a prospective case-control cohort study at Corewell Health William Beaumont University Hospital. Participants were selected based on strict criteria: singleton pregnancies at 14–37 weeks gestation, suspected isolated, non-syndromic CHD confirmed postnatally, and no aneuploidy, syndromic defects, or extracardiac anomalies. Controls were gestational age-matched individuals without CHD.
Saliva samples, collected after fasting, were analyzed using ultra-performance liquid chromatography-mass spectrometry (UPLC-MS). Data was cleaned and normalized, excluding metabolites below detection limits, and advanced ML models—including deep learning (DL), random forest (RF), and support vector machines (SVM)—were employed. The team also incorporated pathway enrichment analysis to pinpoint metabolic disruptions in CCHD cases.
To validate their findings, the researchers used area under the curve (AUC), sensitivity, and specificity metrics, alongside strategies to mitigate overfitting, such as regularization and dropout.
Clinical predictors, such as maternal diabetes and family history, were integrated with metabolite data in AI models to enhance diagnostic accuracy. By combining these data points, the team aimed to reflect real-world scenarios while pushing the boundaries of precision fetal cardiology.
Findings and Analysis
The study looked at how clinical, demographic, and metabolomic factors could be used to identify CHD, with a focus on the more severe form, CCHD. The researchers worked with 80 participants—40 with CHD (split evenly between cyanotic and non-cyanotic cases) and 40 healthy controls. Both groups had similar baseline characteristics, making the comparisons fair.
Using metabolomic profiling, the team identified 468 metabolites, 30 of which showed significant differences in CCHD cases. When these findings were combined with clinical and demographic data, the predictive models performed impressively:
- A model using 20 key markers hit an AUC of 0.8372 in training and 0.8222 in testing, with sensitivity ranging from 91.5 % to 92.5 % and specificity between 87 % and 89 %.
- Another model using only metabolites—25 in total—was just as accurate, achieving an AUC of 0.8488, with 92.5 % sensitivity and 91.0 % specificity.
- For detecting CCHD specifically, a 50-marker model stood out, with an AUC of 0.8198, sensitivity of 92.5 %, and specificity of 87.0 % in the test group.
What’s exciting is that AI models consistently outperformed traditional methods like logistic regression, proving how useful advanced algorithms can be for CHD prediction based on metabolomics.
Insights and Implications
This study is one of the first to explore how maternal saliva and machine learning could work together to detect CCHD in a noninvasive way. The results were impressive, with the models showing strong accuracy and high sensitivity.
A key finding was the disruption of lipid metabolism, with notable changes in triglycerides, diglycerides, and ceramides. These shifts suggest a move toward fatty acid oxidation in CCHD cases, which ties in with earlier studies that found similar patterns in maternal blood and urine samples.
On top of that, the analysis pointed to disruptions in key metabolic pathways, including those involving arachidonic acid, alpha-linolenic acid, and tryptophan. These pathways are crucial for fetal heart development, giving researchers new insights into how CCHD develops and where potential interventions could make a difference.
Of course, there were some limitations, like the relatively small sample size and the lack of fetal blood data to draw more direct connections. But the study’s novelty and potential impact cannot be overlooked. Using saliva as a diagnostic tool is pain-free, accessible, and offers a real alternative to the current prenatal and postnatal screening options.
Conclusion
This study showcased the potential of maternal salivary metabolomics and machine learning as a non-invasive approach to detect CCHD with high accuracy. By uncovering key metabolic changes, particularly in lipid metabolism, the research offers valuable insights into fetal heart development and presents a viable alternative to traditional ultrasound screenings.
While further studies with larger cohorts are necessary, this innovative method has the potential to enhance prenatal diagnostics and support earlier, more effective interventions for congenital heart disease.
Journal Reference
Bahado-Singh et al., 2025. Precision fetal cardiology detects cyanotic congenital heart disease using maternal saliva metabolome and artificial intelligence. Scientific Reports, 15(1). DOI:10.1038/s41598-025-85216-7 https://www.nature.com/articles/s41598-025-85216-7
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.