Researchers have developed machine learning models that analyze routine first-trimester data to identify pregnancies at higher risk of serious complications. The study, published in the Journal of Medical Internet Research, used data from over 770,000 pregnancies across Sweden, Chile, and Singapore, focusing on information available before 14 weeks’ gestation.
The study aimed to improve upon current early prenatal risk assessments, which primarily consider maternal age, prior obstetric history, and existing health conditions. These factors, while important, often fail to capture the full range of medical, social, and demographic influences on pregnancy outcomes.
Model Performance Across Countries
The best-performing models used LightGBM, a machine learning technique, and outperformed existing early risk assessment methods in all three countries. Performance gains were strongest in Sweden and Chile, with a meaningful improvement also seen in Singapore.
The AI systems predicted a composite of serious outcomes, including preterm birth, low birth weight, stillbirth, severe maternal morbidity, and maternal mortality. While the models showed promise, researchers emphasized they should be used as triage aids, not diagnostic tools, to identify pregnancies that may warrant closer monitoring or earlier intervention.
Related: Multiancestry Genetic Score Boosts Alzheimer Risk Forecast
Sweden’s larger dataset, drawn from full national registry records, likely contributed to the model’s stronger performance compared to Singapore, where fewer variables were available for analysis.
Social Factors Influence Risk Predictions
Beyond accuracy, the study highlighted which variables most influenced risk predictions. In Sweden and Singapore, sociodemographic factors like maternal age, BMI, ethnicity, and household income ranked high. In Chile, clinical and obstetric variables, such as antihypertensive use and first-trimester blood sugar levels, carried more weight.
This finding is significant because traditional antenatal risk tools often overlook social determinants of health, which can affect access to care, baseline health status, nutrition, and stress exposure. The study suggests machine learning can help identify patterns across many variables simultaneously, potentially improving pregnancy outcome predictions.
Population Differences Limit Global Model Use
The three cohorts differed in meaningful ways, with Sweden representing a high-resource, lower-morbidity population, Chile having higher rates of preterm birth and obstetric complications, and Singapore’s maternity population being older and more metabolically at risk but with limited data availability. Calibration also varied, indicating that a single global prenatal AI model would not effectively transfer across populations.
Related: ALS Biomarker Reveals New Therapeutic Target
Prospective validation is still needed to confirm whether AI-assisted risk stratification actually improves outcomes. The authors positioned their models as tools to support clinical workflows rather than replacements for clinical judgment.
According to the models, pregnancies with the highest predicted risk of serious complications were more likely to have experienced such an event in the past, have certain clinical characteristics, or come from disadvantaged backgrounds. The study shows the potential of machine learning in improving early prenatal risk assessment, but further research is needed to demonstrate real-world benefits.
The study’s findings highlight the importance of considering social and demographic factors in prenatal risk assessment. Machine learning could help identify combinations of variables that may be overlooked with conventional tools, ultimately supporting a more full understanding of pregnancy risks.
Leave a Reply