Harnessing Machine Learning to Predict Low Birth Weight in Newborns

Research conducted by scientists at the University of São Paulo (USP) has demonstrated that advanced machine learning algorithms can effectively predict which pregnancies are at risk of resulting in low birth weight (less than 2.5 kg). Babies born with low birth weight are significantly more vulnerable, being 20 times more likely to face mortality and at increased risk of developing long-term health issues such as neurological and cardiovascular diseases, diabetes, and growth delays. Early prediction allows healthcare providers to implement targeted interventions, which may include nutritional support, enhanced prenatal care, and lifestyle counseling, ultimately improving health outcomes for both mother and child.

This pioneering study utilized data from 1,579 pregnant women within the Araraquara cohort, a population-based project in São Paulo's interior. Notably, it is the first application of such sophisticated machine learning techniques in Brazil for this purpose, and it addresses a gap in research predominantly based on data from high-income countries.

The team tested four algorithms: Random Forest, XGBoost, LightGBM, and CatBoost, with XGBoost emerging as the most accurate in identifying high-risk pregnancies. The model relies on simple, routinely collected variables such as maternal age, socioeconomic status, anthropometric data, and prenatal care access, making it practical for resource-limited settings.

Findings indicate that the model performs well within southeastern Brazil, but caution is necessary when applying it to other regions like the Amazon or African countries, where population-specific adjustments are required for precise predictions. The research underscores the importance of localized data in refining predictive tools, ultimately aiding public policy and prenatal management.

The study emphasizes that integrating machine learning into prenatal care can facilitate early interventions, reducing low birth weight rates and associated complications. Furthermore, the unique data from the Araraquara cohort, encompassing diverse sociodemographic and environmental factors, enhances the model’s relevance. This approach highlights the potential for artificial intelligence to transform maternal and child healthcare and reduce disparities across different socioeconomic and geographical contexts.

Source: medicalxpress.com