Predictive Model of the Climate Change Impact on Low Birth Weight, a Methodological Approach in Argentina

Introduction: As one of the challenges of this century, climate change has a negative impact on health. The specific application of models to evaluate low birth weight (LBW) in the context of climate change is still limited. Methodology: In 657 urban localities in Argentina, birth weight data of live births from the period 2018–2019 and information on community food environments (CFE) were extracted from various sources of open data and satellite images. Modeling was performed in two stages: 1) using all variables as inputs, and then 2) using only the top 10 most influential variables in predicting LBW. The performance of nine machine learning (ML) models was compared, selecting a final XGBoost model. The models were managed using the MLflow platform. To simulate a possible climate change scenario within 25 years, the annual change rates of the environmental variables included in the top 10 most influential variables were calculated. A spatial autocorrelation analysis was conducted to evaluate clusters, and a hotspot analysis and hotspot comparison were performed to assess the location of clusters and the differences between the original and simulated LBW values. The work was carried out with ArcGIS Pro and QGIS 3.28. Results: The modeling and simulation presented an R2 of 0.88 and 0.87 respectively. Clusters were found in all LBW variables, with high prevalence areas mainly towards the center of the country. The simulated data compared to the original data show a contraction of the high prevalence LBW cluster towards the center. Conclusions: There is an opportunity to apply this methodology for future research that could transform the approach to neonatal health in the context of climate change, providing efficient tools for evidence-based decision-making.