Global patterns of air pollution-attributable neonatal preterm birth mortality rates: a machine learning analysis using LSTM autoencoder and K-means clustering.

Neonatal preterm birth is a leading cause of neonatal mortality worldwide, and maternal exposure to air pollution is increasingly recognized as a contributing factor. Quantifying the global burden of deaths attributable to air pollution is essential for guiding interventions. Using data from the Global Burden of Disease (GBD) study, we examined country-level neonatal preterm birth deaths attributable to air pollution between 1990 and 2023. A Long Short-Term Memory (LSTM) autoencoder was employed to capture temporal patterns and anomalies in mortality rates, and K-means clustering (K-means) grouped countries with similar trajectories. Results showed that mortality attributable to air pollution declined in most countries, with annual slope values ranging from -0.79 to +0.04. Despite these declines, mean mortality rates varied substantially, from 0.03 per 100,000 live births in Japan and Monaco to 29.26 in Mali. Countries, such as Bangladesh, Bhutan, Laos and Guatemala demonstrated steep declines, whereas Mali, Guinea-Bissau, Niger and Sudan maintained persistently high mortality. Clustering revealed four distinct groups of countries, highlighting differences in burden and progress. Overall, neonatal preterm birth mortality attributable to air pollution has declined globally but at uneven rates, with the greatest progress in parts of Asia and the highest persistent burden in sub-Saharan Africa, underscoring the need for targeted interventions.