From climate-health epidemiology to predictive modeling: A narrative review of machine learning applications in maternal health in vulnerable regions

Background: Climate change is increasingly recognized as a significant threat to maternal and neonatal health, particularly in low- and middle-income countries (LMICs). While robust epidemiological evidence links environmental stressors like extreme heat and air pollution to adverse pregnancy outcomes, a gap persists between these findings and current clinical predictive models.

Objective: This narrative review examines the evolution of maternal risk research, tracing its development from climate-health epidemiology to advanced machine learning (ML) predictive modeling.

Methods: We synthesize existing literature across three distinct research phases, highlighting methodological patterns, key limitations, and persistent gaps in the field, with a specific focus on climate-vulnerable regions such as sub-Saharan Africa and South Asia.

Results: Our synthesis reveals that most contemporary ML models rely exclusively on clinical and demographic data, largely neglecting environmental and climate variables. We identify that successful integration faces significant challenges, including the need for interoperable data infrastructure, external model validation across diverse climates, and the active engagement of frontline community health workers.

Conclusion: We propose a "Hybrid Framework" that combines clinical, environmental, and adaptive ML approaches, supported by digital health tools. This model represents the most promising pathway toward building climate-resilient maternal health systems and enhancing health equity in resource-constrained settings.