Modeling the impact of multiple hazards on the Maternal and Child Health System in Zambia

Extreme weather events (floods and heatwaves) are becoming more frequent and intense due to climate change, posing significant risks to maternal and child health (MCH). These events interact in complex ways, occurring as compounding hazards (simultaneous or overlapping events), multiple hazards (independent but co-occurring risks), or cascading hazards (where one event triggers or exacerbates another). Understanding these interactions is critical for assessing their full health impacts and improving health system resilience. To date, health-related research has primarily focused on the effects of each hazard individually. This study employs an integrated framework that combines copula models, Bayesian networks, and machine learning approaches to analyse multi-hazard interactions, focusing on Zambia as a case study. Data, including daily rainfall and temperature, MCH-related datasets, utilisation data, and health system performance metrics – such as antenatal care (ANC), postnatal care (PNC), childhood immunisation, place and mode of delivery, and health service utilisation records – were obtained from Zambia through the REACH project. Daily rainfall was merged with TAMSAT and ERA5 reanalysis data (weather station data) to identify flood and heatwave events across Zambia from 1981 to 2023. Copula models were used to capture non-linear dependencies between heatwaves and floods; Bayesian networks uncovered causal pathways linking hazards with MCH and utilisation outcomes; and machine learning models (e.g., random forests and neural networks) predicted health impacts and identified critical patterns of hazard-MCH interactions. Intermediate variables, such as demand-side factors (e.g., education, wealth, age, etc.) and supply-side factors (e.g., facility density, health worker density, and healthcare financing), were incorporated to improve causal inference and identify actionable pathways. Marginal distributions for temperature and precipitation extremes were modelled using extreme value theory, while copulas quantified the joint probabilities of simultaneous extremes. Bayesian networks provided insights into cascading effects, such as how flooding damages healthcare infrastructure and exacerbates the impact of heatwaves on MCH services. Machine learning models were then trained to predict MCH outcomes (utilisation rates and counts) based on these multi-hazard interactions, leveraging their capacity to handle complex, non-linear relationships. Key results focus on estimating the level of ANC and PNC service disruption caused by compounding hazards, such as simultaneous floods and heatwaves. There is an urgent need for climate-resilient healthcare systems and targeted interventions to mitigate the risks of interacting with climate extremes on MCH. Such disruptions are anticipated to highlight important predictive factors, including increased rates of preterm births and maternal complications. This integrated approach, combining statistical, causal, and predictive tools, offers a holistic framework for analysing multi-hazard interactions and their impact on maternal and child health outcomes. By focusing on Zambia as a case study, this research aims to generate insights that are both contextually relevant and scalable for global application. 

Keywords: Multiple hazards, maternal and child health, machine learning, copula models, Bayesian networks 

 Acknowledgements 

This work was conducted under the framework of the Economic and Social Research Council grant: Building Resilience to Floods and Heat in the Maternal and Child Health Systems in Brazil and Zambia (REACH), Grant Number: ES/Y00258X/1