Using Machine Learning Methods to Estimate Maternal Mortality Ratios

Abstract

Maternal mortality is an ongoing, critical international health challenge, especially in low- and middle-income countries. In 2023, the global maternal mortality ratio (MMR), or the number of maternal deaths per 100,000 live births, was 197. Approximately 95% of maternal deaths occur in low and lower-middle income countries and fragile settings. Many of these deaths could have been prevented by using existing, effective interventions. The World Health Organisation has highlighted how low-quality, sparse data about maternal mortality hinders effective intervention, especially as countries with the highest MMRs tend to have the most missing data. The primary contribution of my thesis to the literature was its development of decision-tree based machine learning models to predict the MMRs of 172 countries between 1985 and 2018 using data from the World Health Organisation and World Bank. In contrast to existing approaches, my proposed models estimated MMR without needing to make explicit assumptions about the underlying distribution of data. My models also used a wider range of socio-economic and health-related features than existing methods. This produced an alternative set of MMR estimates that can be used provide consensus about the true MMR values as well as encourage debate about the validity of different MMR modelling techniques. The best-performing Random Forest Stacking Ensemble achieved a test mean relative error of 0.07 when predicting all MMR data for a specific country and a test mean relative error of 0.37 when forecasting MMR values. Despite being limited by low-quality and sparse input data, my models’ MMR predictions were similar to those produced by the most widely used models in the literature, reinforcing their validity. The socioeconomic and health-related variables with highest predictive power for MMR in my models were established risk factors. My analysis highlighted the importance targeting socio-economic drivers of maternal mortality, such as women’s employment prospects, to successfully reduce maternal deaths.