Although child under-five years mortality remains a major public health concern (i.e., nearly 7% of live births in low- and middle-income countries), researchers have struggled to identify a hierarchy of correlates (representing potential causal factors) that can be replicated and that inform prioritization of prevention and intervention efforts. To address this gap, PBS faculty member Kirby Deater-Deckard has been working in an international collaborative group led by Prof. Gianluca Esposito (University of Trento, Italy, and Nanyang Technical University, Singapore) and Dr. Marc Bornstein (NICHD) that also includes team members at Duke University and UNICEF in New York.
Using data from the ongoing UNICEF MICS data system of surveys, the team applied machine learning techniques to identify the largest and globally most pervasive correlates of mortality (e.g., maternal age, household size, cooking fuel source, water quality) in a sample of nearly 300,000 households in 27 low- and middle-income countries. This was followed by longitudinal analysis showing that improvements in these correlates over time were predictive of country-level decreases in under-five mortality. With UNICEF, the research team is planning to develop operational predictive models that can be used by UNICEF field sites to inform decisions about effective resource distribution for mortality prevention efforts.