Predicting Infant Mortality Risk From Information Available at the Time of Birth

Antonia Saravanou, University of Athens
Clemens Noelke , Brandeis University
Nick Huntington, Brandeis University
Dolores Acevedo-Garcia, Brandeis University
Dimitrios Gunopulos, University of Athens

Using birth certificate data for all registered US births over a three year period (2000-2002), we explore whether it is possible to reliably identify infants at risk of dying before their first birthday using information that is routinely gathered at the time of birth. We use four classifiers from the machine learning (ML) literature to predict mortality before the first birthday, as well as age at death (early neonatal, late neonatal, and postneonatal) and cause of death. We also explore whether the quality of predictions varies by maternal race/ethnicity. We find that the best-performing classifier correctly identifies, at the time of birth, 3 out of 4 infants who die before their first birthday. The resulting risk scores can potentially be used to allocate more intensive care within and beyond the clinical setting to infants with a high predicted mortality risk.

See extended abstract

 Presented in Session 3. Population, Development, & the Environment; Data & Methods; Applied Demography