Identifying Risk of Stillbirth Using Machine Learning

Abstract

(Am J Obstet Gynecol. 2023;229:327:e1–16) A tool that could help identify women at risk for stillbirth would be a welcome addition to clinical practice. Machine learning models, which can look at relationships between variables in large data sets, was investigated in this article. This study’s aim was to refine a machine learning model to predict stillbirth using data available prior to the gestational age of viability (22 to 24 wk). The researchers performed a secondary analysis of data in the Stillbirth Collaborative Research Network (SCRN), a study database of 982 stillbirths and 3000 representative live births from 2006 to 2009. The study included patients who delivered a live or stillborn fetus at >18 weeks’ gestation from 59 hospitals in 5 states (Georgia, Massachusetts, Rhode Island, Texas, and Utah). The data set included 6000 measures collected from interviews with the mother, postmortem pathological exam, cause of death analysis, and medical record Abstractsion. Subsequently 101 risk factor variables were identified and used in the model. These included maternal and biological father demographics, family history, maternal health history, prenatal lab data, fetal genetics, second-trimester screen, prenatal care history, ultrasound, and social determinants of health.