Prediction of stillbirth and perinatal morbidity with machine learning
This project involves application of machine learning methods to predict the risk of stillbirth, preterm birth, and low birth weight by analysing more than 4.5 million births in Australia (WA, NSW, SA, NT). Predicting perinatal mortality and morbidity in early pregnancy remains elusive as clinical management cannot currently make full use of the information provided by the best predictors – family and obstetric history – which could be provided with risk stratification. Stratification of risk is necessary to objectively support clinical decisions for prevention of avoidable perinatal morbidity and mortality. Although single risk factors are aetiologically important, prognosis (prediction) and clinical management require an individualised approach that uses all available risk factors to substantially improve primary prevention.
In over three decades there has been little improvement in the stillbirth rate, which is also markedly higher for Aboriginal Australians than Caucasians and other races/ethnicities.
So far we have discovered that almost half of stillbirths can be potentially identified antenatally based on a combination of current pregnancy complications, congenital anomalies, maternal characteristics, and medical history. Greatest sensitivity is achieved when current pregnancy complications are used for prediction. Machine learning classifiers (ensemble methods) offer some improvement for prediction compared to the classical method most often applied in medical research (logistic regression).
Project lead: Prof Gavin Pereira
Collaborators: Curtin Institute for Computation, Joondalup Health Campus, King Edward Memorial Hospital, Ramsay Healthcare, Royal Women’s Hospital, Telethon Kids Institute, University of Sydney
Publication(s)