Title: Predicting pregnancy complications: Models and challenges
Authors: Shalem Leemaqz - Robinson Research Institute, University of Adelaide (Australia) [presenting]
Gus Dekker - Robinson Research Institute - University of Adelaide (Australia)
Claire Roberts - Robinson Research Institute - University of Adelaide (Australia)
Abstract: An estimated 25\% of first pregnancies are affected by Preeclampsia (PE), preterm birth (PTB), intrauterine growth restriction (IUGR) and/or gestational diabetes mellitus (GDM). However, providing interventions to prevent these pregnancy complications is difficult because we are not able to identify which nulliparous women are at risk. Modelling of these complications is challenging due to the complex underlying relationships with a vast number of potential risk factors and co-existing complications. We propose a tiered prediction strategy using penalised regression and integrating multiple systems to estimate the occurrence and co-occurrence probability of each complication for simultaneous prediction of multiple pregnancy complications. The first tier, aimed at a high sensitivity, classifies women who are deemed to be at risk or at low risk. The second tier, with a higher positive predictive value, further stratifies at risk women into high or moderate risk. We utilised data from the Screening for Pregnancy Endpoints (SCOPE) study which consists of detailed clinical and lifestyle variables of pregnant women in Adelaide, Australia. We have also explored visualisation of the concurrent prediction models and development of a learning system through backpropagation feedback. This may assist in providing tailored antenatal care and early interventions that could benefit both the mother and child.