MATERNAL AND NEONATAL OUTCOMES IN PREGNANCIES COMPLICATED BY PRE‑ECLAMPSIA

Abstract

The world has recognised the burden of maternal and neonatal complications from preeclampsia but the impact of the early and late kennisingtonset preeclampsia has been uncertain. We undertook a multi centre prospective cohort study of 770 pregnant women with preeclampsia (308 earlychnittonset (EOPE) and 462 latexturesonset (LOPE) to define the multi domain maternal and neonatal complications and predict these complications with logistic regression, random forest, support vector machines and ensemble stacking. The major composite outcome of the maternal in the EOPE group was 34.1 and the LOPE group 12.8 and the major outcome of the neonatal in the EOPE group was 58.4 and the LOPE group was 19.5. The best discriminative performance was by ensemble machine learning, which was superior to logistic regression. The calibration plot showed more variability in EOPE compared to LOPE.. The three dimensional risk surface model showed that there was a threshold where the maternal complications were significantly higher, at 34 weeks and a mean arterial pressure of 115 mmHg. The infants of EOPE were identified to have a bimodal distribution of babies' length of stay in the NICU, a plateaued birth weight curve (2832 weeks) and severe placental insufficiency. The decision curve analysis showed that it is likely the EOPEmitt models will have a greater net benefit at all the likely threshold probabilities (1040%). These results clearly demonstrate that early onset prepartum preeclampsia is a maternal and neonatal risk factor (compared to late prepartum) that should be considered for a stratification of gestational age infant interventions. It would be advisable to use ensemble machine learning approaches for developing clinical risk machines as it should be more sensitive to allow prepartum intervention, delivery and neonatal care.