Development and validation of risk prediction models for adverse maternal and neonatal outcomes in severe preeclampsia in a low-resource setting, Mpilo Central Hospital, Bulawayo, Zimbabwe.

ABSTRACT

OBJECTIVES: Hypertensive disorders of pregnancy are major causes of global maternal and neonatal morbidity and mortality. This study aimed to develop and validate models to predict composite adverse maternal and neonatal outcome in severe preeclampsia in low-resource settings. STUDY DESIGN: A retrospective cross-sectional study of women with severe preeclampsia giving birth in a tertiary referral centre in Zimbabwe between 01/01/2014-31/12/2018. Candidate variables identified from univariable logistic regression (p < 0.2) were entered into stepwise backward elimination logistic regression models to predict composite adverse maternal and neonatal outcomes. Models' performance was assessed by the area under the curve of the receiver operator characteristic (AUC ROC). The models were validated internally using bootstrap-based methods and externally using the Preeclampsia Integrated Estimate of RiSk dataset. MAIN OUTCOME MEASURES: The co-primary outcomes were composite adverse maternal outcome and composite adverse neonatal outcome. RESULTS: 549 women had severe preeclampsia from whom 567 neonates were born. The predictive model for composite adverse maternal outcome included maternal age, gestational age on admission, epigastric pain, vaginal bleeding with abdominal pain, haemoglobin concentration and platelets; the AUC ROC was 0.796 (95% CI 0.758-0.833). External validation showed poor discrimination (AUC ROC 0.494, 95% CI 0.458-0.552). The model for composite adverse neonatal outcome included gestational age, platelets, alanine transaminase and birth weight; the AUC ROC was 0.902 (95% CI 0.876-0.927). CONCLUSIONS: While the models accurately predicted composite adverse maternal and neonatal outcomes in the study population, they did not in another cohort. Understanding factors which affect model performance will help optimize prediction of adverse outcomes in severe preeclampsia.