Building and validating 5-feature models to predict preeclampsia onset time from electronic health record data

Abstract Background Preeclampsia is a potentially fatal complication during pregnancy, characterized by high blood pressure and presence of proteins in the urine. Due to its complexity, prediction of preeclampsia onset is often difficult and inaccurate. Methods This study aims to create quantitative models to predict the onset gestational age of preeclampsia using electronic health records. We retrospectively collected 1178 preeclamptic pregnancy records from the University of Michigan Health System(UM) as the discovery cohort, and 881 records from the University of Florida Health System(UF) as the validation cohort. We constructed two Cox-proportional hazards models with Lasso regularization: one baseline model utilizing maternal and pregnancy characteristics, and the other full model with additional lab results, vital signs, and medications in the first 20 weeks of pregnancy. We built the models using 80% of the UM data and subsequently tested them on the remaining 20% UM data and validated with UF data. We further stratified the patients into high and low risk groups for preeclampsia onset risk assessment. Findings The baseline model reached C-indices of 0·64 and 0·61 in the 20% UM testing data and the UF validation data, respectively, while the full model increased these C-indices to 0·69 and 0·61 respectively. Both the baseline and full models contain five selective features, among which number of fetuses in the pregnancy, hypertension and parity are shared between the two models with similar hazard ratios. In the baseline model, history of complicated type II diabetes and a mood/anxiety disorder during the first 20 weeks of pregnancy were important. In the full model, maximum diastolic blood pressure in early pregnancy was the predominant feature. Interpretation Electronic health record data provide useful information to predict gestational age of preeclampsia onset. Stratification of the cohorts using five-predictor Cox-PH models provide clinicians with convenient tools to assess the patients’ onset time of preeclampsia. Funding This study was supported by grants through the NIEHS, NICHD, NIDDK, and NCATS.