Derivation and external validation of risk stratification models for severe maternal morbidity using prenatal encounter diagnosis codes

Objective We sought to develop a prediction model using prenatal diagnosis codes that could help clinicians objectively stratify a women’s risk for delivery-related morbidity. Study design We performed a prospective cohort study of women delivering at a single academic medical center between 2016 and 2019. Diagnosis codes from outpatient encounters were extracted from the electronic health record. Standard and common machine-learning methods for variable selection were compared. The performance characteristics from the selected model in the training data set—a LASSO model with a lambda that minimized the Bayes information criteria—were compared in a testing and external validation set. Results The model identified a group of women, those in the highest decile of predicted risk, who were at a two to threefold increased risk of maternal morbidity. Conclusion As EHR data becomes more ubiquitous, other data types generated from the prenatal period may improve the model’s performance.