(896) Machine Learning Prediction of Gestational Hypertensive Complications at Point-of-Care

Hypertensive disorders of pregnancy (HDP) pose significant risks to both maternal and fetal health, potentially leading to early induction and NICU admission. Early identification and intervention have been shown to prevent up to 80% of preeclampsia cases. This study aims to develop and evaluate a machine learning model to predict preeclampsia using routinely collected prenatal variables.

Study Design:

An extreme gradient boosting model was developed to predict the risk of HDP, inclusive of preeclampsia and gestational hypertension. The model was trained on data from the American College of Surgeons National Surgical Quality Improvement Program database comprising prenatal and demographic variables from a large cohort of pregnant women from 2018 to 2021 and tested on data from 2022.

Inclusion criteria were term deliveries without missing data for hypertensive complications. Patients missing hypertensive complication data were excluded. The model incorporated clinical history, demographic information, and early-stage comorbidities. To address class imbalance, weight scaling algorithms were utilized and patients from 2014-2017 with hypertensive complications to enhance the representation of the positive class, with the exception of 2015 patients due to corrupted data.

Results:

35 factors across 11,298,212 patients were used to develop the model. Of these patients, 1,551,321 (13.7%) had prenatal hypertensive complications, The model was tested on 3,258,442 total patients, of which 907,974 (27.9%) had hypertensive complications. The XGBClassifier model achieved an AUC-ROC of 0.70, with a 74.9% accuracy, 48.2% sensitivity and 0.75 F1 score. 

Conclusion:

Early-stage predictive models used at the point of care can detect nearly half of all hypertensive complications at the point of care with discriminative ability. Early identification through such models may allow for targeted clinical interventions, harboring potential for the prevention of hypertensive cases and cost savings for payors.