Leveraging remote blood pressure monitoring and interpretable machine learning to improve clinical workflows for hypertensive disorders of pregnancy

点击翻译按钮获取中文摘要

Project Summary/Abstract: Hypertensive disorders of pregnancy (HDP) are a leading cause of pregnancy-related deaths in the United States. Specific interventions, such as nutrition counseling and prophylactic aspirin use, are known to prevent the onset and exacerbation of HDP. However, current approaches to identify patients early in pregnancy are limited due to challenges collating patient data from the electronic health record (EHR) and low precision and recall of traditional rules-based medical calculators. Machine learning (ML) methods that can flexibly capture complex relationships between HDP risk factors offer a potential solution, but often only render a static prediction at one time point and do not update as additional information is collected during pregnancy. The objective of this project is to develop a clinically actionable machine learning model that updates dynamically as patients track blood pressure throughout their pregnancies. Specifically, in Aim 1, we will assess the increased predictive power of utilizing blood pressure measurements arising from remote blood pressure monitoring (RBPM) as compared to in-office measurements. We will phenotype patient blood pressure trajectories and investigate associations between phenotypes and HDP diagnosis. In Aim 2, we will use a Bayesian machine learning approach to incorporate the RBPM phenotypes developed in Aim 1 to enhance an existing static HDP model built on EHR data. The developed model will be able to assess patients at multiple time points throughout their pregnancy based on their at-home BP measures. Finally, in Aim 3, we will conduct a mixed-methods study with obstetricians and certified nurse midwives to build a user-centered display that effectively communicates the results from the dynamic model. The project outlined in this proposal will give obstetricians a clinically interpretable tool – BotoML – to help them identify patients that would benefit from intervention early in their pregnancy. The completion of these aims will enable a future Phase II to deploy and prospectively validate BotoML in geographically diverse provider and patient populations.