Predictive Models for Sustained, Uncontrolled Hypertension and Hypertensive Crisis Based on Electronic Health Record...

Area of Science:

• Cardiovascular Medicine
• Health Informatics
• Machine Learning in Healthcare

Background:

• Uncontrolled hypertension poses significant risks, necessitating proactive disease management strategies.
• Early identification of patients with uncontrolled hypertension is crucial for timely intervention.
• Predicting disease progression aids in optimizing patient care pathways.

Purpose of the Study:

• To develop and validate predictive models for sustained uncontrolled hypertension and hypertensive crisis.
• To assess the performance of machine learning models using electronic health record data.
• To evaluate model generalizability across different populations.

Main Methods:

• Utilized a large dataset of 142,897 patients with uncontrolled hypertension.
• Developed and compared four machine learning frameworks: logistic regression, multilayer perceptron, gradient boosting, and random forest.
• Employed 10-fold cross-validation and internal-external cross-validation for robust model assessment.

Main Results:

• The hypertensive crisis model demonstrated strong predictive performance (C-statistic 0.81 internally, 0.79 externally).
• The sustained uncontrolled hypertension model showed moderate performance (C-statistic 0.72 internally, 0.70 externally).
• Both models outperformed default strategies, indicating clinical utility.

Conclusions:

• Electronic health record-based models show promise for predicting hypertensive crisis.
• These models can support population health surveillance and hypertension management efforts.
• Further research is needed to enhance prediction accuracy for sustained uncontrolled hypertension.