Patient Stratification Using Electronic Health Records from a Chronic Disease Management Program

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Chronic Disease Management

Background:

• Hypertension patient care requires effective prognostic stratification for personalized treatment plans.
• Care coordination programs aim to optimize resource allocation and patient outcomes.
• Variability in patient response to therapy necessitates tailored management strategies.

Purpose of the Study:

• To develop a machine learning (ML) framework for prognostic stratification of hypertension patients.
• To enable personalized care plan design and effective resource allocation in care coordination.
• To identify patient subgroups with distinct disease profiles and treatment responsiveness.

Main Methods:

• Utilized a de-identified cohort of 2,521 hypertension patients from electronic health records (EHRs).
• Applied stepwise and logistic regression to identify risk factors for blood pressure changes.
• Employed model-based clustering to group patients based on identified risk factors.

Main Results:

• Identified a set of predictive features including demographics, medications, and diagnostics.
• Achieved an Area Under the ROC Curve (AUC) of 0.71, indicating good predictive performance.
• Clustered patients into four clinically meaningful groups: two severe and two mild disease profiles.

Conclusions:

• Clustering analysis generates homogeneous patient groups, aiding customized care program design.
• Predictive modeling and clustering using EHR data offer a systematic approach for tailored patient management.
• This ML framework supports care providers in optimizing hypertension management based on patient-level factors.