Identifying High-Need Primary Care Patients Using Nursing Knowledge and Machine Learning Methods.

Area of Science:

• Healthcare Informatics
• Computational Medicine
• Nursing Science

Background:

• Machine learning (ML) models can segment patient cohorts using clinical data.
• Integrating clinical knowledge enhances the translational value of ML-driven patient segmentation.
• A practical approach combines medical, behavioral, and social factors for patient segmentation.

Purpose of the Study:

• To demonstrate a pragmatic application of ML for rapid patient cohort generation using unsupervised classification.
• To showcase the increased translational value of ML models through nursing knowledge integration.
• To create meaningful patient cohorts for improved care coordination.

Main Methods:

• Utilized unsupervised k-means cluster analysis on a primary care dataset of 1233 diabetic patients.
• Expert nurses selected variables based on critical care coordination factors.
• Nursing knowledge was applied to interpret and describe psychosocial phenotypes of identified clusters.

Main Results:

• Identified four distinct patient clusters with unique psychosocial need profiles.
• Cluster 1: Diverse females, non-English speakers, low medical complexity, childhood illness history.
• Cluster 2: English speakers with obesity and respiratory disease; Cluster 3: Males with substance use disorder and multiple comorbidities; Cluster 4: Older patients with renal failure.

Conclusions:

• Developed a practical method for analyzing primary care data using ML and clinical expertise.
• The identified clusters allow for direct translation into actionable social and medical care plans.
• This integrated approach enhances patient segmentation for targeted interventions.