Building a Time-Series Model to Predict Hospitalization Risks in Home Health Care: Insights Into Development,...

Area of Science:

• Health Informatics
• Clinical Prediction Models
• Healthcare Disparities

Background:

• Home health care (HHC) serves millions of older adults, yet up to 25% face preventable hospitalizations and emergency department (ED) visits.
• Previous risk prediction models for HHC were limited by underutilized clinical notes, aggregated data, and potential demographic biases.

Purpose of the Study:

• Develop a time-series risk model to predict hospitalizations and ED visits in HHC patients.
• Examine model performance across different prediction windows.
• Identify key predictive variables and assess model fairness across demographic subgroups.

Main Methods:

• Utilized electronic health records, including clinical notes processed via natural language processing, and Medicare claims data.
• Developed a Light Gradient Boosting Machine algorithm for risk prediction, evaluated using 5-fold cross-validation.
• Assessed model fairness across gender, race/ethnicity, and socioeconomic subgroups.

Main Results:

• Achieved high predictive performance (F1 score of 0.84 for a 5-day window).
• Identified 20 top predictive variables, including novel indicators like nurse visit length and frequency.
• Revealed performance disparities, with lower model effectiveness for historically underserved populations.

Conclusions:

• Developed a robust time-series risk model for predicting adverse events in HHC with high accuracy.
• Highlighted the importance of both established and novel risk factors in HHC.
• Emphasized the need for fairness adjustments to ensure equitable risk prediction across all patient populations.