Identifying Predictors of Heart Failure Readmission in Patients From a Statutory Health Insurance Database:...

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Cardiology Research

Background:

• Heart failure (HF) patients are frequently readmitted, often for noncardiovascular reasons.
• Understanding outpatient HF management is key to reducing readmissions.
• Machine learning (ML) on health insurance data offers insights into large, representative populations.

Purpose of the Study:

• To assess ML's ability to predict 1-year all-cause and HF-specific readmissions.
• To identify key predictors of readmission in patients with HF.
• To utilize outpatient statutory health insurance (SHI) data for predictive modeling.

Main Methods:

• Utilized SHI data (AOK Baden-Württemberg) from 2012-2018 for individuals with HF.
• Trained and applied ML algorithms (Random Forest, Elastic Net, etc.) to predict readmissions.
• Included diagnosis codes, drug exposures, demographics, rurality, and disease management program participation as predictors.

Main Results:

• The Random Forest model achieved the highest predictive accuracy (C-statistic 0.68 for all-cause, 0.69 for HF-specific).
• Key predictors for all-cause readmission included pantoprazole use, COPD, atherosclerosis, sex, rurality, and participation in diabetes/CHD management programs.
• HF-specific readmission was strongly associated with HF comorbidities.

Conclusions:

• Identified known HF comorbidities and novel predictors for readmission risk.
• Disease management programs can help target high-risk HF patients with comorbidities.
• Rural location and factors beyond comorbidities significantly influence HF readmission risk, informing outpatient care strategies.