A machine learning model to predict heart failure readmission: toward optimal feature set.

Area of Science:

• Cardiology
• Health Informatics
• Machine Learning

Background:

• High hospital readmission rates for heart failure patients persist.
• Predictive modeling with big data offers potential for risk identification and care management.
• Large datasets can present performance challenges in predictive analytics.

Purpose of the Study:

• To develop a machine learning model for predicting 30-day heart failure readmissions using a national database.
• To identify an optimal feature set maximizing the Area Under the Curve (AUC) for the prediction model.

Main Methods:

• Utilized data from the 2020 Nationwide Readmissions Database for heart failure patients.
• Employed a heuristic feature selection process with logistic regression and random forest models.
• Evaluated model discrimination using accuracy, sensitivity, specificity, and AUC.

Main Results:

• Analyzed 566,019 heart failure discharges; readmission rates were 8.9% (same-cause) and 20.6% (all-cause).
• Random forest models achieved higher AUCs (0.607 same-cause, 0.576 all-cause) than logistic regression.
• Identified optimal feature sets (20-22 variables) including age, payment method, chronic kidney disease, and post-care encounters.

Conclusions:

• The developed model demonstrated comparable discrimination to studies with smaller datasets.
• Reducing dataset size improved performance, highlighting big data complexity.
• Heuristic feature selection effectively leveraged nationwide data for predicting heart failure readmissions.