Predicting Hospital Readmission Rates Using Data Mining Techniques.

Area of Science:

• Health Informatics
• Data Mining in Healthcare
• Predictive Analytics

Background:

• Rising hospitalization costs and increasing patient readmissions strain healthcare resources.
• Effective prediction of readmission risk is essential for optimizing patient care and hospital services.

Purpose of the Study:

• To predict patient readmission risk using data mining techniques.
• To identify key factors contributing to hospital readmissions in a large subspecialty hospital.

Main Methods:

• Retrospective cohort study utilizing the CRISP-DM methodology.
• Analysis of 47,892 electronic medical records from August 2018 to August 2019.
• Application of neural networks and C5 decision tree algorithms for pattern extraction and prediction.

Main Results:

• Neural network model identified discharge type (0.28), inpatient department (0.21), and length of stay (0.16) as significant predictors.
• C5 decision tree highlighted length of stay (0.12), number of medications (0.11), and discharge type (0.10) as influential factors.
• Overall readmission rate was 11.95%, with models achieving 61.2% accuracy in prediction.

Conclusions:

• Discharge type, inpatient department, length of stay, and number of medications are critical factors in patient readmission.
• Data mining models provide valuable insights into readmission risk factors.
• Implementing data mining for readmission prediction can enhance healthcare decision-making and resource allocation.