Federated learning for 30-day readmission prediction: A controlled evaluation of federation effects versus algorithm standardization across 47 healthcare institutions

Key Points

• This research aims to evaluate the effectiveness of federated learning in predicting 30-day hospital readmissions compared to traditional algorithm standardization.
• Conducted a controlled evaluation across 47 healthcare institutions.
• Compared outcomes from federated learning and standardized algorithms.
• Utilized readmission data over a 30-day period.
• Federated learning improved prediction accuracy compared to standardization, enhancing detection rates significantly.
• Achieved notable performance gains in diverse clinical settings, indicating broader applicability.
• Highlighted the potential for federated models to leverage distributed data without compromising privacy.