Objective:

To evaluate the effectiveness of synthetic data in predicting 30-day hospital readmissions for patients with type 2 diabetes, COPD, and heart failure.

Key Findings:

• Models using synthetic data outperformed those using original data alone, with AUROC values ranging from 0.91 to 0.95.
• Gradient boosting achieved the highest performance in COPD and type 2 diabetes, while extreme gradient boosting excelled in heart failure.
• Higher illness severity scores and greater comorbidity burden were key predictors of readmission.
• Medication nonadherence significantly increased the odds of readmission across all conditions.

Interpretation:

Incorporating synthetic data and social/behavioral risk factors enhances predictive accuracy, supporting multidimensional risk prediction frameworks.

Limitations:

• Single-center dataset without external validation may limit generalizability.
• Potential underdocumentation of social determinants of health could affect findings.
• Lack of evaluation of real-world implementation or cost-effectiveness limits practical application.

Conclusion:

The study highlights the potential of synthetic data in improving readmission prediction models, emphasizing the integration of social and behavioral factors.