Predicting 30-day readmission using DRG-based hospitalization data: a large real-world logistic regression model from a tertiary hospital.

Background: Early unplanned readmission is a key quality indicator in Diagnosis-Related Groups (DRG)-based payment systems. Despite China's rapid expansion of DRG reform, evidence on hospital-wide predictors of 30-day readmission using large-scale real-world data from tertiary hospitals remains limited. This study developed and evaluated a DRG-based logistic regression model for predicting 30-day readmission. Methods: We conducted a single-center retrospective study using administrative hospitalization data from a high-volume tertiary hospital in Shanghai, China. We extracted 65,215 inpatient episodes from the hospital (January 2023-December 2024). After excluding discharges in December 2024 due to incomplete follow-up ( Results: The 30-day readmission rate was 13.0%. In unadjusted comparisons, patients who were readmitted had shorter median hospital stays (3 vs. 4 days) and lower total costs. After multivariable adjustment, longer length of stay was associated with increased readmission risk (OR 1.016 per day, Conclusion: Using comprehensive DRG-based real-world data, we developed an interpretable prediction model for 30-day readmission with moderate-to-good discrimination and acceptable calibration. Clinical case-mix captured by DRG categories and patient-level complexity reflected by longer length of stay were key determinants of early readmission. The model may support risk stratification, quality improvement, and performance monitoring in DRG payment environments. The findings may also inform policy discussions on aligning DRG efficiency incentives with patient safety outcomes.