ReAdmit: Predicting Early Unplanned ICU Readmission Using Radiology Notes and Structured Data

Unplanned ICU readmission is associated with poor clinical outcomes and is considered an important indicator of hospital care quality. Accurate prediction of readmission risk can help optimise ICU discharge decisions and improve patient management.

To predict the risk of ICU readmission within 72 h after discharge using structured data and radiology notes collected before ICU discharge for supporting ICU discharge decisions and transitional care management for patients at high risk of early readmission.

A retrospective study was conducted using data from the MIMIC-IV database. The ReAdmit model was developed using six categories of structured data and radiology notes. Patient data from the 24 h before ICU discharge was used for prediction. Clinical BERT was applied to process radiology notes, and the model was built using XGBoost. The model was calibrated with Isotonic Regression before validation. Subgroup analysis was conducted by age, gender and ICU type, and model performance was evaluated using AUROC and standardised readmission rate. The ReAdmit model was compared with the SWIFT score.

Data from 30 714 ICU patients were included, with a 72-h readmission rate of 5.97%. The best-performing model achieved an AUROC of 0.783, outperforming models using only structured data and the SWIFT score (AUROC = 0.625). The median AUROC for subgroup analysis exceeded 0.77, and the overall standardised readmission rate was 1.103 (95% CI: 1.014-1.199).

The ReAdmit model accurately predicts ICU readmission risk within 72 h using routinely collected clinical features and radiology notes, supporting ICU discharge and care transitions.

The ReAdmit model enables early identification of high-risk patients using routine data before ICU discharge. Its clinical integration can aid safer transfer decisions, reduce preventable readmissions and guide post-discharge care such as referrals to high-dependency units or follow-up clinics.