ICU readmissions

ICU readmissions predicts patient risk of readmission to intensive care units at 3, 7, and 30 days using machine learning models trained on the MIMIC-OMOP dataset to support clinical risk stratification.

Key Features:

• Machine Learning Approach: Employs a lightweight tree boosting method for readmission risk prediction.
• Benchmarking on MIMIC-OMOP Dataset: Trained and evaluated on the OMOP Common Data Model (CDM) version of the MIMIC-III dataset (MIMIC-OMOP) for standardized clinical data analysis.
• Time-Specific Predictions: Produces risk predictions at 3-day, 7-day, and 30-day post-discharge time points.
• Performance Metrics: Achieves an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.805 for 3-day readmission predictions, reported to outperform existing solutions.

Scientific Applications:

• Clinical risk stratification: Identifies patients at high risk of ICU readmission to inform targeted clinical interventions.
• Resource allocation and outcome research: Supports planning of healthcare resources and clinical research using OMOP-standardized data to evaluate readmission risk.

Methodology:

Models were trained and evaluated on the MIMIC-OMOP dataset (OMOP CDM version of MIMIC-III) using a lightweight tree boosting method with standardized OMOP-format clinical data.