Machine Learning Model for Predicting AKI Risk in Hemodynamically Stable Sepsis Patients

A study based on the MIMIC IV database developed a machine learning model to predict acute kidney injury (AKI) in sepsis patients. The dataset includes 8,276 hemodynamically stable sepsis patients, with 3,061 (37%) experiencing AKI. The model was developed by Miao He and last updated on 2026-05-26.

Use Cases

• Train a risk prediction model based on nine identified clinical risk factors.
• Benchmark machine learning algorithms like XGBoost and Random Forest for clinical outcome prediction.
• Validate predictive models for sepsis-associated complications using external hospital data.

Strengths

• Includes 8,276 patient records from the MIMIC IV database.
• Identifies nine specific risk factors for model development.
• External validation was performed using clinical data from a hospital.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying data is unknown, which may limit suitability assessment.
• The dataset is a 217.4 KB DOCX file, indicating limited scope and likely contains model documentation rather than raw data.

Provenance

Source

MIMIC IV Database and an external hospital.

Collection Method

Clinical data extraction and analysis.

Freshness

Last updated 2026-05-26 05:26:00; freshness should be verified.