Traumatic Brain Injury Patient Data for Delirium Prediction from MIMIC-IV and eICU-CRD

915 TBI patients from the MIMIC-IV database and 317 from the eICU-CRD database were used to develop a machine learning model for predicting incident delirium. The Random Forest model achieved an internal AUC of 0.819 and an external AUC of 0.706. The study, authored by Cheng Li and last updated in May 2026, includes predictors like Glasgow Coma Scale and invasive ventilation, with SHAP analysis for interpretability.

Use Cases

• Developing early warning systems for delirium based on clinical features like Glasgow Coma Scale and invasive ventilation status.
• Evaluating model generalization across different patient subgroups, such as younger patients or those with extracranial injuries.
• Creating interpretable risk stratification frameworks using SHAP analysis to understand feature contributions.
• Benchmarking machine learning algorithms (e.g., Random Forest) for clinical prediction tasks in traumatic brain injury.

Strengths

• Model was developed and externally validated on two distinct ICU databases (MIMIC-IV and eICU-CRD).
• Performance metrics are provided, including AUC values (0.819 internal, 0.706 external) and clinical net benefits.
• Key predictors are identified and analyzed through SHAP, offering interpretability for clinical use.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying patient data is unknown, which may limit suitability assessment.
• The primary file format is DOCX, which may require extraction or conversion for direct analysis.

Provenance

Source

MIMIC-IV and eICU-CRD databases.

Collection Method

Retrospective study of patients diagnosed with traumatic brain injury.

Freshness

Last updated 2026-05-21 14:01:15; freshness should be verified.