Clinical Records for 2,691 ICU Patients with Traumatic Brain Injury, 2008-2019

A dataset of 2,691 clinical records for ICU patients with traumatic brain injury, used to develop a mortality prediction model. The data includes demographics, comorbidities, vital signs, lab results, severity scores, and interventions, sourced from the MIMIC-IV database between 2008 and 2019. The study by Ning Liu, published on figshare, identified a final set of 12 predictive variables, with in-hospital mortality occurring in 11.7% of patients.

Use Cases

• Training mortality prediction models based on demographic and clinical metrics like age and Glasgow Coma Scale (GCS).
• Analyzing non-linear risk relationships based on features such as body temperature and glucose levels.
• Benchmarking machine learning algorithms for clinical tasks based on the described AUROC and Brier score performance metrics.
• Applying SHAP analysis for model interpretability based on the identified top mortality drivers like anion gap.

Strengths

• Includes 2,691 patient records with a defined outcome (11.7% in-hospital mortality).
• Data spans 12 years (2008-2019) from the established MIMIC-IV clinical database.
• The underlying study employed a rigorous, multi-method feature selection process to identify 12 key predictors.
• The final XGBoost model achieved a reported AUROC of 0.873, indicating strong predictive performance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the downloadable data file is unknown, which may limit suitability assessment.
• Data may reflect temporal and institutional bias inherent to the single-source MIMIC-IV database.

Provenance

Source

MIMIC-IV database

Collection Method

Retrospective analysis of clinical records

Time Range

2008-2019

Freshness

Last updated 2026-04-23 04:19:10; freshness should be verified.

Geography

null