Data Sheet 1_Deep learning guided propofol ketamine dosing and inflammation trajectories i

Data Sheet 1 contains data from a retrospective multi-cohort analysis of 614 perioperative episodes in elderly burn patients. The study, authored by Xiaohui Yuan and last updated in May 2026, used data from the MIMIC-IV and eICU databases to develop machine learning models for predicting inflammatory marker trajectories and optimizing anesthesia dosing. External validation was performed on 206 independent episodes.

Use Cases

• Training predictive models for inflammatory marker trajectories based on IL-6, TNF-α, and CRP measurements.
• Developing offline reinforcement learning agents for personalized propofol-ketamine dosing recommendations.
• Analyzing the relationship between anesthetic dosing and hemodynamic stability based on mean arterial pressure (MAP) targets.
• Validating machine learning models for detecting inflammatory spikes in elderly burn patients.

Strengths

• Dataset includes 614 perioperative episodes with external validation on 206 independent episodes.
• Model performance metrics are provided, including a 70.1% improvement in IL-6 forecasting error over baseline.
• Data is sourced from two established public ICU databases (MIMIC-IV and eICU).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is specific to elderly burn patients (≥65 years), limiting generalizability to other populations.

Provenance

Source

MIMIC-IV and eICU Collaborative Research Database.

Collection Method

Retrospective multi-cohort analysis of ICU data.

Freshness

Last updated 2026-05-18 05:41:24; freshness should be verified.

Geography

Academic medical center ICU and 208 community and academic hospitals in the United States.