ICU Patient Subsyndromal Delirium Risk Factors and Prediction Model

A prospective cohort study collected data from ICU patients between October 2024 and May 2025 to predict subsyndromal delirium (SSD). The dataset was used to compare seven machine learning algorithms, with XGBoost identified as the best model (AUC=0.84). The study, authored by Ying Li and shared on figshare, identified four key predictive factors for SSD.

Use Cases

• Building a risk prediction model for subsyndromal delirium based on identified factors like vasoactive drug use and monthly household income.
• Comparing the performance of machine learning algorithms like Random Forest, XGBoost, and Logistic Regression for clinical prediction tasks.
• Identifying high-risk ICU patient subgroups for targeted intervention based on feature importance analysis.

Strengths

• The study reports a specific SSD prevalence rate of 37.158% within its cohort.
• Model performance is quantified with an AUC score of 0.84 for the best-performing algorithm (XGBoost).
• Feature importance analysis provides quantified weights for four predictive factors.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale modeling.
• Data is from a single-center prospective study; the authors note a need for validation in larger, multicenter cohorts.

Provenance

Source

figshare

Collection Method

Prospective cohort study of ICU patients.

Time Range

October 2024 to May 2025

Freshness

Last updated 2026-04-28 05:24:01; freshness should be verified.

Geography

null