Table 5_Short-term risk stratification using parallel admission and reassessment features

684 infection-related pediatric intensive care unit admissions from the PhysioNet Paediatric Intensive Care database were analyzed to develop a risk prediction model. The study by Yuhang Wang, last updated in June 2026, compared LASSO, random forest, XGBoost, and a stacked ensemble to predict major adverse events within 72 hours. The random forest model achieved an AUC of 0.724 in an internal test set.

Use Cases

• Developing early risk prediction models based on admission and reassessment features mentioned in the description
• Comparing machine learning algorithms (LASSO, RF, XGBoost) for clinical outcome prediction based on the described methodology
• Conducting SHAP analysis to interpret feature contributions for risk stratification as performed in the study

Strengths

• Model performance metrics are reported, including an AUC of 0.724 and PR-AUC of 0.741 for the random forest model in the internal test set
• The study uses a defined cohort of 684 infection-related PICU admissions from a known database (PhysioNet)
• Validation was performed in both an internal test set and a de-identified timestamp-ordered validation cohort

Limitations

• The dataset is very small at 23.8 KB, indicating limited scope and likely summary-level results rather than raw patient data
• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment for direct ML training

Provenance

Source

PhysioNet Paediatric Intensive Care database

Collection Method

Retrospective analysis of PICU admissions with infection

Freshness

Last updated 2026-06-04 05:44:15; freshness should be verified