Table 3: Short-term Risk Stratification for PICU Patients with Infection

The PhysioNet Paediatric Intensive Care database provided 684 infection-related PICU admissions for this study. Yuhang Wang developed and validated an early risk prediction model for 72-hour major adverse events using admission and reassessment features. The model was last updated on June 4, 2026.

Use Cases

• Predicting 72-hour major adverse events based on admission (M0) features.
• Assessing risk stratification performance using early reassessment (M1) features.
• Comparing machine learning models like LASSO, random forest, and XGBoost for clinical prediction.
• Conducting decision curve analysis for model utility assessment.

Strengths

• Model performance was validated in an internal test cohort and a de-identified timestamp-ordered validation cohort.
• The random forest model achieved an AUC of 0.724 in the internal test set and 0.718 in the validation cohort.
• The dataset is licensed under CC-BY-4.0, facilitating open use.

Limitations

• The dataset is very small at 18.7 KB, indicating limited scope.
• Column-level documentation is absent; field semantics must be inferred after download.
• Calibration was noted to be attenuated in the later validation cohort, suggesting a need for recalibration.

Provenance

Source

PhysioNet Paediatric Intensive Care database

Collection Method

Features extracted from admission (0–6 h) and reassessment (12–36 h) windows.

Freshness

Last updated 2026-06-04 05:44:14