Table 2_Short-term risk stratification using parallel admission and reassessment features

684 infection-related pediatric ICU admissions from the PhysioNet database were used to develop a risk prediction model for major adverse events within 72 hours. The study compared LASSO, random forest, XGBoost, and a stacked ensemble, with random forest achieving an AUC of 0.724 in an internal test set. The model framework uses features extracted from admission (0–6 hours) and early reassessment (12–36 hours) windows.

Use Cases

• Developing early risk prediction models for major adverse events based on admission and reassessment features.
• Comparing machine learning algorithms like LASSO, random forest, and XGBoost for clinical prediction tasks.
• Validating risk stratification models in a de-identified timestamp-ordered validation cohort.
• Performing SHAP analysis to interpret the complementary contributions of different feature sets.

Strengths

• Model performance was assessed with multiple metrics including AUC, PR-AUC, calibration, Brier score, and decision curve analysis.
• The study includes an internal test cohort and a separate de-identified timestamp-ordered validation cohort.
• The dataset is derived from the authoritative PhysioNet Paediatric Intensive Care database.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The model's calibration was good in the internal test set but attenuated in the later validation cohort, suggesting a need for recalibration.

Provenance

Source

PhysioNet Paediatric Intensive Care database

Collection Method

Features were extracted from admission (0–6 hours) and early reassessment (12–36 hours) windows for 684 infection-related PICU admissions.

Freshness

Last updated 2026-06-04 05:44:14; freshness should be verified.