PICU Infection Patient Risk Stratification Model with Admission and Reassessment Features

684 infection-related pediatric intensive care unit admissions from the PhysioNet Paediatric Intensive Care database form the analytic cohort. The dataset was used to develop and validate a risk prediction model for 72-hour major adverse events, authored by Yuhang Wang and last updated in June 2026. It compares LASSO, random forest, XGBoost, and a stacked ensemble model.

Use Cases

• Training risk prediction models based on admission (M0) and reassessment (M1) feature windows.
• Benchmarking machine learning algorithms like random forest and XGBoost for clinical time-series data.
• Validating model calibration and discrimination using metrics like AUC, PR-AUC, and Brier score.
• Conducting feature importance analysis (e.g., SHAP) to understand contributions of parallel assessment frameworks.

Strengths

• The analytic cohort includes 684 infection-related PICU admissions with a defined primary outcome.
• Model performance was evaluated in both an internal test set and a de-identified timestamp-ordered validation cohort.
• The random forest model achieved an AUC of 0.724 and a PR-AUC of 0.741 in the internal test set.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying data is unknown, which may limit suitability assessment.
• The dataset is very small at 17.9 KB, indicating limited scope, likely containing only model results or a summary.

Provenance

Source

PhysioNet Paediatric Intensive Care database

Collection Method

Retrospective analysis of admission and reassessment features extracted from defined time windows (M0: 0–6h, M1: 12–36h).

Freshness

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