An ensemble boosting model for predicting transfer to the pediatric intensive care unit

Area of Science:

• Clinical informatics
• Machine learning in healthcare
• Pediatric critical care medicine

Background:

• Early detection of patient deterioration is crucial for timely intervention.
• Predicting pediatric intensive care unit (PICU) transfers aids in resource allocation and patient management.
• Current methods may not fully capture the complexity of pediatric patient decline.

Purpose of the Study:

• To develop a data-driven pediatric early deterioration indicator.
• To predict the likelihood of patient transfer from a general ward to the PICU.
• To create a tool for clinicians to anticipate critical care needs.

Main Methods:

• Utilized 5.5 years of electronic health record data from two medical facilities.
• Developed machine learning classifiers including adaptive boosting and gradient tree boosting.
• Created an ensemble model and evaluated its generalizability across facilities, comparing it to a modified pediatric early warning score (PEWS).

Main Results:

• The machine learning ensemble model demonstrated superior performance over the modified PEWS baseline.
• Achieved higher accuracy (0.77 vs. 0.69), sensitivity (0.80 vs. 0.68), specificity (0.74 vs. 0.70), and AUROC (0.85 vs. 0.73).

Conclusions:

• Data-driven machine learning algorithms can significantly improve PICU transfer prediction.
• These algorithms outperform expertly defined systems like modified PEWS.
• Careful algorithm training is necessary to prevent bias in predictive outcomes.