Machine learning prediction of mortality in pediatric fungemia using the Candida score

Area of Science:

• Critical Care Medicine
• Infectious Diseases
• Pediatrics

Background:

• Pediatric fungemia, a bloodstream fungal infection in children, is associated with high mortality rates, especially in pediatric intensive care units (PICUs).
• Accurate prediction of mortality risk is crucial for timely and effective clinical management of pediatric fungemia.

Purpose of the Study:

• To evaluate the predictive performance of the Candida Score combined with clinical variables for mortality in pediatric fungemia.
• To compare the discrimination of a multivariable logistic regression model against machine learning algorithms (Random Forest, Gradient Boosting Machine).

Main Methods:

• A retrospective analysis of 85 pediatric fungemia cases from a single PICU (2016-2020).
• A prespecified multivariable logistic regression model was used as the primary analysis.
• Random Forest and Gradient Boosting Machine were employed as exploratory comparative models.
• Model discrimination was assessed using a held-out test set and validated through 10-fold cross-validation and bootstrap resampling.

Main Results:

• The study included 85 pediatric patients with a median age of 6 months; 45.9% of patients died.
• On the test set, logistic regression achieved an AUC of 0.800.
• Random Forest demonstrated the highest discrimination with an AUC of 0.861, followed by Gradient Boosting (AUC 0.847).

Conclusions:

• Integrating the Candida Score with clinical predictors shows potential for stratifying mortality risk in pediatric fungemia.
• Machine learning models, especially Random Forest, may offer superior discrimination compared to traditional logistic regression in this context.
• These findings are exploratory and necessitate external validation in larger, multicenter cohorts before clinical implementation.