Can 8 lab markers predict severe pediatric flu?

Blood glucose was the strongest predictor in a machine learning model with an AUROC of 0.88, although sensitivity reached only 50%.

1

The retrospective study included 436 children hospitalized with influenza at one Chinese hospital from January 2022 to May 2023; 114 (26%) had severe disease.
2

A random forest model using 8 laboratory markers had the highest testing-cohort AUROC among 6 algorithms, at 0.88.
3

The random forest model had 84% accuracy, 50% sensitivity, 97% specificity, and positive and negative predictive values of 84% each.
4

Blood glucose was the most influential predictor; higher glucose, aspartate aminotransferase, and procalcitonin contributed to predictions of severe influenza.
5

The study was limited by its retrospective, single-center design, small testing cohort, lack of external validation, and inability to establish prospective prediction or improved outcomes.

1