Diagnostic potential of a gradient boosting-based model for detecting pediatric sepsis.

Abstract Pediatric sepsis is a major cause of mortality of children worldwide. However, there is still a lack of easy-to-use predictive tools that can accurately diagnose sepsis in children. This study aimed to develop an optimal gene model for the diagnosis of pediatric sepsis using statistics and machine learning approaches. Combining gene expression profiles from a training cohort of 364 pediatric samples with a Least Absolute Shrinkage and Selection Operator analysis produced eighteen genes as diagnostic markers. With the implementation of a Gradient Boosting algorithm, a model designated PEDSEPS-GBM, that aggregated these markers was developed with optimal performance for the diagnosis of pediatric samples in the validation and two independent cohorts. Moreover, a web calculator with a user-friendly interface was established for PEDSEPS-GBM. This study presents a diagnostic model that holds great potential for the detection of pediatric sepsis, and demonstrates the biologic and clinical relevance of this model.