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

Area of Science:

• Genomics and Bioinformatics
• Pediatric Medicine
• Computational Biology

Background:

• Pediatric sepsis is a critical global health issue, contributing significantly to child mortality worldwide.
• Current diagnostic methods for pediatric sepsis lack user-friendliness and optimal accuracy, necessitating improved tools.
• Early and accurate diagnosis is crucial for effective sepsis management in children.

Purpose of the Study:

• To develop a precise and accessible diagnostic model for pediatric sepsis using statistical and machine learning techniques.
• To identify a robust set of gene expression markers indicative of pediatric sepsis.
• To create a clinically applicable tool for the early detection of pediatric sepsis.

Main Methods:

• Utilized gene expression profiles from 364 pediatric samples in a training cohort.
• Employed Least Absolute Shrinkage and Selection Operator (LASSO) analysis to identify 18 diagnostic genes.
• Developed the PEDSEPS-GBM model using a Gradient Boosting algorithm, validated across multiple cohorts, and created a web calculator.

Main Results:

• Identified 18 key genes as diagnostic markers for pediatric sepsis.
• The PEDSEPS-GBM model demonstrated optimal diagnostic performance in validation and independent cohorts.
• A user-friendly web calculator was developed for the PEDSEPS-GBM model, facilitating clinical application.

Conclusions:

• The developed PEDSEPS-GBM model offers a highly promising tool for the accurate and early detection of pediatric sepsis.
• The study highlights the biological and clinical relevance of the identified gene markers and the diagnostic model.
• The accessible web calculator can aid healthcare professionals in diagnosing pediatric sepsis more effectively.