Refining empiric subgroups of pediatric sepsis using machine-learning techniques on observational data

Abstract

Sepsis contributes to 1 of every 5 deaths globally with 3 million per year occurring in children. To improve clinical outcomes in pediatric sepsis, it is critical to avoid "one-size-fits-all" approaches and to employ a precision medicine approach. To advance a precision medicine approach to pediatric sepsis treatments, this review provides a summary of two phenotyping strategies, empiric and machine-learning-based phenotyping based on multifaceted data underlying the complex pediatric sepsis pathobiology. Although empiric and machine-learning-based phenotypes help clinicians accelerate the diagnosis and treatments, neither empiric nor machine-learning-based phenotypes fully encapsulate all aspects of pediatric sepsis heterogeneity. To facilitate accurate delineations of pediatric sepsis phenotypes for precision medicine approach, methodological steps and challenges are further highlighted.

Keywords: biomarker; clustering; machine-learning; observational data; pediatric sepsis.