Sepsis mortality prediction using machine learning and deep learning - a systematic review

Sepsis, a critical infection-induced inflammatory condition, poses substantial global health challenges, demanding timely detection for effective intervention. This article explores the application of machine learning (ML) and deep learning (DL) in predicting sepsis using electronic health record (EHR) to enhance patient outcomes. A comprehensive search across PubMed, IEEE Xplore, Google Scholar, and Scopus yielded 39 studies meeting stringent inclusion criteria. Predominantly retrospective (n = 34) and geographically diverse, these studies, spanning North America (n=19), Asia (n=13), Europe (n=6), and Australia (n=1), exhibited varied datasets, sepsis definitions, and prevalence rates, necessitating data augmentation strategies. Heterogeneous parameter usage, diverse model distribution, and inconsistent quality assessments were identified. Despite differences, longitudinal data showcased the potential of early sepsis prediction. The review outlines the challenges posed by disparate funding and article quality correlation, emphasizing the need for standardized evaluation metrics. In conclusion, this systematic review highlights the promising role of ML/DL methodologies in sepsis detection and early prediction through EHR, underscoring the imperative for standardized approaches and comprehensive quality assessments.