Modern machine learning techniques used in prediction of sepsis and bloodstream infection in burn patients: A systematic review.

INTRODUCTION

Burn patients are a group highly prone to sepsis and bloodstream infections (BSIs) due to immune dysregulation, skin barrier loss, and complex inflammatory responses. Traditional diagnostic criteria for sepsis are often unreliable in this population. While conventional regression models and laboratory biomarkers have been widely studied, they have limitations. Machine learning (ML) techniques have been used for this purpose. However, no systematic review has yet summarized these applications specifically for burn care. This review aims to synthesize and critically appraise current evidence on ML models for predicting sepsis and BSI in burn patients.

METHODS

We systematically searched three databases: PubMed, Scopus, and Web of Science. Eligible studies included original research applying modern ML algorithms to predict sepsis or BSI in burn patients. Two independent reviewers screened and extracted data. Risk of bias was assessed using the PROBAST+AI tool.

RESULTS

Following the search and screening process, seven studies were selected for final inclusion, comprising four studies focused on developing predictive models for sepsis and three aimed at predicting BSI in burn patients. One out of seven studies used multi-center data. Included studies had sample sizes ranging from 82 to 302. The data types were heterogeneous. The majority of the studies used clinical and laboratory data. Notably, only one study performed temporal validation beyond internal resampling.

CONCLUSION

Despite the promising advancements identified in this review, the field of using AI techniques to predict sepsis and bloodstream infections in burn population remains underexplored. The limited number of studies, coupled with small sample sizes, indicates that this field is still an emerging area of research.