Clinical phenotyping of bloodstream infections: a review of current evidence

Bloodstream infections (BSIs) are a leading cause of morbidity and mortality, yet their clinical heterogeneity continues to challenge effective patient stratification and treatment optimisation. In other heterogeneous conditions such as sepsis, data-driven clinical subphenotyping has identified reproducible subgroups with distinct outcomes and treatment responses. Whether similar approaches can be applied to BSIs to improve clinical management and trial design is an area of growing interest.

We aimed to review the current evidence on the use of data-driven phenotyping and unsupervised machine learning techniques to identify clinical subphenotypes in BSIs. Methodological approaches used to derive subphenotypes, and their implications for clinical practice and trial design are elucidated.

A literature search was performed using PubMed/MEDLINE and the Semantic Scholar AI-assisted search tool. Reference lists of included studies were hand-searched to identify additional publications.

Data-driven subphenotyping has been most extensively studied in Staphylococcus aureus bacteraemia (SAB), where latent class analysis and cluster analysis point towards distinct subphenotypes with significantly different mortality rates across independent international cohorts. Emerging evidence extends to mixed-pathogen BSI cohorts in the intensive care unit and in immunocompromised populations including solid organ transplant recipients. Bedside tools including online calculators, and simplified scoring systems have been developed to facilitate rapid phenotype assignment. However different studies use varying methodological approaches and there is limited data on validation of these.

Clinical subphenotyping offers a promising framework for personalising antimicrobial therapy, improving risk stratification, and enriching clinical trial populations. Future key priorities include extending phenotyping to underrepresented BSI aetiologies, integrating clinical phenotypes with biological endotypes, and conducting phenotype-stratified interventional trials.