Revising Host Phenotypes of Sepsis Using Microbiology

Area of Science:

• Critical Care Medicine
• Infectious Diseases
• Bioinformatics and Computational Biology

Background:

• Sepsis exhibits significant heterogeneity in pathogens, host responses, organ dysfunction, and outcomes.
• Existing clinical and biological sepsis phenotypes lack the integration of crucial pathogen data.
• The impact of microbiological variables on sepsis classification remains largely unexplored.

Purpose of the Study:

• To investigate the role of pathogen data in classifying sepsis phenotypes.
• To compare sepsis phenotypes derived from clinical data alone versus combined clinical and microbiological data.
• To analyze the impact of phenotype reclassification on treatment effects, specifically for recombinant human activated protein C (rhAPC).

Main Methods:

• Secondary analysis of the PROWESS study data involving 1,690 severe sepsis patients.
• Latent class analysis (LCA) was employed to identify distinct sepsis phenotypes.
• Two models were developed: a 'host model' using only clinical variables and a 'host-pathogen model' incorporating clinical and microbiological data (infection site, pathogen type, antimicrobial resistance).

Main Results:

• Latent class modeling identified 4 distinct phenotypes in both the host and host-pathogen models.
• Incorporating microbiological data led to a phenotype reassignment for 46% of patients, increasing model probability (0.95 to 0.97).
• Phenotype characteristics, particularly for 'beta' (abdominal infections) and 'gamma' (pulmonary infections) types, were significantly refined by pathogen data, influencing coagulation marker levels and 28-day mortality differences.

Conclusions:

• Sepsis phenotype classification is significantly modified by the inclusion of microbiological data alongside clinical variables.
• The host-pathogen model enhances cluster cohesiveness and homogeneity, providing a more accurate representation of sepsis heterogeneity.
• Refined phenotypes alter the observed treatment effects of rhAPC, highlighting the importance of pathogen-specific classification for personalized medicine.