Abstract
Sepsis induces intense, dynamic and heterogeneous host response modulations. Despite improvement of patient management, the risk of mortality and healthcare-associated infections remains high. Treatments to counterbalance immune response are under evaluation, but effective biomarkers are still lacking to perform patient stratification. The design of the present study was defined to alleviate the limitations of existing literature: we selected patients who survived the initial hyperinflammatory response and are still hospitalized at day 5-7 after ICU admission. Using the Immune Profiling Panel (IPP), a fully automated RT-qPCR multiplex prototype, we optimized a machine learning model combining the IPP gene expression levels for the identification of patients at high risk of worsening, a composite endpoint defined as death or secondary infection, within one week after sampling. This was done on 332 sepsis patients selected from two retrospective studies. The IPP model identified a high-risk group comprising 30% of patients, with a significant increased proportion of worsening events at day 28 compared to the low-risk group (49% vs. 28%, respectively). These preliminary results underline the potential clinical application of IPP for sepsis patient stratification in a personalized medicine perspective, that will be confirmed in a larger prospective multicenter study.