Combined Biomarker and EMR Data for Heterogeneous Treatment Effects and Surrogate Endpoints in Sepsis

Principal Investigator/Program Director (Last, first, middle): Reddy, Jr., Bobby Project Summary: Sepsis is a poorly understood clinical syndrome characterized by dysregulated host response to infection. The complexity and heterogeneity of the host response has frustrated attempts at developing effective treatments. In partnership with 6 U.S. hospitals, Prenosis amassed NOSIS, one of the world’s largest datasets and biobanks that combines biomarker and clinical data for patients suspected of infection, housing over 60,000 plasma or serum samples from over 12,000 patients. We also curated a dataset of dense time-series data from each patient’s Electronic Medical Record (EMR), including demographics, vitals, lab results, interventions, outcomes, and many other parameters. In this project, Prenosis will build upon previous work to conduct targeted analyses of the individual treatment effects of corticosteroids on septic patient outcomes. Using propensity score matching on patient baseline data (their biomarker/EMR profile at the time they were suspected of serious infection), we have identified 1,350 pairs of treatment and control patients whose samples have already been collected and stored in the biobank. We will leverage our existing pipeline to measure the 40 core biomarkers on 8,100 samples for these 2,700 patients: the pre-treatment sample closest to the time at which the treatment patient received steroids, the sample closest to 24 hours after the treatment patient received steroids, and the patients’ final samples to understand how treatment impacted biomarker profile and how these changes are associated with clinical outcomes of interest. These data and analyses will serve as the basis for at least three immediately commercializable products. First, elucidating which patients are likely to benefit or suffer from treatment with corticosteroids will improve diagnostic and clinical decision support products that are deployable within Prenosis’ existing ImmunixTM platform. Second, this will demonstrate the NOSIS dataset is a powerful platform for predictive enrichment of clinical trials. Estimating individual treatment effects in NOSIS identifies the optimal patient subpopulations to be recruited for clinical trials, which would otherwise fail. Third, understanding treatment effects on biomarker profile and corresponding association with clinical outcomes will establish the NOSIS host response profile as effective surrogate endpoints for clinical trials.