PROJECT SUMMARY Sepsis is a disorder that contributes to approximately 1 in 3 hospital deaths in the United States and 1 in 5 deaths worldwide. Advancing sepsis management has been challenging in part due to the heterogeneity of septic patients in demographics, comorbidities, infectious source, microbiologic etiology, and level of organ dysfunction. Our group has previously identified sepsis subclasses that differ in prognosis and response to treatment, suggesting that precision medicine may improve sepsis care. Glucocorticoids (GCs) are commonly used in septic patients despite inconsistent results from randomized controlled trials (RCTs), and they are an ideal candidate to develop a precision medicine approach. Recently, two of the largest RCTs ever conducted to test efficacy of GCs in sepsis (APROCCHSS and ADRENAL) also demonstrated conflicting results. Reconciling discordant results between trials has proved challenging with traditional methods but may be facilitated by state-of-the-art computational approaches which incorporate machine learning to estimate the conditional average treatment effect based on individual covariate patterns. In this proposal, we will create a ‘knowledge network’ using clinical and biologic data from 4 RCTs of GCs in sepsis (APROCCHSS, ADRENAL, ESCAPe, HYPRESS) and electronic health record data (Sepsis Endotyping in Emergency Care project). In Aim 1, we will utilize unsupervised and supervised learning approaches using clinical data from RCTs to characterize heterogeneity of treatment effect, identify subclasses that benefit, and develop a treatment policy to reduce 90-day mortality. In Aim 2, we will use causal Bayesian modeling approaches that incorporate RCT and EHR data to identify effect modifiers and confounders of GC therapy and mortality. We will use these results to develop a treatment policy to reduce 90-day mortality. Secondary analyses will compare RCT-only policies in Aim 1 to RCT-EHR policies in Aim 2. In Aim 3, we will perform cytokine assays and RNAseq using samples from the ADRENAL, ESCAPe, and HYPRESS trials to identify endotypes that benefit from GCs. We have assembled a multidisciplinary team of clinical trialists, biostatisticians, computational biologists, and critical care specialists with an established track record of collaboration for this proposal. Successful completion of our Aims will reconcile discordant results of prior RCTs testing GCs in sepsis, develop a treatment policy that can be deployed in EHRs, and improve design of future RCTs.