Improving Characterization of Individualized Treatment Effects in Critical Illness

PROJECT ABSTRACT Sepsis and acute respiratory distress syndrome (ARDS) are common and potentially fatal critical care disorders in need of improved therapeutic approaches. Developing effective treatment strategies for sepsis and for ARDS has been challenging in part due to disease heterogeneity with variability in patient demographics, comorbid conditions, severity of illness, inciting insults, and host response. Understanding why patients respond differently to therapies in the intensive care unit is essential. This variability is commonly referred to as heterogeneity of treatment effect (HTE). Recent studies identified subphenotypes in sepsis and in ARDS based on clinical variables or biologic signatures and tested for HTE across subgroups. My recent research has focused on an alternate method of investigating HTE by estimating individualized treatment effects (ITEs) using machine learning-assisted approaches. Computational approaches that estimate ITEs model the difference in an outcome for an individual on treatment versus on placebo or usual care. ITEs predict whether a patient will benefit from or be harmed by a treatment and provide an estimate of the magnitude of benefit or harm. ITEs vary for patients based on their unique covariate patterns and may have advantages in uncovering treatment heterogeneity when the variables that determine treatment response differ from those contributing to subphenotype membership. Additional tools including SHapley Additive exPlanations provide information on variable importance for ITEs lending insight into potential mechanisms contributing to treatment response. Preliminary studies from our group using ITE-based approaches in datasets from previously completed randomized clinical trials have uncovered heterogeneous responses at an individual level even when no significant difference was observed in the overall trial. Specifically, our work has yielded promising signals for heterogeneity to (1) resuscitation strategies in sepsis using clinically available variables; (2) response to glucocorticoids in septic shock using cortisol levels; and (3) enteral nutrition strategies in ARDS using endocrine biomarkers. The current proposal will continue our work to improve characterization of treatment responses in sepsis and ARDS. In this R35, we will leverage existing cohorts and biospecimen repositories to provide biologic and mechanistic insights into the ITE signatures identified in our preliminary work. We noted in our preliminary studies that for some treatments clinically available variables measured immediately prior to intervention were not sufficient in predicting ITEs. Thus, in this R35, we will incorporate biologic data from multi-omics platforms to improve and refine ITE predictions. The MIRA award will be the ideal mechanism to fund this work as it facilitates the flexibility to rapidly adapt to new methodologies emerging in precision medicine and will allow for the investigation of multiple therapies for sepsis and for ARDS which may vary significantly in the variables determining treatment response.