Sepsis is a common cause of pediatric multiple organ dysfunction syndrome (MODS), which, in turn, is a final common pathway for death in most critically ill children. Unfortunately, approaching sepsis as a uniform clinical syndrome has contributed to dozens of failed drug trials and hundreds of single biomarkers demonstrating poor performance. In light of this, some sepsis researchers have focused their efforts in discovering sepsis subtypes that represent the specific pathobiological pattern underlying the syndrome in different subgroups of patients. These subtypes, in theory, are associated with therapy response and outcomes and could be used to guide individualized therapy. While optimal methods to uncover sepsis subtypes are still unclear, there are strong advantages to using a phenotype-based approach, where the clinical characteristics of patients are used as surrogates of the underlying pathobiology. We hypothesize that the clinical characteristics of children with sepsis can be adequately quantified using calibrated pediatric organ dysfunction scores and that the early patterns of organ dysfunction can be used to identify novel phenotypes of sepsis with prognostic and therapeutic relevance. Existing pediatric organ dysfunction scores are weighted in a multivariable model to predict mortality and are not designed to represent the progressive loss of function of each individual organ system. In light of this, we recently adapted and validated a pediatric version of the Sequential Organ Failure Assessment score (pSOFA) which allocates similar weight to each failing organ system. However, the validation of pSOFA was performed in a single center and no attempts were made to calibrate the organ-specific subscores to optimize the grading of each organ dysfunction (i.e. respiratory, cardiovascular, hepatic, renal, neurologic, and hematologic). In this study, we aim to: 1) re-calibrate and validate the pSOFA subscores using a multi-center observational cohort of critically ill children with confirmed or suspected infection; and (2) analyze the early patterns of organ dysfunction in critically ill children with sepsis using the pSOFA subscores in order to identify novel phenotypes of sepsis with prognostic and therapeutic relevance. We will use Subgraph Augmented Non-Negative Matrix Factorization to model and visualize the patterns of organ dysfunction during the acute phase of sepsis. These patterns will form the basis for the characterization of novel sepsis phenotypes. We will then evaluate whether the novel sepsis phenotypes are associated with outcomes in a validation cohort and independently associated with response to two common adjuvant treatments: corticosteroids and albumin infusions. The results from this study will be used as the foundation for follow-up studies to characterize the underlying molecular and cellular perturbations associated with the identified phenotypes.
Sepsis is a life-threatening infection that affects many children and can frequently result in the failure of multiple organs, such as the lungs, heart, or kidneys. There are no existing effective treatments that are specifically designed for sepsis, but there is evidence that some subgroups of patients with sepsis respond differently to some treatments and likely share some distinguishing features at the genetic and cellular level. The best way to identify these subgroups of patients still remains unclear, so we are proposing an approach that involves applying advanced data analysis techniques to the clinical data contained in the electronic medical records of thousands of children with sepsis, particularly the data that describes how their organs are functioning.
