ABSTRACT
Sepsis-associated multiple organ dysfunction syndrome (MODS) affects approximately 1 in every 6 children
admitted to a pediatric intensive care unit (PICU) in the U.S. and is associated with about 1 in every 3 PICU
deaths. Despite this, the current management approach for pediatric sepsis and sepsis-associated MODS and
the use of adjuvant therapies remains highly variable in the PICU. This is in part due to the heterogeneity of
sepsis and the high variability in which it presents and evolves in children. This heterogeneity can delay the
recognition of sepsis and the early identification of patients at high risk of poor outcomes, and has limited the
development of targeted interventions. Unraveling this heterogeneity and identifying high-risk sepsis
phenotypes with prognostic and therapeutic relevance is considered a major research priority in the path
towards precision medicine in sepsis. However, for precision medicine strategies to be most clinically impactful
for children with sepsis, it is critical that they are implementable in real-time, generalizable to more settings,
and cost-effective. This would enable timely enrollment in enriched randomized controlled trials and initiation of
targeted interventions with potential for higher public health impact. We previously identified and characterized
the persistent hypoxemia and shock phenotype in critically ill children with sepsis-associated MODS using a
data-driven approach. This phenotype was present in almost a third of children with sepsis-associated MODS,
had features of systemic inflammation and endothelial activation, was highly reproducible, and was
independently associated with mortality, accounting for more than two-thirds of all deaths in this population.
Furthermore, in a causal inference analyses, patients with persistent hypoxemia and shock were associated
with a higher likelihood to benefit from intravenous hydrocortisone and albumin when compared to other
patients. Importantly, phenotypes with similar characteristics in terms of hypoxemia, shock, inflammation, and
endothelial activation has been identified in prior studies of sepsis and acute respiratory distress syndrome in
children and adults. However, the persistent hypoxemia and shock phenotype currently requires longitudinal
data for up to 72 hours to be classified, so there is a need to determine whether it can be predicted early in the
course using routinely collected clinical data. Furthermore, amongst patients who achieve a high probability of
having the phenotype, there is a need to determine which common adjuvant therapies are associated with
heterogeneity of treatment effect (HTE) in order to inform future clinical trials. In this study we aim to: (1) Derive
and externally validate a prediction model of persistent hypoxemia and shock in children admitted to the PICU
with confirmed or suspected infection; (2) Determine which common adjuvant therapies and therapeutic goals
are associated with HTE amongst children with high probability of having persistent hypoxemia and shock; and
(3) Implement and prospectively evaluate the performance of the prediction model of persistent hypoxemia and
shock in the EHRs of three institutions.
