Abstract
Worldwide, nearly 3 million children die each year from sepsis. Red blood cells (RBC) are transfused in ~50%
of children with septic shock, with the intent to enhance oxygen delivery, help resolve shock, and prevent
organ dysfunction. However, RBC transfusion has repeatedly been associated with adverse outcomes in
critical illness, suggesting harm. For most children with septic shock we lack data to identify who should or
should not be transfused. Interventional trials and RBC transfusion guidelines have used hemoglobin
concentration alone to inform RBC transfusion decisions. However, this is likely not the best way to decide
when to transfuse RBCs to a patient with shock. Our preliminary data in septic children suggest that
hemodynamic measures better identify children who might benefit from RBC transfusion and that children with
severe immune suppression may be at greater risks of transfusion-related harm. Lastly, our data show that
RBC transfusion effects likely differ based on RBC unit collection, storage, and processing methods, and
resultant quantities of soluble mediators. Together, these data support our hypothesis that physiologic and
blood product factors will predict when children with septic shock will benefit from, or be harmed by, RBC
transfusion. The overall goal of our research program is to develop a personalized-medicine approach to
RBC transfusion in children with septic shock. Our objective for this proposal is to build decision support tools
that can be used in future clinical trials to identify when children with septic shock should or should not be
transfused with RBCs. Our proposal uses robust machine learning techniques that are designed to use a very
large number of dynamic variables to define personalized treatment decisions that maximize clinical outcomes.
With these methods, and an agnostic, data-driven approach, the resultant decision rules will be free of clinician
bias and can consider a much wider array of clinical parameters than have traditionally been used in
transfusion trials. We will conduct a multicenter prospective observational study of 660 children with septic
shock to accomplish the following specific aims: Aim 1: Determine which clinical, hemodynamic, and blood
product-specific factors should drive RBC transfusion decision-making in children with septic shock.
Using data from the electronic medical record, we will use outcome-weighted learning to build dynamic
transfusion algorithms to minimize predicted daily organ dysfunction (PELOD2 scores). Aim 2: Identify
immune phenotypes that predict differential response to RBC transfusion in children with septic
shock. We will collect serial blood samples from subjects to measure a validated panel of inflammatory and
immune function biomarkers and test the hypothesis that outcomes related to RBC transfusion will differ based
on pre-transfusion levels of biomarkers of inflammation and/or immune suppression. We expect our data to
transform the design of interventional trials of data-driven RBC transfusion algorithms to shift the transfusion
paradigm away from hemoglobin-based strategies and improve outcomes for critically ill septic children.