Biomarker-enhanced Artificial Intelligence-Based Pediatric Sepsis Screening Tool Towards Early Recognition and Personalized Therapeutics

PROJECT SUMMARY
The overall objective of this proposed STTR effort is focused on the derivation and validation of a commercialized
biomarker-enhanced artificial intelligence (AI)-based pediatric sepsis screening tool (PSCT) that can be
incorporated into emergency department (ED) workflows to enhance early recognition and the initiation of timely,
aggressive personalized sepsis therapy.
The early recognition and timely personalized management of sepsis remain among the greatest challenges in
pediatric emergency medicine. The early ED recognition of established or impending critical sepsis is hampered
by high prevalence of common febrile infections, poor specificity of discriminating features, capacity of children
to compensate until advanced stages of shock, and delays/limited sensitivity of infection confirming
microbiological tests. In a recent study, 47% of 7 million cases of sepsis admitted to ICUs had negative cultures.
Improved diagnostics are needed to distinguish between sterile inflammation, viral infection, and bacterial
infection in patients with suspected sepsis. While useful, commonly used laboratory-based diagnostics such as
WBC, CRP, PCT and lactate of limited utility in the management of pediatric sepsis. Novel panels of biomarkers
for the Pediatric Sepsis Biomarker Risk Model (PERSEVERE) have shown to be effective in prediction of
deterioration and mortality in immunocompromised patients. The performance of PERSEVERE biomarkers in a
more undifferentiated population of children with possible sepsis, where the aim is to identify those that are about
to deteriorate remains unknown.
Septic patients, especially when critically ill, represent a highly heterogenous population. The role of the host-
specific dysregulated immune response in the pathophysiology of sepsis, coupled with the diversity of
phenotypes, highlights the need for a precision medicine PSCT approach that identifies patients who are most
likely to benefit from targeted interventions such as restrictive fluid resuscitation where early vasoactive therapy
is initiated rather than repeated fluid boluses.
Automated sepsis screening tools in the market today are generally brittle, embedded modules in a large EHR
system that exhibit poor specificity and positive predictive value, ignore important evidence available in free text
notes, and do not reflect decision-making criteria used by expert ED physicians in initiating sepsis care. We
believe there is a significant need for a continuously learning commercial PSCT that leverages widely available
EHR interface standards to deliver the combined analytic power of expert knowledge, biomarkers, NLP and
machine learning to enhance early pediatric sepsis recognition and detect phenotypes that can predict treatment
responses/outcomes towards personalized medicine.
.PROJECT NARRATIVE
In hospitals today the subtle symptoms of sepsis that can masquerade as many less dangerous childhood
illnesses frequently result in delayed recognition. Moreover, despite the high diversity of pediatric sepsis, the
treatment delivered in cases of severe sepsis is generally “homogenized” and may not optimally meet the unique
needs of a specific patient. Our goal is to develop an artificial intelligence (AI)-based pediatric sepsis screening
tool that can be routinely used in emergency departments to improve both early recognition and personalized
treatment, known to be associated with reduced mortality in infected children.