Project Summary/Abstract
Anaphylaxis can be life-threatening but exacts an outsized toll on patients and the healthcare system. This
system’s burden is due in part to one size fits all national guidelines that necessitate emergency department
(ED) care for all patients and = 4-6-hour observational periods that are agnostic to reaction severity and patient
characteristics. This generic approach is ill-suited to most reactions, thereby contributing to ED overutilization,
long ED wait times, ED overcrowding, avoidable hospitalizations, and inflated healthcare costs. The proposed
K23 study seeks to deliver personalized anaphylaxis care and will further develop my skillset as an independent
translational investigator. I hypothesize that machine learning can be applied to my actively enrolling KL2 cohort
to develop prediction models capable of customizing ED observation periods in real-time based on patient-
specific risk. The rationale for this hypothesis comes from preliminary KL2 data (n = 91) in which 51.6% of
patients had complete symptom resolution within 2 hours from the first epinephrine dose. The next logical step
is to expand my prospective cohort and employ machine learning to develop a model capable of predicting
time-to-symptom resolution in real-time to reduce observation periods, prevent unnecessary hospitalizations,
and identify patients who may not require ED care and can instead be safely managed at home. Additionally,
there is an unmet need for clinically actionable biomarkers to enhance patient risk stratification and enable
customizable therapeutic approaches; however, there is an incomplete understanding of immunological
mechanisms contributing to human anaphylaxis pathogenesis. Most research has focused on the role of IgE-
mediated mast cell and basophil activation, yet emerging data support my second hypothesis that neutrophils
are key drivers of anaphylaxis that govern divergent progression into either refractory or biphasic courses. The
rationale for this hypothesis comes from preliminary KL2 data in which 100% of patients with severe reactions
exhibited elevated markers of neutrophil activation and levels were dynamic over time. The specific aims of
this application are 1) To derive a dynamic clinical prediction model to customize anaphylaxis
observation periods. Simultaneously, I will acquire training in biostatistical and biomedical informatics,
machine learning, dynamic prediction modeling, and mediation analysis; and 2) Demonstrate that serum
biomarkers of neutrophil activation predict reaction severity and clinical courses. This aim will help me
acquire essential training in translational approaches to neutrophil biology and will provide a mechanistic
gateway for future research on the immunology of anaphylaxis courses. Findings from this study will result in a
novel prediction model to optimize anaphylaxis observation periods and prevent unnecessary hospitalizations.
These outcomes will aid my transition to independent investigator status by conducting validation studies of the
predictive model in a multicenter cohort. Study findings will also open avenues of investigation on the
integration of neutrophil measures into clinical care and the prognostic machine learning tool.