An automated system to differentiate Kawasaki disease from febrile illness with real life clinical datasets in New York City

ABSTRACT – Kawasaki disease (KD) is the most common cause of acquired heart disease in
children. Treatment with intravenous immunoglobulin (IVIG) reduces the incidence of coronary
aneurysms and risk of long-term cardiovascular complications. IVIG is recommended to be
given within 10 days of illness; however only 4.7% receive the correct diagnosis at the first
medical visit. Timely and accurately diagnosis of KD is critical, yet there isn’t a gold standard
diagnostic test. A challenge of diagnosis is that the clinical signs of KD overlap those of other
pediatric febrile illnesses. We previously applied statistical learning using clinical and laboratory
test variables to differentiate KD from febrile illnesses and validated the algorithm in five
children’s hospitals in the US. Results showed its potential of being a computer-assist tool of
decision making at point of care in the settings where echocardiography would not be readily
available. Before translation and commercialization, the algorithm needs to be validated in a
large, diverse population and integrated into a patient surveillance platform as a real-time
screening tool for healthcare providers to use. In this project, we propose three specific aims to
address the central hypothesis that a KD screening tool incorporating our previously identified
and newly found patient-level variables in the electronic health record (EHR) can differentiate
KD from clinically similar febrile illnesses in an ethnically diverse pediatric population in New
York City (NYC). We will collaborate with Healthix, the nation’s largest public health information
exchange (HIE) with data of over 16 million patients from NYC. In Aim 1, we will set up a
pediatric EHR warehouse of patients with KD and other febrile illnesses from Healthix NYC data
sources. In Aim 2, we will identify features that are differentially expressed between patients
with KD and patients with other febrile illnesses, and develop an improved algorithm to
differentiate KD from other febrile illnesses. Finally, we will integrate the algorithm into the HBI
Spotlight Solutions. The Spotlight Solutions include a healthcare surveillance platform with high-
capacity data infrastructure and risk engines to offer AI solutions to providers. We expect
ultimately an HIE-based pediatric KD assessment system will be ready to alert HIE participating
providers for timely evaluation, treatment and follow up for the long-term cardiovascular
sequelae in NYC and other communities.Accomplishing the proposed project shall contribute to addressing the problem of
delayed diagnosis of Kawasaki disease (KD). This rapid, low-cost KD screening
system is expected to assist clinical decision-making and enable timely evaluation and
actionable treatment for patients’ better health outcomes in medical facilities. The results
will also add to the knowledge of clinical patterns associated with patients of different
racial and ethnic groups.