Abstract O.32: An Improved Point-of-care Differentiation Of Kawasaki Disease From Other Febrile Illnesses

Abstract

Background: Kawasaki disease (KD) shares many clinical features with other pediatric febrile illnesses (FC). Clinical confusion can lead to a missed or delayed diagnosis that increases the risk of coronary artery damage. In the present study, we improved our previous KD diagnostic algorithm for point-of-care diagnosis. Methods and Results: We reviewed clinical records of 534 acute KD and 318 FC patients (development dataset) and subsequent 268 acute KD and 161 FC patients (validation dataset). KD subjects met the American Heart Association definition. Using clinical data and lab test results, we integrated our previously developed linear discriminant analysis (LDA)-based clinical model with a newly developed decision tree-based algorithm to improve KD diagnosis. To train the decision trees, sub cohorts were constructed based upon the 5 KD classic criteria. Our 1 st clinical model (LDA) stratified the subjects into FC (FC diagnosis, negative predictive value NPV >=95%), undecided (88/802 KD, 167/479 FC), and KD (KD diagnosis, positive predictive value PPV >=95%) subgroups. The subsequent 2 nd clinical model (decision trees) further classified the undecided group into FC, undecided, and KD subgroups, resulting in a much-improved algorithm with only 59/479 FC (Specificity>76%) and 26/802 KD (Sensitivity>95%) undetermined. Conclusions: Our computer-based algorithm that incorporates only clinical findings and readily available clinical laboratory data now has sufficient sensitivity and specificity in distinguishing acute KD from FC patients that a multicenter, prospective clinical trial is warranted to test the performance of the diagnostic algorithm against the gold standard of expert clinical opinion.