1515. Distinguishing Kawasaki Disease Initially Presenting as Cervical Lymphadenopathy from Bacterial Cervical Lymphadenitis Using the Decision Tree Analysis

BACKGROUND: Kawasaki disease (KD) can present initially with fever and cervical lymphadenopathy before other characteristic clinical signs appear. It is important to recognize KD at an early stage for appropriate treatment to be started and reduce cardiac complications. We sought to identify variables which distinguish early stage of node first presentation of KD from bacterial cervical lymphadenitis (BCL). METHODS: We retrospectively reviewed records of patients presenting with cervical lymphadenopathy and fever as primary diagnosis who discharged from the inpatient ward between January 2010 and December 2017. Only the patients who underwent radiologic studies (CT or US) were included. Patients were divided into 3 groups by their final diagnosis; CL with abscess (Group A), CL without abscess (Group B), and Node first presentation of KD (Group C). Initial clinical, laboratory, and imaging characteristics of the 3 groups were compared. Sequential laboratory data performed on the third hospital day was used to calculate their percentage differences. A clinical prediction model for KD in patients with CL was constructed using the decision tree analysis. RESULTS: Patients with KD had a longer duration of fever after admission and longer duration of fever prior to final diagnosis compared with BCL patients. Compared with group A or B, group C patients had higher neutrophil counts, neutrophil-lymphocyte ratio, C-reactive protein (CRP), erythrocyte sedimentation rates (ESR), alanine transaminase (ALT), aspartate transaminase (AST). In radiologic studies, no differences were observed in node sizes or palatine tonsil enlargement among the 3 groups. Retropharyngeal edema was more commonly associated with group C than group A and B combined, but was not significantly different between group A and C. By decision tree analysis, factors contributing to classification included presence of abscess, CRP percentage change, and neutrophil percentage change (Figure 1). Strong predictivity of the decision tree was observed in classifying Group C, with an overall accuracy of 87.9% for Group A, 82.6% for Group B, and 91.4% for Group C. CONCLUSION: Decision tree analysis showed that the presence of abscess, CRP percentage change, and neutrophil percentage change could classify KD from BCL with an accuracy of 91.4%. [Image: see text] DISCLOSURES: All authors: No reported disclosures.