Diagnosis of Kawasaki Disease (KD) can be confusing in the absence of a confirmatory test or pathognomonic finding, especially when clinical criteria are incomplete (iKD). Delays in treatment are associated with an increased risk of coronary artery (CA) complications. We have lately proposed serum NT-proBNP as an adjunctive diagnostic test. We retrospectively tested a new algorithm to aid in diagnosis based on NT-proBNP (Z-score for age), coronary artery dilation (CAD) at onset, and abnormal serum albumin or C-reactive protein (CRP). The aim of the study was to assess the performance of the algorithm with respect to CAD outcome (aneurysm, dilation, or occult dilation) and to compare its performance to the 2004 American Heart Association (AHA) algorithm. The algorithm was tested on 81 KD patients who had NT-proBNP on admission at our institution between 2007 and 2013. Age at diagnosis was 3.2 ± 2.6 years, with a median of 5 diagnostic criteria (range 3-6), of whom 31/81 (38.3%) had iKD. Aneurysms occurred in 16/81 (19.8%); higher prevalence in iKD, 12/31 (38.7%) versus 4/50 (8.0%) (p=0.001). CAD affected 35/81 (43.2%), and 30/81 (37.0%) had occult CAD. Using this algorithm, 80/81 (98.8%) were to be treated, based on high NT-proBNP alone for 54/81 (66.7%), onset CAD for 12/81 (14.8%), and high CRP or low albumin for 14/81 (17.3%). Results were similar when the algorithm was applied to patients with complete or incomplete criteria. The only patient “not-to-treat” with our algorithm had iKD and occult CAD. Applying the AHA algorithm to our patient cohort, 12/30 (40.0%) of eligible patients would not have been referred for treatment, including 3 (25.0%) with coronary artery aneurysm, 6 (50.0%) with CAD and 3 (25.0%) with occult CAD. This NT-proBNP based algorithm proved superior to the 2004 AHA algorithm to identify and refer for treatment patients at risk of CA involvement, 30/31 (96.8%) versus 18/30 (60.0%), p<0.001. This NT-proBNP based algorithm is efficient to identify and treat patients at risk of coronary involvement, despite an apparent selection bias of CA involvement. This study paves the way for a prospective validation trial of the algorithm.View Large Image Figure ViewerDownload (PPT)