, et al
Point-of-care differentiation of Kawasaki disease from other febrile illnesses
J Pediatr
; doi:

Investigators from Stanford University and the University of California at San Diego studied the pattern of clinical findings and laboratory test results in prospective cohorts of febrile children to develop and validate an algorithm to differentiate Kawasaki disease (KD) from other febrile illnesses. They postulated that such an algorithm could be incorporated into information technologies as an inexpensive point-of-care tool to assist clinicians in diagnosing KD. Four cohorts of children were randomized: children thought to have KD for the training phase (development of the algorithm) (n=276); children with fever not due to KD (febrile controls [FC]) for the training phase (n=243); children thought to have KD for the testing phase (to validate the algorithm internally) (n=136); and FC for the testing phase (n=121). The KD cohorts included children diagnosed within 10 days of fever onset using the American Heart Association guidelines (fever for ≥3 days and the presence of at least 4 of the following: rash, conjunctival injection, oral mucosal changes, extremity changes, or cervical lymphadenopathy [>1.5 cm], or fewer than 4 of these findings with documented coronary artery abnormalities). The FC cohorts included children seen in the emergency department with ≥3 days of fever and ≥1 finding of KD. All FC had a clinically-determined or culture-proven etiology (viral infection for 256, bacterial infection for 76, other infammatory process for 29, and combined viral and bacterial infections for 3). The cohorts were demographically similar, except that the KD cohorts were younger.

Clinical findings (presence of fever within 24 hours of presentation, the 5 criteria for KD, and number of days since fever onset) and laboratory data (complete blood count, C-reactive protein, gamma-glutamyl transferase, alanine aminotransferase, and erythrocyte sedimentation rate) were recorded. These clinical findings and laboratory test results from the training cohorts were analyzed by logistic regression to ascertain adjusted odds ratios and likelihood P values. Using linear discriminant analysis, subjects were classified based on clinical findings, laboratory results, or both. Receiver operating characteristic analyses allowed generation of a KD score (high, low, or intermediate), stratifying subjects as >95% positive predictive value of KD, >95% negative predictive value of KD (thus, FC), or indistinguishable using the algorithm.

Clinical variables that were significantly different between the KD and FC cohorts included conjunctival injection, extremity changes, oropharyngeal changes, and fever. Laboratory values that differed significantly between the cohorts included age-adjusted hemoglobin, platelet count, and erythrocyte sedimentation rate. Optimal performance of the algorithm was noted when clinical findings and laboratory test results were combined. Using the combined results, the algorithm correctly diagnosed (produced a high score for) 81.2% of KD patients during training and 74.3% of KD patients during the testing phase. The algorithm correctly diagnosed (produced a low score for) 67.5% of FC patients during training and 62.8% of FC patients during...

You do not currently have access to this content.