Kawasaki syndrome is an acute vasculitis of infancy and childhood which results in coronary artery lesions in 15 to 25% of affected children. Treatment with intravenous gamma globulin and aspirin with the first 10 days of illness has been demonstrated to reduce the prevalence of coronary lesions. Even when treated promptly, however, approximately 6% of children with Kawasaki syndrome develop coronary artery abnormalities. We propose to construct a predictive instrument for the development of coronary artery abnormalities in patients with acute Kawasaki syndrome who develop such lesions despite early treatment with intravenous gamma globulin (IVGG). Our predictive instrument will be developed using the data base from the NHLBI-funded study on Treatment of Kawasaki Syndrome with Intravenous Gamma Globulin (HL34545), the largest prospectively-collected data base available in the United States. We will separate patients treated with IVGG in whom the coronary arteries were normal at the time of treatment into a development data set and test data sets. We will use multiple logistic regression and will validate models using test data sets. The predictive instrument will be based on sociodemographic variables, temperature, and the following laboratory parameters measured during the acute phase of the disease: serum IgG, white blood count, differential- white-cell count, absolute band count, absolute lymphocyte count, hemoglobin, hematocrit, platelet count, alpha-1-antitrypsin, albumin, and C-reactive protein. Variable selection will be determined by clinical considerations, significance testing, and overall model evaluation. We will evaluate models using the area under the ROC curve, a simultaneous measure of sensitivity and specificity; the R2 statistic, a measure of overall model fit; and the calibration curve and associated Hosmer-Lemeshow chi-square statistic. The output of the predictive instrument will be the predicted probability of developing coronary artery abnormalities. A ranking of risk into low, medium and high may be of more use to clinicians than a continuous probability between zero and one. We will propose cutoffs for use in a clinical setting and will validate them on the test data sets. Although several scoring systems have been constructed to identify those children at highest risk for development of coronary artery abnormalities, none is based solely on risk factors measured early enough in the course of the disease to be useful in patient management. Furthermore, the scoring systems were based on patients who were treated with aspirin alone, not with IVGG in addition to aspirin. The development of a risk score will allow physicians to identify low risk children in whom extensive and frequent cardiac testing may be unnecessary, as well as high-risk children who require closer monitoring and may be candidates for additional therapies.
