Abstract: Objective To question the prediction model of intravenous gamma globulin (IVIG) resistance in children with Kawasaki disease (KD).Methods Medical records of KD patients, who were first diagnosed and treated in Children's Hospital of Fudan University, were retrospectively collected. All eligible KD patients were randomly assigned to the establishment group and the validation group at the ratio of 7∶3. A prediction model of IVIG resistance was constructed in the establishment group through univariate and multivariate logistic regression analysis, and then validated in the validation group. In addition, KD children were stratified by gender, age of onset, fever days before initial IVIG, KD type, etc. A new prediction model was constructed and validated separately in each layer. Finally, the published prediction models of IVIG resistance, which were established based on clinical manifestations and laboratory indicators, were applied to the whole sample population to evaluate the predictive value.Results A total of 1,360 KD children were enrolled, including 875 males (64.3%). The median age of onset was 1.8 (0.9, 3.2) years. There were 171 patients in the IVIG resistant group, 1,189 in the IVIG sensitive group, 952 in the establishment group and 408 in the validation group, respectively. Demographic characteristics, main clinical symptoms, laboratory indicators, and the rate of IVIG resistance and coronary artery lesions were not significantly different between the establishment group and the validation group (P >0.05). In the establishment group, the constructed prediction model included 1 score each for male, age of onset ≥2 years, N%≥0.75, Hb≥110 g·L^-1 and 2 scores each for fever days before initial IVIG ≥5 days, ALB≥34 g·L^-1, Na⁺≥133 mmol·L^-1. The AUC was 0.818 (95% CI: 0.774-0.861). With a cutoff point of ≥5 scores, the sensitivity and specificity were 0.767 and 0.726, respectively. In the validation group, the AUC was 0.777 (95% CI: 0.712-0.842) with the sensitivity and specificity of 0.627 and 0.776, respectively. When 11 published prediction models of IVIG resistance were applied to the whole sample population, the sensitivity and specificity were 0.272-0.799 and 0.412-0.926, respectively. Of the prediction models established in our and other studies, none had both sensitivity and specificity ≥75%.Conclusion Demographic characteristics, clinical symptoms, and laboratory indicators of KD children were insufficient to establish a clinically useful prediction model of IVIG resistance.

Key words: IVIG resistance, Kawasaki disease, Prediction