We have shown previously that patients with Kawasaki Disease (KD) resistant to IVIG are at risk for the development of coronary artery abnormalities. Prediction of IVIG resistance with the Kobayashi and Egami scores has been successful in Japanese children. However, they lack sensitivity when applied to North American patients. There has only been one published North American score (San Diego), and it is not currently used in clinical practice. We aimed to develop a new risk assessment score to detect IVIG resistance in KD and compare its performance to the Kobayashi and Egami scores in this single centre multiethnic Canadian cohort.

Methods:

Data from our retrospective cohort of patients with KD treated with at least one dose of IVIG (2 g/kg) and low dose ASA (< 10 mg/kg/day) between 01/2004 and 12/2014 were used. IVIG resistance was defined as the requirement for further treatment after the first dose of IVIG. Using available laboratory data and clinical characteristics at diagnosis, we developed a new risk assessment scoring system. ROC (receiver operative characteristic) analysis was used to examine the predictability of individual variables based on their optimal levels of sensitivity and specificity. Only variables with a minimum AUC (area under the curve) of 0.5 were considered for the new scoring system. Score weights were allotted to each selected variable based on their strength of predictability; weight 1 was assigned if the AUC was between 0.5 and 0.6, and weight 2 or 3 was assigned if the AUC was >0.6. Scoring criteria was optimized after testing for various combinations of variables. Finally, logistic regression was carried out to verify the strength of association using the final model.

Results:

From our cohort of 269 patients, analysis was carried out in those with complete KD (n=206) with all the necessary laboratory tests to construct the various scores (Kobayashi=129; Egami=120; New Score=137). Variables with AUC <0.5 were not included in the scoring system: fever duration at diagnosis, sodium and albumin. Variables with an AUC >0.5 were included: age, percent neutrophils, platelet count, CRP, AST and ALT. From these, we generated a 10-point score with an ideal cutoff of 4. Our new scoring system predicted IVIG resistance with a sensitivity of 70.8% (95% CI 55.9-83%), specificity of 62.9% (52.0–72.0), PPV of 50.7% (38.2-63.2), NPV of 80% (68.7-88.6) and AUC of 0.72 (0.63-0.81). After adjusting for age, sex, and fever duration in the logistic regression, patients with the positive new score (>=4) showed a strong, predictable association with IVIG resistance (OR=4.28; 95% CI=1.95-9.37). In our cohort, the sensitivity of the Kobayashi and Egami scores was relatively low at 38.6% and 29.7% respectively, but specificity was high at 87.1% and 87.6%.

Conclusion:

Our new risk assessment score performed better than the Kobayashi and Egami scores in predicting IVIG resistance in KD in this multiethnic Canadian cohort. However, it did not show additional utility compared to the San Diego score. Improvements are required to the proposed new score, and further analysis for this is currently underway.