Clinical Management
Jigna Narendra Bathia, MBBS, DCH, DNB (Pediatrics), MRCPCH
Post Doctoral Fellow
Institute of Child Health, Kolkata
Kolkata, West Bengal, India
Introduction: Various risk scoring systems have been developed to predict IVIG resistance in KD, however, they have not been found to be useful in other ethnicities.
Objectives: The aim of the study is to produce a set of cutoff values for the parameters that would best predict IVIG resistance in KD in the Indian cohort of patients.
Methods: It is a retrospective analytical study on data of patients diagnosed as KD from January 2018 to April 2021, from Institute of Child Health, Kolkata. The basic framework of the 3 Japanese scoring methods were retained. Datas was divided into training and test datasets. The training dataset was used to produce the scoring mechanism, and testing dataset used to compare accuracy of proposed new score with the established methods. The training dataset had 70 patients, 22 being IVIG resistant. The testing dataset had 45,15 were IVIG resistant.
Scores based on original Kobayashi, where the cutoff (=/ >5), all the original Kobayashi predictors (but new cutoff value predicted for each), and the greater/lower than signage corresponding to each predictor was re-used to produce new possible modified Kobayashi scores. On Sano and Egami, process repeated in a similar way. Another new scoring method, using a structure similar to the established scoring methods, but selecting predictors based on Logistic Regression and testing different values of the cutoff for best performer.
For each score, the corresponding predictors and their list of values on which to test them were used to generate a grid. and possible new modified scores were then proposed for all three Japanese score. For each proposed score, the set of predicted labels for IVIG Resistance was calculated iteratively for each set of points on the grid, and the sensitivity and specificity were calculated. The set of values of the predictors that resulted in the best prediction accuracy were noted. The choice of predictor values that give the highest mean accuracy was chosen.
Results: The Kobayashi based approach produced best results. Using the same variables used in Kobayashi, and running an analysis on the training dataset, we came up with the following new values of the variables. (Table 1) Table 2 demonstrates the sensitivity and specificity of the proposed and the original Kobayashi score.
Conclusion: The proposed score can be seen to have better sensitivity. If validated, the new score can act asa screening tool to predict IVIG resistance in the Indian cohort.