Multivariate logistic regression and ROC curve analysis
for predicting initial IVIG resistance
To not affect the prediction efficiency, the correlation between
univariate factors and coagulation profiles were tested by Pearson
analysis. The strong variables were removed from the model (Supplemental
material 2). Multivariate logistic regression analysis showed that
longer PT, APTT, higher D-dimer and lower ATIII activity were
independent risk factors for initial IVIG resistance in patients with KD
(Table 2).
The validity of the serum coagulation profile in predicting initial IVIG
resistance in patients with KD was assessed using ROC curve analysis.
The areas under the curve (AUC) of the various variables for predicting
IVIG resistance were 0.662, 0.599, 0.684, and 0.706 for PT, APTT,
D-dimer, and ATIII, respectively. The PT, APTT, D-dimer and ATIII cutoff
values of 13.95 s, 41.15 s, 1.48 mg/l, and 89.5% yielded sensitivities
of 73%, 32%, 71%, 81%; specificities of 55%, 88%, 62%, 51% for
predicting initial IVIG resistance, respectively (Table 3 and Figure
2A).