Abstract:

Objective To investigate the clinical features of pediatric Kawasaki disease (KD) complicated with pleural effusion (PE) and construct a clinical prediction model. Methods A retrospective review was conducted on clinical data of 462 children with KD from June 2017 to June 2023. Eleven KD children with PE were selected as the KD-PE group, and 118 patients without PE were selected as the KD control group. Using propensity score matching, the clinical data including clinical manifestation, auxiliary examination, treatment process and complications were collected and analyzed. Univariate and Lasso-Logistic regression were used to analyze influence factors of PE in children with KD. A line chart model for predicting KD-PE was constructed, and the prediction efficiency of the model was evaluated. Results The respiratory symptoms were generally mild in the KD-PE group, with 10 cases exhibiting mild PE. Two variables of C-reactive protein (CRP, OR=1.045, 95%CI: 1.009-1.082) and albumin (ALB, OR=0.755, 95%CI: 0.591-0.964) were screened out by Lasso regression and multi-factor Logistic regression to construct the prediction model. The regression equation was Logit (P)=2.221+0.044×CRP-0.281×ALB. The area under the ROC curve of the prediction model was 0.957 (95%CI: 0.911-1.000), and the Hosmer-Lemeshow test showed that the curve of the calibration model was well coincident with the curve of the actual model (χ²=4.320, P=0.827). The evaluation model using clinical decision curve and clinical impact curve had good clinical applicability. Conclusion KD complicated with PE in children is rare, elevated CRP is a risk factor for KD-complicated PE, and elevated ALB is a protective factor. The model has higher accuracy, discriminatory power and net benefit.

Key words: mucocutaneous lymph node syndrome, pleural effusion, C-reactive protein, serum albumin, nomograms