Risk factors and an early predictive model for Kawasaki disease shock syndrome in Chinese children

Children who were initially diagnosed with KD in the Children’s Hospital Capital Institute of Pediatrics and Beijing Children’s Hospital from June, 2015 to July, 2023 were enrolled. We used the propensity score matching by age and gender to pair KDSS patients and KD patients without shock syndrome in a ratio of 1:4.

The eligible criteria were based on the American Heart Association Guidelines and Kanegaye criteria [1, 18]. The exclusion criteria were as follows: 1) Combined with sepsis shock or other cardiovascular, hypotension diseases; 2) Incomplete clinical records.

The conduct of this study was approved by the Ethics Committee of the Capital Institute of Pediatrics (Approval Number SHERLL2023007).

We collected age, gender and laboratory results which were conducted prior to the onset of KDSS and treatment with IVIG. Laboratory results including white blood cell (WBC), ANC, PLT, Hgb, CRP, procalcitonin (PCT), erythrocyte sedimentation rate (ESR), ALT, AST, Scr, Alb, interleukin-6 (IL-6), interleukin-8 (IL-8) and interleukin-10 (IL-10) were shared in both hospitals. If the laboratory measurements were repeated, we recorded the earliest indexes for further analyses.

Patients were tested for IL-6, IL-8 and IL-10 at admission before the onset of KDSS and treatment with IVIG. Serum cytokine concentrations were measured by chemiluminescence (Siemens Immulite 1000 chemiluminescence analyzer) in the Children’s Hospital Capital Institute of Pediatrics, where the reference ranges of IL-6, IL-8 and IL-10 were 0–3.4 ng/L, 0–62 ng/L, 0–9.1 ng/L, respectively. Concentrations of the three cytokines were measured by using the fluorescence-activated cell sorting (BD FACSCalibur™ Flow Cytometry instruments) in the Beijing Children’s Hospital, where the reference ranges of IL-6, IL-8, IL-10 were ≤ 5.4 ng/L, ≤ 20.6 ng/L, ≤ 12.9 ng/L, respectively.

Kolmogorov–Smirnov test was used to test the normality of continuous variables. Depending on whether the parameters were in Gaussian distributions, continuous variables are reported as median (interquartile ranges) or mean ± standard deviation, and between-group differences were compared using Two-sample Student’s t-test or Mann–Whitney test. Categorical variables are expressed as numbers (percentage) and compared using χ² test. A two-sided P value < 0.05 was considered statistically significant.

Spearman correlation coefficients were used to quantify relationship between the laboratory results. Univariable logistic regression and forward stepwise logistic regression were used to select significant and independent risk factors for KDSS. To facilitate clinical application, continuous variables were transformed into dichotomous variables. The value of IL-10 was normalized by using the clinical reference values. The optimal cut-off values of PLT, CRP, PCT and Alb were chosen according to the maximum Youden’s index. Finally, we used multivariate logistic regression to construct the predictive model. A nomogram plot was used to visually illustrate this model.

Internal validation was performed by using the bootstrapping method with 1000 repetitions. The receiver operator characteristic curve (ROC) and the calibration curve analysis were used to evaluate the discrimination and calibration of the predictive model in both development and case-independent validation datasets. Sensitivity, specificity, and the area under the receiver operating characteristic (AUC) were calculated. Decision curve analysis (DCA) was used to evaluate the net benefits of the predictive model. All statistical analyses were performed using the R software version 4.2.3 and IBM SPSS Statistics 27.0.1 software.

During the study period, a total of 3086 patients who were initially diagnosed with KD were included; 459 were excluded for missing clinical data; leaving 2627 in the final analysis. There were 2587 KD patients from the Children’s Hospital Capital Institute of Pediatrics, of whom 73 (2.8%) were KDSS. KDSS patients were older than KD patients without shock [3.57 ± 2.16 vs. 2.34 ± 1.59, P < 0.001], while no statistically significant difference in sex distribution was observed between the two groups [boys: 37 (50.7%) vs. 1539 (61.2%), P = 0.09]. There were 40 KD patients from the Beijing Children’s Hospital, and they met our inclusion and exclusion criteria for the KDSS. The average age was 3.59 (interquartile range: 2.42, 5.73) years and 26 (65%) were boys.

After propensity scores matching by age and gender, a dataset comprising 73 KDSS patients and 292 KD patients without shock syndrome from the Children’s Hospital Capital Institute of Pediatrics was formed for analyzing the risk factors and constructing a predictive model. Additionally, another dataset consisting of 40 KDSS patients from the Beijing Children’s Hospital and 160 KD patients without shock from the Children’s Hospital Capital Institute of Pediatrics was formed to validate this model. Demographic information and laboratory results between the two groups in development and case-independent validation datasets are presented in Table 1 and Supplemental Table 1, respectively.

As shown in Table 2, 9 candidate variables: WBC, ANC, PLT, CRP, PCT, ESR, Scr, Alb, IL-10 were associated with the risk of KDSS. We removed ANC from the further analysis because high correlation was found between WBC and ANC (Spearman’s r = 0.9; P < 0.001). By using the forward stepwise logistics regression, PLT, CRP, PCT, Alb and IL-10 were teased out as significant and independent risk factors of KDSS (Table 2).

For better popularizing in clinical practice, we normalized IL-10 by using the clinical reference values. PLT, CRP, PCT and Alb were transformed into dichotomous variables according to the maximum Youden’s index. The optimal cut-off points are shown in Supplementary Table 2. Incorporating these dichotomous indicators into multivariate logistic regression analysis revealed that IL-10 > reference value, PLT < 260 × 10⁹/L, CRP > 80 mg/ml, PCT > 1 ng/ml, and Alb < 35 g/L were independent risk factors for KDSS [IL-10 > reference value, odds ratio 2.836 (95% CI: 1.346–5.976), P = 0.006; PLT < 260 × 10⁹/L, odds ratio 3.343 (95% CI: 1.679–6.658), P = 0.001; CRP > 80 mg/ml, odds ratio 2.736 (95% CI: 1.293–5.789), P = 0.009; PCT > 1 ng/ml, odds ratio 6.151 (95% CI: 2.818–13.424), P < 0.001; Alb < 35 g/L, odds ratio 3.361 (95% CI: 1.581–7.145), P = 0.002]. A nomogram visualized the predictive model was constructed (Fig. 1).The score points for PLT < 260 × 109/L, CRP > 80 mg/ml, PCT > 1 ng/ml, and Alb < 35 g/L can be read on the Points per item axis. By summing up scoring points (range: 0–5) from all factors in the nomogram, we can calculate the total points that correspond to the Total Points axis. Subsequently, a vertical line is drawn downwards towards the Risk axis in order to determine the probability of KDSS. For instance, assuming a child with PLT at 310 × 10⁹/L (0 points), CRP at 126 mg/dl (3 points), PCT at 1.2 ng/ml (5 points), Alb 33.9 g/L (3.5 points) and IL-10 > reference value (3 points), the total points were 14.5, the estimated probability amounts to 65%.

The AUC for the nomogram was 0.91 (95% CI: 0.87–0.94) (Fig. 2a). The specificity and the sensitivity were 80% and 86%, respectively. After 1000 Bootstrap resampling internal validations, the AUC reached as high as 0.89. The calibration curve revealed a good fit during internal validation in patients with a 40%-80% probability of KDSS actually (Fig. 2b).

In the case-independent validation dataset, the model yielded an AUC value of 0.90 (95% CI: 0.71–0.86) with a specificity and sensitivity of 66% and 77%, respectively (Fig. 2d). Figure 2e shows good predictive accuracy of the nomogram in the case-independent validation dataset. The decision curve analyses revealed the predictive model has good application value (Fig. 2c and f).