Evaluating the performance of egami, kobayashi and sano scores in predicting IVIG resistance in infant kawasaki disease

Abstract

Background

This study aimed to evaluate the effectiveness of Egami, Kobayashi and Sano scores in predicting intravenous immunoglobulin (IVIG) resistance in infant Kawasaki disease (KD), considering its unique clinical presentation.

Methods

We retrospectively analysed 143 infants aged < 12 months and diagnosed with KD at a single centre from 2019 to 2023. Patients were divided into IVIG-resistant and IVIG-responsive groups. Demographic, clinical and laboratory data were compared between the groups. The diagnostic performance of Egami, Kobayashi and Sano scores in predicting IVIG resistance was evaluated using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and area under the receiver operating characteristic curve (AUC). Additionally, we developed a new scoring system based on significant predictors identified in our cohort.

Results

Among 143 infants, 45 (31.5%) showed IVIG resistance. The IVIG-resistant group had a significantly higher rate of coronary artery lesions (15.6% vs. 5.1%, p = 0.036). Incomplete KD was observed in 61.5% of cases. Egami, Kobayashi and Sano scores exhibited low sensitivity (35.6%, 55.6% and 20%, respectively) and moderate specificity (77.6%, 63.3% and 95.9%, respectively) in predicting IVIG resistance. The AUC ranged from 0.583 to 0.674, indicating poor to fair discriminative ability. Our newly developed scoring system, based on total bilirubin and albumin levels, showed similar performance (AUC 0.633) to existing scores.

Conclusions

Existing Japanese risk scoring systems and our newly developed score showed limited effectiveness in predicting IVIG resistance in infant KD. The high proportion of incomplete presentation and IVIG resistance in infants highlights the need for age-specific risk assessment and management. Further research is necessary to develop more sophisticated, dedicated prediction model for IVIG resistance in infants with KD.

Background

Kawasaki disease (KD), a systemic vasculitis, is an acute, self-limited febrile illness with an unknown aetiology. KD primarily affects children aged under 5 years and is the most common cause of acquired heart disease in developed countries. If not properly treated, complications, such as coronary artery lesion (CAL), can develop [1].

The current standard therapy for KD is intravenous immunoglobulin (IVIG) and aspirin (ASA). However, despite appropriate treatment, 15–25% of patients treated with IVIG show resistance and require additional therapy [2]. Patients with IVIG resistance have an increased risk of developing CAL. Therefore, early identification of patients likely to show IVIG resistance is crucial for timely management and prevention of complications [3, 4].

Risk scoring systems were developed in Japan, including the Egami score [5], the Kobayashi score [6], and the Sano score [7], to predict IVIG resistance. These scoring systems demonstrated high sensitivity and specificity in predicting IVIG resistance in Japanese KD patients [8].

Compared to other age groups, infants with KD pose a unique challenge due to the less frequent manifestation of characteristic KD symptoms, such as eye or oral mucosal changes [9]. Young age itself is a major risk factor for CALs in Kawasaki disease [10, 11]. The atypical presentation in infants can make timely diagnosis challenging. Given these factors, predicting IVIG resistance is particularly important in this age group to optimize treatment strategies and potentially reduce the risk of CALs in infant KD.

This study aimed to evaluate the efficacy of Egami, Kobayashi and Sano scores, which are commonly used risk-scoring systems for predicting IVIG resistance, in infant KD.

Methods

This study included 143 patients diagnosed with KD aged < 12 months at a tertiary medical centre (Kyungpook National University Children’s Hospital, Daegu, Korea) from January 2019 to December 2023. Patients treated at other hospitals or with underlying diseases, such as congenital heart disease, were excluded. The electronic medical records of the enrolled patients were analysed retrospectively.

The diagnosis of KD was established according to the guidelines published by the American Heart Association (AHA) in 2017 [1] Complete KD was diagnosed in patients who had a fever lasting ≥ 5 days (with the first calendar day of fever being illness day 1) and ≥ 4 of the 5 principal clinical features (erythema and cracking of lips, strawberry tongue, and/or erythema of oral and pharyngeal mucosa; bilateral bulbar conjunctival injection without exudate; rash; erythema and oedema of the hands and feet in the acute phase and/or periungual desquamation in the subacute phase; cervical lymphadenopathy [≥ 1.5-cm diameter]). According to the AHA guidelines, incomplete KD was diagnosed in patients with a fever lasting ≥ 5 days and at least 2 compatible clinical criteria or in infants with unexplained fever lasting ≥ 7 days and laboratory or positive echocardiogram findings, such as coronary artery aneurysm.

After diagnosis, all patients were treated with 2 g/kg IVIG. Concurrently, they received 30–50 mg/kg/day ASA until they reached an afebrile state after the IVIG treatment. Once the patient remained afebrile for at least 2 days, the ASA dose was decreased to 3–5 mg/kg/day.

According to the AHA guidelines, IVIG resistance was defined as the presence of recrudescent or persistent fever at least 36 h after the IVIG infusion [1]. Patients who exhibited IVIG resistance received retreatment with 2 g/kg IVIG or 30 mg/kg methylprednisolone for 3 consecutive days.

Echocardiography was performed for all patients within 1–2 days after admission during the acute phase to evaluate for CALs. Echocardiography was performed again 6–8 weeks after the treatment during the convalescent phase to re-confirm the presence of CALs. Coronary artery luminal dimensions were converted to body-surface-area-adjusted z-scores. Patients were considered to have CAL if the z-score of at least one of the specified arteries (left main coronary artery, left anterior descending coronary artery or right coronary artery) was ≥ 2.5 [12].

Patients were divided into two groups based on the presence or absence of IVIG resistance: IVIG-resistant and IVIG-responsive groups. Demographic data, clinical information and laboratory results were compared and analysed between the groups.

The following demographic variables and clinical information were analysed by reviewing the medical records: age at onset, sex, body weight, illness duration, total hospitalisation duration, the number of clinical symptoms meeting the principal criteria and the presence of incomplete KD and CALs.

Laboratory tests were performed on the first day of admission before the IVIG treatment. The following parameters were measured: white blood cell (WBC) count, percentage of neutrophils in WBCs (% neutrophils), percentage of lymphocytes in WBCs (% lymphocytes), percentage of monocytes in WBCs (% monocytes), percentage of eosinophils in WBCs (% eosinophils), haemoglobin, red blood cell distribution width (RDW), platelet count, aspartate aminotransferase (AST), alanine aminotransferase (ALT), albumin, total bilirubin, sodium, erythrocyte sedimentation rate (ESR), C-reactive protein (CRP) and N-terminal pro-brain natriuretic peptide (NT-proBNP).

The Egami, Kobayashi and Sano scores, risk-scoring systems used to predict IVIG resistance in KD, were applied to all subjects. Table 1 presents the information used to calculate the scores [5,6,7]. The cut-off values for each scoring system were determined based on the original studies that developed these systems. Patients were classified as high-risk if they had an Egami score of ≥ 3, a Kobayashi score of ≥ 4 or a Sano score of ≥ 2. The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of each scoring system were calculated to evaluate the efficacy of each scoring system in predicting IVIG resistance in infant KD.

Table 1 Scoring systems to predict IVIG resistance

Statistical analysis

Statistical analyses were performed using IBM SPSS Statistics for Windows, version 26.0 (IBM Co., Armonk, NY, USA). Continuous variables were expressed as the mean ± standard deviation or the median and interquartile range (IQR, 25th–75th percentile) according to normal distribution. Nominal variables were expressed as numbers and percentages. The 2 independent samples t-test was used to compare groups of normally distributed data, whereas the Mann-Whitney test was used to compare groups of non-normally distributed data. The χ² test was used for nominal variables. To identify potential IVIG resistance risk factors, variables showing significant differences between two groups were subjected to multivariate logistic regression analysis. Receiver operating characteristic (ROC) curves were used to determine the optimal cut-off values for parameters found to be significant in the multivariate analysis. Based on these cut-off values, a simple scoring system was developed, assigning 0 or 1 point for each parameter. The discriminatory capacity of each scoring system, including the newly developed one, was evaluated using the area under the ROC curve. Statistical significance was set at p < 0.05.

Results

Patient characteristics and clinical manifestations

From January 2019 to December 2023, 143 infants aged < 12 months were diagnosed with and treated for KD at our single centre over 5 years. Among them, 95 (66.4%) were male. The mean age of the cohort was 6.9 ± 2.6 months, and the mean hospitalisation duration was 5.59 ± 2.21 days. The mean duration of fever before treatment was 4.5 ± 1.97 days. The most common symptom was skin rash (95.8%), while the least common symptom was cervical lymphadenopathy (12.6%). Incomplete KD was observed in 88 (61.5%) cases, and CALs were detected in 12 (8.4%) patients. Forty-five patients (31.5%) showed resistance to IVIG treatment (the IVIG-resistant group), while the remaining 98 patients responded to the first IVIG treatment (the IVIG-responsive group).

Comparisons between IVIG-resistant and IVIG-responsive groups

Compared to the IVIG-responsive group, the IVIG-resistant group showed a significantly higher CAL rate (5.1% vs. 15.6%, p = 0.036) and longer hospitalisation duration (4.68 ± 1.11 days vs. 7.58 ± 2.67 days, p < 0.001). Age at onset, sex, body weight, fever duration before treatment initiation and clinical symptoms did not differ between the groups (Table 2).

Table 2 Comparison of the demographic and clinical characteristics between the two groups

Three laboratory findings before treatment initiation showed statistically significant differences between the groups. The IVIG-resistant group showed a higher median CRP (52.7 mg/L [IQR 33.4–89.3] vs. 39.1 mg/L [IQR 27.1–57.0]; p = 0.018) and median total bilirubin (0.49 mg/dL [IQR 0.32–0.7] vs. 0.38 mg/dL [IQR 0.24–0.52]; p = 0.003). Additionally, the median albumin was significantly lower in the IVIG-resistant group (41 g/L [IQR 38–44] vs. 43 g/L [IQR 40–46]; p = 0.008). Other laboratory findings did not show significant differences between the groups (Table 3).

Table 3 Comparison of laboratory findings between the two groups

Evaluation of existing IVIG resistance prediction scores

The effectiveness of Egami, Kobayashi and Sano scores in predicting IVIG resistance was evaluated by applying these scores to both groups. Table 4 demonstrates the distribution of high and low risks within each group according to each scoring system. Diagnostic tests and ROC curve analysis were performed to evaluate the actual IVIG resistance diagnosis based on Egami, Kobayashi and Sano scores. First, using the Egami score, 38 patients were classified as high-risk with 35.6% sensitivity, 77.6% specificity, 42.1% PPV and 72.4% NPV. The AUC of the Egami score was 0.583 (95% confidence interval [CI] 0.481–0.684). Second, using the Kobayashi score, 61 patients were classified as high-risk with 55.6% sensitivity, 63.3% specificity, 41% PPV and 75.6% NPV. The AUC of the Kobayashi score was 0.610 (95% CI 0.507–0.713). Lastly, using the Sano score, 13 patients were classified as high-risk with 20% sensitivity, 95.9% specificity, 69.2% PPV and 72.3% NPV. The AUC of the Sano score was 0.674 (95% CI 0.573–0.774) (Table 4; Fig. 1).

Table 4 Diagnostic performance of scoring systems

Fig. 1

ROC curves for predicting IVIG resistance in infant Kawasaki Disease using existing scoring systems. The graph shows the performance of Egami (AUC = 0.583, 95% CI: 0.481–0.684), Kobayashi (AUC = 0.610, 95% CI: 0.507–0.713), and Sano (AUC = 0.674, 95% CI: 0.573–0.774) scores. The diagonal line represents random chance (AUC = 0.5)

Development and evaluation of a new scoring system for ivig resistance prediction

Based on the findings from our multivariate analysis, we developed a new scoring system using the significant predictors of IVIG resistance in our infant population. Total bilirubin and albumin were identified as significant predictors (p < 0.05) in the multivariate analysis.

We determined the optimal cut-off values for these parameters using ROC curve analysis. The cut-off value for total bilirubin was 0.46 mg/dL, and for albumin, it was 3.3 g/dL. We assigned 1 point for total bilirubin ≥ 0.46 mg/dL and 1 point for albumin < 3.3 g/dL, creating a simple scoring system with a range of 0–2 points.

Using ROC curve analysis, we evaluated the performance of this new scoring system. At a cut-off score of 1, the sensitivity was 57.8%, and the specificity was 67.4%. The AUC for this new scoring system was 0.633 (95% CI 0.533–0.733).

Discussion

This study investigated the clinical characteristics of KD in infants and evaluated the effectiveness of existing IVIG resistance prediction scoring systems in this age group. Infants with KD presented with a higher proportion of incomplete clinical symptoms compared to older children, which is consistent with previous reports. Moreover, the existing risk scoring systems, such as Egami, Kobayashi and Sano scores, had limited effectiveness in predicting IVIG resistance in infant KD. The atypical presentation and the lack of reliable predictive tools pose a challenge for the early identification and timely treatment of high-risk infants. Furthermore, we observed a higher rate of IVIG resistance in infants compared to the general paediatric population with KD. These results underscore the unique features of infant KD and highlight the need for age-specific risk assessment and management.

Administering IVIG treatment timely is crucial to prevent CALs in KD. The diagnosis of KD is based on clinical criteria; therefore, the most critical aspects of timely diagnosis and treatment are meticulous history-taking and thorough physical examination. However, clinical symptoms may not be as apparent in infants as in other age groups. Furthermore, unexplained high fever and marked irritability may be the only presenting symptoms of KD in very young infants [13]. Consequently, incomplete clinical presentation in infants can delay diagnosis and treatment, increasing the risk of CALs. In a Japanese nationwide survey of KD involving all paediatric age groups, incomplete KD accounted for 20.6% of cases [14]. However, the present infant-focused study identified incomplete KD in most subjects (88 patients, 61.5%). Additionally, while IVIG resistance is generally reported in 15–25% of KD patients, this study, limited to infants, found IVIG resistance in 31% of patients. The IVIG-resistant group had a statistically significantly higher incidence of CALs than the IVIG-responsive group (15.6% vs. 5.1%; p = 0.036). Therefore, a high degree of suspicion for KD is essential when infants present with prolonged fever, and the ability to predict IVIG resistance is also necessary to prevent CALs.

Scoring systems were developed in Japan to predict IVIG resistance, with Egami, Kobayashi and Sano scores being the most representative systems [8]. The Egami score was derived from a study conducted in 2006 on 320 Japanese patients and demonstrated a sensitivity of 78% and a specificity of 76% in predicting IVIG resistance. Additionally, it showed a sensitivity of 61% and a specificity of 81% in predicting CALs [5]. The Kobayashi score was also developed in 2006 and was based on 750 Japanese KD patients. This score demonstrated a sensitivity of 86% and a specificity of 67% in predicting IVIG resistance [6]. The Sano score was developed in 2007 from a study involving 112 Japanese patients and showed a sensitivity of 77% and a specificity of 86% in predicting IVIG resistance [7]. All three scoring systems demonstrated high efficacy in predicting IVIG resistance in their respective studies. However, when applied to non-Japanese populations, they were not as effective. When the three scoring systems were applied to a North American population, the specificity was good, but the sensitivity was low, failing to accurately identify IVIG resistance [15]. Similarly, studies involving other non-Asian populations found that the Japanese scoring systems were ineffective in predicting IVIG resistance [16, 17]. A study conducted on 350 Koreans evaluated the effectiveness of Kobayashi and Egami risk scores and showed their good specificity but low sensitivity [18]. However, subjects aged < 2 years were excluded for propensity score matching in this study. We evaluated whether these scoring systems effectively predicted IVIG resistance in infant KD. All three had an AUC < 0.7, with the Sano score showing very high specificity (95.9%) but very low sensitivity (20%). Our findings suggest that Egami, Kobayashi and Sano scores have limited effectiveness in predicting IVIG resistance in infant KD. In an attempt to address this limitation, we developed a new scoring system based on the significant predictors identified in our multivariate analysis: total bilirubin and albumin. However, the performance of our new scoring system was comparable to that of the existing scoring systems evaluated in this study. This result further emphasizes the complexity of predicting IVIG resistance in infants with KD and highlights the need for more sophisticated models tailored to this specific age group.

The limited predictive power of both existing and our new scoring systems in this infant population underscores the unique challenges in managing KD in very young patients. It suggests that the pathophysiology and risk factors for IVIG resistance in infants might differ from those in older children, necessitating a different approach to risk stratification and treatment decision-making in this age group.

Currently, studies focusing solely on infant KD are lacking, while the few known studies were conducted with very small sample sizes. An Italian cohort study involving 32 infants found a similarly high proportion of incomplete KD, accounting for most cases [11]. Another study involving 17 infants aged < 6 months also found a high proportion of incomplete KD and emphasised the need for high suspicion of KD for young infants with unexplained fever beyond 5 days [19]. The present study included a relatively large number of infants and compared their characteristics between IVIG-resistant and IVIG-responsive groups. The factors that showed statistically significant differences were CRP, total bilirubin and albumin, which may help predict IVIG resistance in infant KD. However, our attempt to create a new scoring system based on these factors demonstrated that simple, linear models might not be sufficient to capture the complexity of IVIG resistance prediction in infants.

This study has several limitations. Due to its retrospective design, there is always a possibility of bias, which might have influenced data accuracy and reliability. Additionally, as a single-centre study, the generalisability of the findings may be limited, as the patient population and clinical practices may differ from other institutions or regions. Lastly, the sample size may be relatively small, possibly affecting the statistical power of the study and the ability to detect significant associations or differences between the groups. These limitations should be considered when interpreting the results and drawing conclusions from this study.

Nevertheless, the study provides valuable insights into the unique characteristics of infant KD and the need for a new approach to IVIG resistance prediction in this age group. Our findings suggest that future research should focus on developing more complex, possibly machine learning-based models that can better capture the intricate interplay of factors contributing to IVIG resistance in infants with KD. Additionally, multi-centre studies with larger sample sizes are needed to validate these findings and potentially identify additional risk factors specific to infant KD.

Conclusion

We confirmed the limited effectiveness of existing Japanese risk-scoring systems in predicting IVIG resistance in infants with KD. Our attempt to develop a new scoring system specifically for infants also showed limited predictive power, highlighting the unique features of infant KD and the complexity of IVIG resistance in this age group. The high proportion of incomplete KD and IVIG resistance in infants emphasizes the need for heightened clinical suspicion and potentially more aggressive initial management strategies in this vulnerable population.

Further research, particularly multi-centre studies with larger sample sizes, is needed to develop and validate more sophisticated, age-specific predictive models for infant KD. This will improve our understanding of IVIG resistance risk factors in this population and help optimize treatment strategies to reduce CAL incidence. Our findings underscore the need for a tailored approach to risk assessment and treatment in infant KD, potentially leading to more personalized and effective management strategies in this unique patient group.