Novel Score to Predict Immunoglobulin Resistance in Kawasaki Disease

To evaluate existing scoring systems and develop a new model to predict intravenous immunoglobulin (IVIG) resistance in patients with Kawasaki disease (KD). A retrospective cohort study performed between 2004 and 2017 identified 115 patients treated with IVIG for classic or incomplete KD. In our practice, IVIG resistance was defined as fever for > 24 h and patients were divided into responders and non-responders. A univariate analysis was performed to identify independent predictors of IVIG resistance. The predictors were combined into a new scoring system and compared with existing scoring systems. Sixty-five patients had classic KD and 50 had incomplete KD. Among the 115 patients, 80 (69.6%) responded and the remaining 35 were resistant (30.4%) to IVIG. Of the 35 resistant patients, 16 patients had incomplete KD. Hispanic children comprised 43% of our sample population. Coronary artery abnormalities developed in 14 of the 35 IVIG-resistant patients (39%). Univariate analysis showed that IVIG-resistant patients were older and present with lower platelets, potassium, and creatinine (P < 0.05). Multivariate logistic regression analysis used platelets, potassium, body surface area (BSA), and creatinine to devise the Las Vegas Scoring System (LVSS), which demonstrated a sensitivity of 76.2% and a specificity of 68.6%. Compared to published data, we observed a higher rate of IVIG resistance and coronary artery abnormalities in our patient population. The LVSS (using platelets, potassium, BSA, and creatinine) showed higher specificity and comparable sensitivity to other scoring systems devised to predict IVIG resistance.


Introduction
Kawasaki disease (KD) is the most common cause of acquired heart disease in children in developed countries. It mostly affects children < 5 years of age. The incidence of Kawasaki disease (KD) in the USA has been estimated from hospitalization rates at 20.8 per 100,000 [1,2]. Timely initiation of treatment with intravenous immunoglobulin (IVIG) has reduced the incidence of coronary artery aneurysms from 25 to 4% [3,4]. Nearly 15% to 23% of patients with KD develop persistent fever at least 24 h after the end of one dose of IVIG infusion and are termed IVIG resistant [5,6]. Because IVIG-resistant patients are at a higher risk for coronary artery aneurysms, it is important to identify these patients who might benefit from more aggressive initial therapy [7].
Three groups devised scoring systems to predict resistance to IVIG, namely Egami, Kobayashi, and Sano [8][9][10]. These have not been adequately validated in the North American patient population. The Egami score had a very low sensitivity for identifying IVIG non-responders with Kawasaki disease in a Midwestern US cohort [11]. Hence, those models are not found to be clinically useful in predicting responses to initial treatment with IVIG [12,13].
The purpose of this study is to evaluate sensitivity and specificity of existing scoring systems in a North American cohort of patients with KD. Additionally, we have developed a scoring system to predict IVIG non-responders for whom alternative intensive primary treatment could be explored.

Methods
This retrospective cohort study was done at University Medical Center (UMC, Las Vegas) between January 2004 and December 2017 after obtaining IRB approval. Patients aged less than 18 years at the time of admission for KD (classical or incomplete) who were treated with IVIG were identified from hospital coding data. Patients with KD who were not treated with IVIG were excluded. The diagnosis of classic KD was based on the presence of fever and at least 4 of the 5 principal clinical features (conjunctival injection, oral mucosal changes, polymorphous rash, lymphadenopathy, and extremity changes) [3]. Incomplete KD was diagnosed when all 3 below criteria were fulfilled: (1) unexplained fever associated with two or three of clinical criteria; (2) elevated C-reactive protein (CRP) level ≥ 0.5 mg/dL and/ or erythrocyte sedimentation rate (ESR) ≥ 15 mm/h; and (3) at least 3 supplemental laboratory criteria or positive echocardiogram. Patients' demographics, clinical features, laboratory findings (complete blood count, C-reactive protein, ESR, comprehensive metabolic panel), and echocardiographic results (coronary artery abnormality, left ventricular dysfunction, pericardial effusion, and mitral regurgitation) were recorded. Coronary artery abnormalities include aneurysm (Z score > 2.5), dilation/ectasia (Z score 2-2.5), and prominence/perivascular brightness. In our practice, IVIG resistance was defined as persistent fever for > 24 h after IVIG completion [3]. According to this definition, the patients were divided into the responsive and non-responsive IVIG groups.
The two groups were compared using χ 2 test or Fisher's exact test for nominal data and using unpaired Student's t test or Mann-Whitney U test for continuous data. A univariate analysis was performed on the clinical and laboratory variables to identify independent predictors of IVIG resistance. A multivariate logistic regression analysis was performed on the factors obtained from the univariate analysis. The predictors were combined into a new scoring system and compared with existing scoring systems. The discriminatory capacity of the scoring system was assessed using the area under the receiver operating characteristic (ROC) curves. Also, sensitivity and specificity of three existing scoring systems were calculated based on our data. A P < 0.05 was considered as either significantly different or correlated.

Results
During a 14-year period, 115 patients were admitted for KD and treated with IVIG. Ten patients who were not treated with IVIG and 2 patients with missing clinical data were excluded. The mean age of the cohort was 35 ± 32 months with male preponderance of 63%. The mean length of stay (LOS) in the hospital was 3.4 ± 1.61 days. Our population was predominantly Hispanic (43%). The most common symptom was conjunctival injection (88%), whereas induration/erythema of extremities was least common (45%). Mean duration of fever was 5.7 ± 3.3 days. Of note, 75 patients had a recent ED or office visit and 71 were treated with antibiotics prior to hospitalization. 70% of the patients received aspirin and IVIG before 6 days since the onset of symptoms, while 95% received initial treatment before 10 days. Aspirin was dosed at 80-100 mg/kg/day, while 2 g/kg of IVIG was used as the standard treatment.
Sixty-five patients had classic KD (56.5%) and 50 had incomplete KD (43.5%) diagnosed based on American Heart Association (AHA) criteria. The number of patients who responded to IVIG was 80 (69.6%) and the remaining 35 were resistant (30.4%). The clinical and laboratory data between the responsive and resistant groups are similar (Table 1). IVIG-resistant patients received a 2nd dose of IVIG in 34 cases and additional steroids were used to treat 5 patients. Of the 35 resistant patients, 16 (45.7%) patients had incomplete KD, whereas of the 80 responder patients, 34 (42.5%) patients had incomplete KD. Patients with incomplete KD had significantly lower age, WBC, creatinine, and AST than classic KD (P < 0.05, each). In the incomplete KD group, 68% were IVIG responsive and 32% were IVIG resistant (P < 0.05). Coronary artery abnormalities (aneurysm/dilation/ectasia/prominence) developed in 14 out of the 35 IVIG-resistant patients (40%) which is comparable to 31 out of 80 IVIG-responsive patients (39%) (P = 0.9). Coronary arteries were not measured in 60 patients. They were mostly evaluated to be normal (n = 45), few were subjectively thought to be dilated/ectatic (n = 6) or prominent (n = 9).
Univariate analysis showed that IVIG-resistant patients are older in age with lower platelets, serum potassium, serum creatinine, and had longer LOS. Multivariate 1 3 logistic regression analysis found platelets, potassium, BSA and creatinine to be independent predictors of IVIG resistance ( Table 2). In the LVSS, dichotomous variables were weighted based on the odds ratios of significant predictors as follows: platelets ≥ 380 × 10 3 /mm 3 (1 point), serum potassium ≤ 4 mEq/L (2 points), BSA ≥ 0.5/m 2 (1 point), and serum creatinine ≥ 0.4 mg/dL (2 points). Using a cut-off point of LVSS ≥ 4, we could identify the IVIGresistant group with 76% sensitivity and 68% specificity. The area under the ROC curve was 0.796. We also tested the Egami, Kobayashi, and Sano scoring systems and their sensitivity and specificity were compared. Our new scoring system has higher specificity than existing scoring systems (Fig. 1).

Discussion
Our study evaluated the Egami, Kobayashi, and Sano scores that were originally derived from the Japanese population. We found that they had higher sensitivity (67% to 95%) but lower specificity (9% to 23%) to predict IVIG resistance in our US dataset (Table 3). Similarly, Sleeper et al. evaluated them in North American cohort and found that sensitivity of these three risk scores was low (33% to 42%) and the specificity was high (85% to 87%) [12]. Tremoulet et al. found poor sensitivity of Egami score (38%) and devised a San Diego score with 73% sensitivity and 67% specificity [7]. The Egami and Kobayashi scoring systems had low sensitivity (14-16%) and high specificity (85-86%), while the San Diego system had high sensitivity (95%) but very low specificity (3%) when evaluated in China [14]. Hence, these scoring systems have limited utility in predicting IVIG resistance outside of the derived population. Moreover, these disparities could be due to regional differences in ethnic diversity of populations across North America which indirectly impacts disease severity and response to IVIG. A Pediatric Health Information System (PHIS) database study of 4811 patients from 27 US hospitals reported 60% males, median age 3.4 years, and 16% were Hispanics [5]. Our study had similar age and gender disposition but 43% identified as Hispanics. In another study, male sex was found to be an independent risk factor in North American cohort unlike the Japanese datasets which could reflect the genetic differences between cohorts as opposed to IVIG resistance being a sex-linked trait [12]. Although there was a male preponderance, our results do not corroborate the above preposition. In addition, the fact that KD is diagnosed later in the USA than Japan, with an average of 5.5 to 6.0 days of illness, perhaps one of the reasons why the Japanese scores are not satisfactory. For example, liver dysfunction, emphasized in the Japanese scores, tends to be more prominent in the early stages of the disease.
Kanamitsu et al. showed that the proportion of patients identified as being at high risk for IVIG resistance using the Kobayashi, Egami, and Sano risk scores, was not significantly different between the complete and incomplete KD groups [15]. The proportion of IVIG-resistant patients was 22% in typical KD patients and 17.6% in incomplete KD patients (P = 0.5) in the above study. Similarly, there was no significant difference in IVIG resistance between classic and incomplete groups in our study (29 vs 32%, P = 0.8). Moreover, IVIG resistance in this cohort is 30% which is higher than 9-23% reported in other studies [6,7,16]. Our study included patients with incomplete KD. In contrast, the three Japanese scoring systems to predict IVIG resistance have mostly excluded incomplete KD. This could explain some of the differences observed in this study. This is supported by higher sensitivity of scoring systems (San Diego and Las Vegas) developed in populations that included incomplete KD.
In our entire cohort, 39% had coronary artery abnormalities (dilation/ectasia/aneurysm and prominence). Out of this, 24% had dilation/ectasia/aneurysm and the remaining 15% were 'prominent.' The rate of coding of coronary artery abnormalities (CAA) for Hispanic patients was significantly higher than the rates for other racial and ethnic groups [5]. This is consistent with 39% of our patients who had CAA (mostly dilatation or prominence) when compared to 15-25% reported in prior studies [17,18]. This could explain the higher rates of CAA in this primarily Hispanic population. Male sex, older age, higher platelet count, lower albumin levels, and higher CRP levels were risk factors for cardiac lesions reported in one multicenter study [19]. We did not examine risk factors for CAA in our cohort as the higher proportion of incomplete KD could be a confounding factor. In addition, 13% of CAA were reported as dilated or prominent without Z-score measurements, which would hinder evaluating Son or other predictive scores for CAA [20,21]. Our novel scoring system included platelet count, potassium, creatinine, and BSA. The level of platelet-derived microparticles was higher in children with KD that was resistant to IVIG as compared to children with IVIGresponsive disease [22]. Hence, high platelet count reflects the severity of the underlying inflammation or severity of disease reported. On the contrary, both Egami and Kobayashi scores have platelet count less than 300 × 10 3 /mm 3 as risk factor for IVIG resistance. While liver enzymes such as ALT, GGT, and bilirubin are parameters for IVIG resistance in other scoring systems, potassium and creatinine are novel risk factors that have not been previously reported. This could imply that severe systemic inflammation and vasculitis involving the renal tubules and arterioles alters the potassium homeostasis and creatinine clearance.
The Las Vegas Scoring System could be used to develop an optimal treatment for patients who are predicted to be refractory to IVIG in future. The intensified primary IVIG therapies with steroids, cyclosporine, and infliximab have been proposed to reduce the incidence of refractory to IVIG, resulting in the decrease in CAA as well as medical costs by preventing unnecessary use of IVIG.
Limitations of our study include retrospective nature of the study, and lack of validation dataset to assess the reliability of risk factors that we identified in the novel scoring system. Future studies could evaluate cytokines or whole exome sequencing to identify IVIG resistance. Our practice was to consider persistent fever > 24 h as IVIG resistance which is more aggressive than > 36-48 h used in similar studies. This could have resulted in higher rates of IVIG resistance.

Conclusion
Higher rates of IVIG resistance and coronary artery abnormalities are seen in our predominantly Hispanic population. Our novel scoring system using platelets, potassium, BSA, and creatinine as predictors of IVIG resistance has a sensitivity of 76.2% and a specificity of 68.6%. The LVSS has comparable sensitivity and higher specificity than existing scoring systems to predict IVIG resistance.

Author Contributions All authors reviewed the manuscript.
Funding Resident Research Grant awarded by UNLV School of Medicine.