The effect of vitamin D status on the occurrence of Kawasaki Disease: a meta-analysis

This systematic review protocol was submitted for registration to the International Prospective Register of Systematic Reviews (http://​www.​crd.​york.​ac.​uk/​prospero) on 11 November 2023 and published on 21 November 2023 (Registration ID: CRD42023399850). The study adheres to the PRISMA guidelines [30] and has completed its checklist. Searches were conducted in PubMed and Web of Science datasets for studies in English exploring the relationship between vitamin D status and KD. The search terms used were (Kawasaki disease OR Kawasaki-like syndrome OR Kawasaki syndrome OR mucoskin lymphnode syndrome) AND (vitamin D OR 25(OH)D OR 25(OH)D3 OR 25-hydroxyvitamin D3). A similar search was also conducted in China National Knowledge Infrastructure (CNKI) and Wanfang datasets using the terms “Vitamin D” and “Kawasaki Disease” for relevant reports in Chinese. Only studies published until February 2023 were included in this study, and any disagreements were resolved through discussion until consensus was reached or with the input of a third author.

The studies included in this analysis should satisfy the following eligibility criteria: (i) there must be two types of participants included, based on the diagnosis guidelines from the American Academy of Pediatrics and Cardiology Society [4], the Kawasaki Disease Research Committee in Japan in 2002 [31], and Editorial Committee of Chinese Journal of Pediatrics [32]: those who were diagnosed as KD patients with or without coronary artery lesions (CAL), and healthy individuals who are matched for age and sex; (ii) data on the mean level of 25(OH)D along with its standard deviations (SD) and/or standard error (SE) must be present for both groups of KD patients and healthy controls; and (iii) studies must offer other relevant information that highlights the difference of vitamin D status between KD patients and controls. Conversely, studies irrelevant to the association between vitamin D and KD, repeated reports or meeting summaries, and studies that lack original data will not be included in this study.

This meta-analysis involved two independent authors who extracted specific data, which included: (i) the first author, time of publication, and sample size; (ii) the mean serum vitamin D level, standard deviation, and/or standard error for both KD and control groups; (iii) ethnicity, mean age of subjects, proportion of females in the sample, and complications of participants; and (iv) other format data that could be used to calculate the effect size value. As the main complication of KD, the difference between patients with CAL and without CAL was often discussed [15, 20, 26], and patients were categorized into three groups—KD patients with CAL (CAL group), KD patients without CAL (NCAL group), and mixed group—and were used in subsequent subgroup analysis to determine the influence of patient complication on the relationship between vitamin D status and KD. When recording the concentration value of serum 25(OH)D, the data of different units were converted and unified as ng/mL. Additionally, the quality of all articles was evaluated by two authors, in accordance with the Newcastle–Ottawa Scale (NOS) assessment scale [33].

Two R packages, meta and metafor, were employed to conduct meta-analysis [34, 35]. The effect size of each study was measured using standard mean difference (SMD) and odds ratio (OR) effect values, alongside their 95% confidence intervals (95% CIs), to compare KD and control groups. As per Cuijpers’ recommendation [36], the more conservative random-effects model was used, which assumes that all studies stem from “multiple populations” and the true effect is normally distributed. To assess heterogeneity among the included studies, both Cochran’s Q-statistic and I 2 statistics were employed.

Studies with extreme effect sizes, also known as outliers, may raise concerns and distort the overall results of a meta-analysis [35]. To address this issue, an R program called influence.analysis was utilized. This program detects and removes outlier(s) among all eligible studies using various influence measures (e.g., DIFFITS, Cook’s distance, covariance ratio, etc.). After removing the outlier(s), the meta-analysis was conducted again. To account for the significant heterogeneity (P_het ≤ 0.05 or I ² ≥ 25%) observed, the sources of heterogeneity will be considered [37]. Begg’s rank correlation test and Egger’s weighted regression test will be conducted to assess the risk of publication bias. The results of the trim-and-fill test will be visualized through a funnel plot. Additionally, the leave-one-out function will be used to examine the sensitivity of the results. To maintain the Type I error rate, the “permutest” command of the Metafor package will be utilized to conduct a permutation test with 1,000 iterations.

In order to investigate the potential causes of the observed heterogeneity and determine their impact on the correlation between vitamin D and KD, a meta-regression analysis was conducted using a mixed-effects model (which utilized a random-effects model within subgroups and a fixed-effects model among subgroups) [38]. Additionally, the model test (Q_M) and goodness of fit test (Q_G) were employed to gauge the effect of moderators on the relationship between vitamin D status and KD, and to determine whether other factors may contribute to the variability in effect size [35]. This meta-regression analysis involved five types of information per sample, namely, publication date, mean age of participants, the percentage of females, race, and presence of KD complications, all of which were considered as possible covariates. All significance tests were two-tailed, with a significance threshold set at 0.05.

Initially, 82 publications pertaining to the correlation between vitamin D status and KD were examined based on the inclusion criteria, as depicted in Fig. 1. Upon review of the title and abstracts, 50 studies were excluded as they were deemed irrelevant to the relationship between vitamin D and KD. Out of the remaining 32 studies, two reports were found to be duplicated, five were reviews, four were clinical trial studies, and two focused on cellular research work. After excluding two studies lacking effective data, a total of 17 eligible studies were identified and included in the ensuing meta-analysis. The eligible studies comprised 23 samples, consisting of 1,394 KD patients and 1,557 healthy controls. Of these KD patients, 272 had CAL, 423 did not suffer from CAL complications, and the remaining did not report their complications. The study included mostly Han Chinese participants (72.6%), with one sample (12.7%) from Japan [16] and two samples (14.7%) from Italy [13, 18]. This distribution is consistent with the higher incidence of KD observed in Northeast Asian countries like China, Japan, Korea, and Taiwan compared to other developed nations such as the United States and Europe. The age of the participants ranged from 0 to 5 years, with a mean age of 2.8 ± 0.8 years, and 41.9% were females. Detailed information regarding the eligible studies is provided in Table 1.

In the primary meta-analysis, Fig. 2 depicts the variance in vitamin status observed in KD patients compared to healthy control participants, with the pooled difference estimated under the random-effect model. Results indicate that patients with KD had significantly lower 25(OH)D levels than healthy controls (studies N = 23; SMD (95%CI): -0.94 (-1.72 to -0.15) ng/mL; overall effect: z = -2.35, P = 0.019). This corresponds to an increased risk of KD onset (OR: 5.50, 95%CI: 1.32 to 22.73) per 1-SD decrease in the level of vitamin D among children. The influence.analysis test found that the study of An et al. [25] was an outlier with low-quality data and an extreme effect size (Data S1 and Figure S1). Upon its removal, results consistently showed a significant effect of serum 25(OH)D levels on KD (studies N = 22; SMD (95%CI): -1.30 (-2.05 to -0.55) ng/mL; overall effect: z = -3.40, P < 0.001; Table 2). That was to say, for per-SD decrease in serum 25(OH)D level in children, the risk of KD increased by 11.11 (95%CI: 2.89 to 43.48) times. The permutation test with 1,000 iterations confirmed this finding (P = 0.007). However, a high degree of heterogeneity was still observed among these studies (Cochran’s Q = 1,206, df = 21; and I ² = 98.26; P < 0.001). Sensitivity analysis showed that the result was robust, with the effect size (from -1.46 to -1.14 ng/mL) remaining within the range of the overall effect’s 95%CI interval when any single study was removed (Figure S2). Notably, neither Begg’s rank method (Tau = -0.06, z = 0.37, and P = 0.714) nor Egger’s linear regression test (t = 0.53, df = 20, and P = 0.603) detected significant publication bias in this meta-analysis.

Further meta-analyses were conducted to compare subgroups. The studies were divided into two groups based on the nationality of the subjects: Chinese, and other countries. The results showed a consistently significant difference in serum 25(OH)D levels between children with KD and healthy controls in both groups (with Chinese samples: N = 19, SMD (95%CI): -1.22 (-2.14 to -0.31) ng/mL, overall effect: z = -2.61, P = 0.009; with other countries samples: N = 3, SMD (95%CI): -1.97 (-3.18 to -0.76) ng/mL, overall effect: z = -3.19, P = 0.001; see Table 2). However, when considering the complications of patients, inconsistent results were found between the three groups (mix; NCAL; and CAL). The significant effect size of serum 25(OH)D level was only observed in NCAL patients, mixed patients, and healthy controls (two Ps < 0.019; see Table 2 and Fig. 2).

To investigate the potential reasons for the significant heterogeneity observed in the studies included in our analysis, a meta-regression analysis was conducted. The findings revealed that the variables we examined (e.g., participant age, gender composition, publication date, race, and complications) did not contribute significantly to the observed heterogeneity (all P > 0.05). Furthermore, results from our model indicated that these five variables had no significant impact on the effect size of serum 25(OH)D levels on KD (Q_M = 2.89, df = 6, P = 0.82; Table S1). However, the goodness-of fit analysis showed that this model was incomplete and did not provide an accurate estimate of the true effect size of 25(OH)D on KD (Q_G = 834.96, df = 12, P < 0.01; Table S1).