Non-SARS, non-MERS human coronavirus infections and risk of Kawasaki disease: a meta-analysis

Aim: To study the association between non-SARS, non-MERS human coronavirus (HCoV) infections and Kawasaki disease (KD). Methods: Meta-analysis of observational studies published until 1 May 2021. Results: Out of 571 papers retrieved through database search, 10 provided data of 17,732 children. Age ranged from 2 months–14.9 years with 66% being male and 71% being complete KD. Compared with controls, there was an increased risk of developing KD in those detected to have HCoV infection (OR: 2.3 [95% CI: 1.06–4.99]; p = 0.03). The GRADE evidence for all outcomes was of ‘low-certainty’. Conclusion: A ‘low certainty’ of evidence suggests an increased risk of KD in children infected with HCoV. We need multi-center, prospective studies to support or refute this finding. PROSPERO protocol registration: CRD42021251582.

infection (non-SARS and non-MERS) and KD. We tried to find out if HCoV infection was increasing the risk of development of KD in children of all age groups.

Materials & methods This systematic review protocol is registered at PROSPERO: CRD42021251582
Types of studies Observational studies that aimed at studying the association of HCoV infection with KD in both community and hospital settings were included.

Types of participants
Children of all age groups/sexes were included.
Types of exposure Participants in exposure group included those with HCoV infection (confirmed by either RT-PCR or serology) and KD. In control group, children had HCoV infection but no features of KD.

Types of outcome measures Primary
• Risk of development of KD in those infected with HCoVs.
No language restrictions were applied. The search results were reviewed to identify relevant studies.

Data extraction
The data were extracted by using a pre-designed data extraction form and tested a priori. We independently extracted the following information from each study: author, year, location (country), study design, setting (hospital or community), method of recruitment, inclusion criteria, risk of bias, participants (age, sex, sample size and diagnosis), exposure (type of viruses, methods employed for diagnosis and method of ascertainment), outcomes (outcome definition, valid unit of measurement, time points of collection and reporting), lost to follow up and miscellaneous data (key conclusions, references to other studies and additional data required).

Assessment of risk of bias in the included studies
We independently assessed quality by using the Newcastle-Ottawa scale (NOS) for observational studies [10]. It assesses the quality of studies under three major domains: selection of studies, comparability and outcome/exposure. Any disagreements were resolved by discussion.
Dealing with missing data The missing data were described that included dropouts in selected studies. Difference in dropout rates may lead to biased estimates of the effect size. Furthermore, bias may arise if the reasons for dropouts vary across different groups in different studies.

Data synthesis
Data were analyzed using Review Manager (RevMan, v.5.3, The Cochrane Collaboration, Copenhagen, Denmark) [11]. Data from included studies were pooled and expressed as odds ratio (OR) with 95% CI. A p-value < 0.05 was considered statistically significant. Inter-study heterogeneity was assessed by Cochrane's Q (χ 2 p < 0.10) and quantified by I 2 . An I 2 ≥50% indicated 'substantial' heterogeneity and ≥75% indicated 'considerable' heterogeneity [12]. Causes of substantial or considerable heterogeneity were explored and, sensitivity and/or sub-group analyses were carried out where feasible. We used both random-and fixed-effect models to see any difference in pooled effect. We constructed the funnel plot from primary outcome to ascertain any publication bias [13].

Grade of evidence
We used GRADE Profiler software (v.3.2) to assess certainty of evidence [14]. The software uses five parameters for rating the certainty of evidence. The parameters used are risk of bias, inconsistency of results or unexplained heterogeneity, indirectness of evidence, imprecision of results and publication bias.

Description of studies
Of 571 total citations retrieved, full text of 11 papers was assessed for eligibility, and one duplicate paper was excluded ( Figure 1) [15]. Of the remaining 10 eligible studies (17,732 children), we were able to do a meta-analysis of seven studies including 17,615 children (cases: 555; controls: 17,060) ( Table 1). Age range of included children was 0.2-14.9 years, 66% were male and 71% had classic KD. Eight studies were case-control studies [16][17][18][19][20][21][22][23] and two were prospective studies [24,25]. The studies had been conducted in the following countries: USA (n = 5), Japan (n = 2), Taiwan (n = 2), The Netherlands (n = 1) and Germany (n = 1). Two studies employed serology (antibody assay) and immunofluorescence for diagnosis of HCoV infection, whereas the rest employed viral detection (RT-   PCR and/or culture) method [20,21]. Respiratory samples (from nose, nasopharynx or throat) were used by studies that used viral detection methods. Different studies have assessed the association of different serotypes of HCoVs with KD, and details are depicted in Table 1.

Risk of bias in included studies
None of the studies was having a risk of bias in selection of cases, but four were at a high risk of bias for selection of controls [17,18,20,21]. Two studies were at high risk of bias for comparability of cases and controls [18,21]. All studies were having a low risk of bias for exposure parameters ( Table 2).

Risk of development of KD in those infected with novel HCoV
Overall analysis: Seven studies were able to contribute data to the pooled analysis. Both fixed-and random-effect models were employed to see whether the findings were significant or not, as significant heterogeneity was noted among included studies (I 2 = 58%; p = 0.03). There was an increased risk (odds) of developing KD in those detected to have infection with HCoV as found in both the models (random-effect model Subgroup analysis: we separately analyzed studies employing different diagnostic methods for HCoV (RT-PCR/viral detection or antibody assay). Overall analysis was not significant for studies employing either RT-PCR/viral detection (OR: 3.22 [95% CI: 0.79 to 13.08]; p = 0.1) or antibody assay (OR: 1.75 [95% CI: 0.96 to 3.18]; p = 0.07). As there was significant heterogeneity among studies, we did sensitivity analysis by removing one study [16]. Pooled data was significant without any remarkable heterogeneity (

Risk of development of KD in those infected with different serotypes of HCoV
We separately analyzed data from included studies that reported different serotypes of HCoV (NL63, OC43, 229E). However, none of the serotypes was individually associated with risk of development of KD.

Publication Bias
Funnel plot was asymmetrical showing publication bias ( Figure 3).

Grade (certainty) of evidence
Evidence generated was of 'low certainty' for all outcomes (primary and secondary). Likely reasons were heterogeneity among studies, poor methodological design and selective outcome reporting. A detailed analysis of the summary of evidence for primary outcome is provided in Table 3.

Summary of evidence
After an extensive search of the literature, we included ten studies with data of 17,732 children. As compared with controls, those having documented infection with HCoV were at a higher risk of developing KD. No particular serotype of HCoV was found to be associated with this increased risk. The evidence, however, is of 'low certainty'. In a case-control study published in 2005, Esper et al. first demonstrated the association between HCoV infection and KD [16]. Chang et al. confirmed this association in 2014 [23]. However, further case-control studies failed to replicate this association [17,20,21] which was likely due to relatively small sample sizes in the two older studies. Of the ten studies included in the present review, eight used RT-PCR for detection of HCoV, and two used serological assays. Use of serological assays may have been the reason for low detection rates of HCoV in these studies, as sensitivity of serological assays is lower than RT-PCR. Furthermore, there is a significant variability among studies based upon the interval between diagnosis of KD and RT-PCR testing for HCoV infection. Additionally, only a few studies have documented presence of concomitant respiratory symptoms (due to HCoV infection) and occurrence of KD.
In the present meta-analysis, we have analyzed the association of HCoV infections with KD. This has not been evaluated comprehensively to date. Recognition of association between HCoV with KD predates the KD or KDlike illnesses (in the form of multisystem inflammatory syndrome in children) seen in the ongoing SARS-CoV-2 pandemic by about a decade and a half. KD, KD-like illness (MIS-C/PIMS-TS) and severe COVID-19 infection induced vasculitis represent a similar disease spectrum characterized by a dysregulated immune response. It suggests that SARS-CoV-2 may just represent a 'new kid on the block' of various coronavirus infections associated with KD. In fact, KD has also been reported in association with the H1N1-pdm09 influenza infection that resulted in a pandemic a decade ago [26]. Besides, several other infectious illnesses have also been associated with KD [2]. Hence, there is a distinct possibility that there may be a surge in cases of KD or KD-like illnesses in future pandemics as well.
Exact etiology of KD remains an enigma. Many infectious illnesses can trigger KD in a genetically susceptible host. Of the various microbial agents, respiratory viruses have come up as the most important triggers for KD [2]. Causality has not been proven for any of the associated infectious agents to date, although, a KD-specific RNA respiratory virus has been postulated to exist based largely on autopsy studies [27]. Epidemiological profile of KD also favors the theory of an infectious trigger as it is predominantly seen in children below 5 years, is associated with seasonal clustering of cases with occasional epidemics, and the clinical features (such as fever, rash, cervical lymphadenopathy) mimic an infectious disorder.
It is plausible that HCoV infections may trigger KD illness in susceptible individuals. A number of genetic polymorphisms have been associated with an increased risk of developing KD in certain populations. Variability in HCoV-associated KD in different studies can therefore be easily explained by genetic differences among study populations [28]. It is possible that HCoV may be a risk factor for KD in populations with certain polymorphism(s) and not in others.
The exact pathogenesis of KD following HCoV infection is poorly understood and multifactorial. Endothelial dysfunction or inflammation is the hallmark of both KD and KD-like conditions that follow HCoV and SARS-CoV2 infection. HCoV has spike like protein that binds to angiotensin converting enzyme 2 (ACE2) receptors on surface of the target cell. A systematic review and meta-analysis have documented a significant risk of KD with ACE2 polymorphism [29]. Additionally, ACE2 polymorphisms have also been associated with other systemic vasculitides. SARS-CoV-2 has been shown to modulate ACE2 expression resulting with elevated TNF-α production [30]. Thus, modulation of ACE2 expression as a result of coronavirus infection or genetic polymorphisms may be one of the unifying pathogenetic mechanisms between the KD/KD-like illnesses following coronavirus infection and KD not associated with coronavirus infection.

Strengths & limitations
A comprehensive literature search, robust analysis and grading of evidence are strengths of the present review. Inclusion of only retrospective studies into the pooled analysis is a limitation (albeit unavoidable) of our review as the small sample size and retrospective nature may have introduced bias to overall estimate of results. Another limitation is that only three authors were involved in search methodology, risk of bias assessment and data extraction. This overlapping role might have introduced some bias in the review process. Furthermore, all studies were hospital based and were conducted mostly in developed countries. Hence, it may not be prudent to extrapolate the findings of this meta-analysis to the community at large.

Conclusion
A 'low certainty' of evidence suggests an increased risk of KD in children infected with HCoV. We need multicenter, prospective studies to support or refute this finding. The findings will be of great public health importance as KD/KD-like illnesses are closely associated with SARS-CoV-2 pandemic.

Future perspective
There should be elucidation of actual link between Kawasaki disease, and human coronaviruses (both SARS and non-SARS) at molecular level so that targeted therapy can develop as a form of prophylaxis.

Summary points
• Many studies have analyzed the association between SARS-CoV-2 and Kawasaki disease (KD)/KD-like illnesses with conflicting results, however, systematic review(s) or meta-analyses that have studied the association between human coronavirus (HCoV) infection (non-SARS and non-MERS) and KD are lacking.