N-terminal pro-brain natriuretic peptide as a biomarker for predicting coronary artery lesion of Kawasaki disease

Coronary artery lesion (CAL) caused by Kawasaki disease (KD) is currently the most common acquired heart disease in children in many countries. Nevertheless, there is no single useful marker existing for predicting CAL of KD. Recently, many reports have noted that N-terminal pro-brain natriuretic peptide (NT-proBNP) can be utilized as a biomarker to predict CAL. Thus, we perform a meta-analysis to ascertain the diagnostic value of NT-proBNP in detecting CAL of KD in the acute phase. PubMed, the Cochrane Central Register of Controlled Trials, EMBASE, and China National Knowledge Infrastructure were searched to detect relevant publications. Finally, eight eligible studies were included. The overall diagnostic sensitivity and specificity were 0.84 (95% confidence interval [CI]: 0.78–0.89) and 0.71 (95% CI: 0.68–0.75), respectively. The area under the summary receiver operating characteristic curves value (SROC) curve was 0.8582 ± 0.0531. Moreover, the overall sensitivity and specificity across five studies adopted the threshold of approximately 900 ng/L were 0.82 (95% CI: 0.73–0.89) and 0.72 (95% CI: 0.68–0.76), respectively. SROC was 0.8868 ± 0.0486. This meta-analysis would be the first one to describe the role of NT-proBNP in detecting CAL of KD. We register this study with PROSPERO (CRD42019130083).

Publication bias. We used funnel plots and the Deeks' test to assess publication bias in the included studies. Each dot plots in these graphs represented a study. The distance between each dot and the vertical line indicated bias in each study. Symmetric distribution indicated no publication bias existed. Funnel plots in Fig. 6A-F present a degree of symmetry, suggesting that there is no potential for publication bias among the included articles.

Discussion
The diagnostic accuracy of NT-proBNP in detecting KD with CAL in the acute phase was systematically evaluated in our study. Finally, we found that NT-proBNP can be a valuable biomarker for predicting CAL of KD. The previous meta-analysis 10 showed that NT-proBNP level was mildly higher than in KD patients of the acute phase compared with the febrile control patients. In our study, we found that the mean NT-proBNP level in KD patients with CAL was much higher than in KD patients without CAL (approximately 2500 ng/L vs. 800 ng/L) (Table 1). However, the precise mechanism of elevated NT-proBNP in KD patients is still unclear 21 . Nevertheless, factors known to affect the NT-proBNP levels, such as cardiac function 22 or inflammatory cytokines 23 , were not evaluated in all the included studies, which could cause biased results. Therefore, further well-designed studies are needed to evaluate the value of NT-proBNP in predicting CAL of KD.
The present study suggested that the overall diagnostic sensitivity and specificity of NT-proBNP for diagnosis KD with CAL in the acute phase were 0.84 and 0.71, AUC of SROC was 0.8582. Then, we evaluated five studies, which all had thresholds at approximately 900 ng/L. Finally, we found the overall diagnostic sensitivity and specificity of NT-proBNP (threshold ≈ 900 ng/L) for diagnosis KD with CAL were 0.82 and 0.72, AUC of SROC was 0.8868, which was slightly higher than the overall diagnostic accuracy of NT-proBNP. In general, those found suggested that NT-proBNP can be used as a biomarker for detecting CAL of KD. Additionally, the diagnostic accuracy in the threshold of about 900 ng/L shows a little higher than the overall diagnostic accuracy, which still only indicates that NT-proBNP has diagnostic value, but cannot indicate that 900 ng/L is the recommended threshold. The specific threshold range needs to be determined by more well-designed clinical studies with larger sample size. Besides, the meta-regression showed that the differences among the study countries were the source of heterogeneity.
Furthermore, we conducted two subgroups analysis by the study design and the total sample size, and these results suggested that both the type of research and the total sample size might be the sources of heterogeneity. Besides, the calculated AUC of value was 0.9065 ± 0.0577 for the prospective group and 0.8976 ± 0.0522 for the total sample size (n > 100) group, which showed prospective studies with large sample size (n > 100) were superior designed studies to evaluate the value of NT-proBNP in predicting CAL of KD.   www.nature.com/scientificreports www.nature.com/scientificreports/ Although CAL in KD may be transient or permanent 24 , all the included studies did not provide follow-up time for CAL, which may bias the diagnostic value of NT-proBNP. Further rigorous studies, with unified inclusion and exclusion criteria and a consecutive enrolment design, needed to evaluate the diagnostic value of NT-proBNP in KD with CAL diagnosis.
Our meta-analysis has several limitations. First, the number of included studies was small (n = 8), and all of them were conducted in Asian populations, which means these results may not generalize to other areas. Second, no included articles combine NT-proBNP with other laboratory tests, such as ESR, CRP, to identify the diagnostic accuracy of KD with CAL, which could work as a better method for detection. Third, all of the follow-up time of the included articles were unclear, which may lead to the deviation of CAL diagnosis, which could further affect the accuracy of NT-proBNP in predicting CAL of KD.
In conclusion, despite these limitations, this is the first meta-analysis that showed NT-proBNP could be used as a biomarker for detecting CAL of KD. Besides, further well-designed studies with a large sample size are needed to strictly evaluate the value of NT-proBNP in predicting CAL of KD.

Materials and Methods
Study protocol and ethics statement. We performed this analysis following a predetermined protocol according to the recommendations of Deeks 25 . The data collection and reporting were by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement 26 (S Table 1). Due to it is a systematic literature study, ethical approval was not necessary. The protocol for this analysis was registered with PROSPERO (CRD42019130083). Study selection. Articles were screened preliminarily by title and abstract after citations selected by the systematic search. And then, potentially relevant articles were retrieved by full text, while assessed for compliance to inclusion and exclusion criteria.
Inclusion criteria: (1) all cases must meet the KD diagnostic criteria; (2) randomized or non-randomized controlled, cohort studies, clinical trials evaluating NT-proBNP in blood samples; (3) contained the data that can calculate true positive (TP), false negative (FN), false positive (FP), and true negative (TN), such as specificity, sensitivity and sample size; (4) all studies had KD with non-CAL subjects as the control group; (5) The samples were taken from patients with acute KD before initial IVIG treatment. www.nature.com/scientificreports www.nature.com/scientificreports/ Exclusion criteria: (1) reviews, editorials, abstracts, letters, expert opinions, conferences articles, or case reports without controls; (2) unable to construct 2 × 2 table; (3) duplicated publications.
Data collection and assessment of study quality. The eligibility of studies was assessed by two investigators (Xiaolan Zheng, Yi Zhang) independently by the title and abstract. At the same time, the divergences and the quality of reports were determined by a third reviewer (Yifei Li) according to inclusion or exclusion criteria. The quality assessment of all included reports was evaluated by the two investigators (Xiaolan Zheng, Lei Liu) independently following the QUADAS list 27 . As well-conducted research might score lower in the absence of relevant parts of the methodology and results, the assessments were reported in descriptive form only. Finally, the data from which can calculate TP, FP, FN, and TN were extracted by two investigators (Xiaolan Zheng, Peng Yue). evaluation indicators. We measured the following indicators of NT-proBNP: sensitivity, specificity, DOR, and SROC. Sensitivity represented the proportion of patients in KD patients with CAL, which were correctly identified by the positive results of NT-proBNP. Besides, specificity expressed the KD cases with non-CAL that were correctly identified by the negative results of NT-proBNP. Also, the DOR more reliably defined a summary of test performance, rather than merely pooling specificity and sensitivity in individual reports. DOR was an independent index with a range of 0 ~ infinity. The higher the DOR, the better the discrimination 28 . The SROC was plotted by combining sensitivity and specificity. Furthermore, AUC was calculated as a global measurement of test performance 29 , and the closer the AUC was to 1, the better the test performance would be. publication bias. Funnel plots and the Deeks' test were used to assess the publication bias. It indicated a potential publication bias when the asymmetric distribution of data dot in the funnel plot with a quantified result of P < 0.05 30 . Heterogeneity and meta-regression. The heterogeneity of pooling sensitivity and specificity were examined by the x 2 test, while the heterogeneity of pooling DOR was examined by the Cochran Q test. The I 2 test in every pooling analysis to quantitatively was also conducted to assess the proportion of total variation in the study. I 2 value would range from 0 to 100%, with values of 25, 50, and 75%, respectively, as evidence of low, moderate, and high heterogeneity 31 . The threshold effect was suggested by a curvilinear shape in the SROCs. Furthermore, the meta-regression was carried out to detect the potential factors that would cause heterogeneity. All the possible factors were extracted from the baseline measurement and original testing procedures, which were included in meta-regression. The meta-regression can determine the correlation between the potential factors and the existing heterogeneity. The factor should have a dramatic impact on the homogeneity of the enrolled studies with a P-value < 0.05 when a significant difference was discovered. Sensitivity analysis. We performed the sensitivity analysis to determine the influence of individual studies on the results. Meta-Disc 1.4 32 was used for detecting threshold effects in reports.
Statistical analysis. Meta-Disc 1.4 was utilized to perform data analysis. Besides, STATA 15.1 (Stata Corporation, College Station, Texas, USA) was utilized to assess the publication bias and perform meta-regression analysis. Homogenous results utilized the random-effects model for statistical analysis, while the heterogeneous (I 2 < 50%) results utilized a fixed-effects model, and the data were presented using a forest map.

Data availability
The authors confirm that all the data based findings are fully available without restriction. All relevant data are included in the paper and references.