The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement was used as a guideline for this study.

We conducted searches in multiple databases, including the Cochrane Library, PubMed, MEDLINE, EMbase, CNKI, WanFang, the Chinese Biomedical Database, and CQVIP, up until November 2023. The search terms used were combinations of the following keywords: children, pediatric; kawasaki disease; C-reactive protein (CRP); albumin (ALB); and C-reactive protein to albumin ratio (CAR). Both subject and free words were used in the search, and the strategies were finalized after conducting several pre-searches. Additionally, we reviewed the references of the included articles and other relevant studies. There were no language restrictions applied.

According to the Cochrane Collaboration Network’s systematic evaluation manual, the inclusion and exclusion criteria for this meta-analysis were formulated in strict accordance with the participants, interventions, comparators, outcomes, and study design (PICOS) principles.The inclusion criteria were as follows: (1) The research objectives involved conducting a cohort study to evaluate the early diagnostic value of CAR in IVIG-resistant KD; (2) The study subjects were KD patients who met the relevant diagnostic criteria; (3) Four-fold table data could be directly obtained from the literature or calculated; (4) The sources of reagents and detection methods were clearly defined, and the detected specimens were serum.

The exclusion criteria were as follows: (1) Descriptive studies and case-control studies; (2) Data that was not completely provided or was difficult to extract; (3) Repeated reports (inclusion or exclusion depending on the quality of the literature), reviews, conference papers, or only abstracts where the full text was not available.

For each study, two authors (Jue Liu & Xingguang Chen) independently extracted data and assessed quality. Any disagreements were resolved by the corresponding author (Yiqun Teng & Jian Jin). A standardized data extraction form was used, including:1) Baseline characteristics of the literature: publication date, country or region, number of cases, age, study type, time span, diagnostic criteria, the cut-off value of CAR; (2) Information on diagnostic test parameters: True positive (TP), false positive (FP), false negative (FN), and true negative (TN).

The included studies underwent quality assessment using the Quality Assessment of Diagnostic Accuracy Studies 2.0 (QUADAS-2.0) [15]. Each item in the included studies was evaluated as “yes”, “no”, or “unclear”. A score of 1 was assigned for “yes,” and 0 for “no” or “unclear.” The literature was considered of higher quality if its score exceeded 7.

Stata 15.0 software (StataCorp LLC, North Carolina, USA) was used for statistical analysis. The heterogeneity of the included studies was assessed using the Q test and I2 index. If no heterogeneity (p > 0.1 or I2 < 50% for Q test) was detected for SEN, SPE, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), a fixed-effects model based on the Mantel-Haenszel method was used for pooling. Additionally, the summary receiver operator characteristic (SROC) curve was plotted, and the area under the curve (AUC) was calculated. Otherwise, if there was heterogeneity, the random-effects model using the DerSimonian-Laird method was applied. Clinical utility assessment was conducted using Fagan’s nomogram. Publication bias was assessed using Deek’s funnel plot. Bivariate boxplots were employed to identify literature with high heterogeneity. Cohen’s Kappa analysis was used for the diagnostic consistency analysis.