Methods
Search strategy. Four electronic databases were systematically searched from July 2005 to March 2021: PubMed, Embase, Cochrane Library and Web of Science. The following keywords were used in the literature search: “5-Fluorouracil (5-FU)”, “5-Fluorouracil (5-FU) resistance”, “Colorectal Cancer(s)”, “colorectal carcinoma(s)”, “colorectal neoplasm(s)”, “colorectal tumor(s)”, “Colorectal cancer(s) recurrence”, “colorectal carcinoma(s) recurrence”, “colorectal neoplasm(s) recurrence”, “colorectal tumor(s) recurrence”, “Fusobacterium nucleatum ”, “Fusobacterium spp.”, “F. nucleatum ”, “Fn ”, “Bacteroides fragilis ”, “Bacteroides spp.”, “B. fragilis ”, “Escherichia coli ”, “Escherichia spp.” and “E. coli ”. Furthermore, the listed referent studies and relevant review articles were also examined.
Inclusion and exclusion criteria . The inclusion criteria were as follows: (1) Correlation of F. nucleatum to 5-FU resistance or CRC recurrence; (2) A clinical study of CRC patients with intact data ofF. nucleatum positive and F. nucleatum negative cases,B. fragilis positive and B. fragilis negative cases as well as E. coli positive and E. coli negative cases; (3) Contains clear recurrence rates data; (4) All patients were treated with standard 5-FU-based regimen; (5) The diagnosis of CRC progression should be based on histology; (6) The detection of F. nucleatum ,B. fragilis and E. coli should be based on quantitative polymerase chain reaction analysis, fluorescence in situhybridization or 16S rRNA sequencing; (7) The samples (feces or tissues) should be stored at -20°C to -80°C soon after collection; (8) The articles should be original articles.
The exclusion criteria were as follows: (1) Exact data on recurrence rates are not given; (2) Studies that did not include the necessary data for calculating true‑positive (TP), false‑positive (FP), true‑negative (TN) and false‑negative (FN) of F. nucleatum , B. fragilisor E. coli in 5-FU resistance of CRC; (3) Use of other treatments or no chemotherapy; (4) Letters, reviews, conference abstracts and duplicate publications.
Data extraction and assessment. Based on the selection and inclusion criteria, two authors independently screened the title, abstract and full text of the retrieved studies. The third author excluded the irrelevant studies and cross-checked the data. The following information from each article was extracted: the first author’s name, the year of publication, sample size, sample type,F. nucleatum , B. fragilis and E. coli positive and negative groups, and recurrence of CRC patients after 5-FU treatment. The parameters in results were summarized by bivariate mixed‑effects models. The pooled TPs, FPs, TNs and FNs, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and their 95% confidence interval (CI) were extracted for mapping forest plot. The data of the included studies were extracted into a spreadsheet and copied into Review Manager 5.4.
According to the Quality Assessment of Diagnostic Accuracy Studies (QUADAS), which is recommended by the Cochrane Collaboration. The quality of each study was assessed by QUADAS tool which includes 14 items covering patient spectrum: reference standard, disease progression bias, verification bias, review bias, clinical review bias, incorporation bias, test execution, study withdrawals and indeterminate results. Each of the 14 items in the QUADAS checklist is scored as “yes”, “no” or “unclear”. QUADAS tool provides more transparent rating of bias and applicability of primary diagnostic accuracy studies. If the QUADAS score is less than 10 points, the study is identified as low methodological quality [35].
Statistical analysis. All analyses were conducted by the Review Manager 5.4 software. The available data were analyzed in the meta-analysis, and the outcomes are presented as forest plots and funnel plot. To calculate the combined OR and its 95% confidence interval (CI ), heterogeneity was assessed using P -values in the pooled analyses which represents the percentage of total variation across the studies. If the P value was less than 0.01, then the summary estimate was analyzed in a random-effects model (DerSimonian-Laird method). Otherwise, a fixed-effects model (Mantel-Haenszel method) was applied. In addition, publication bias was detected by visual examination of the funnel plot symmetry, with asymmetry suggesting possible publication bias. It was also assessed by the Begg and Egger test in the meta-analysis. If the P value was less than 0.05, the study is classified as publication bias and the meta‑trim method would be conducted.
Table1. Characteristics and summary results of included studies