Results
Literature screening and characteristics of eligible studies. A total of 233 articles were identified using the search strategies, including 113 articles from PubMed, 5 articles from Cochrane library, 115 articles from Web of Science. Totally, 184 articles were excluded after careful filtration. Among them, 148 articles were duplicates and 36 had inappropriate abstracts and titles. Then the left 49 full-text articles were assessed for their eligibility, which contains 8 articles without diagnostic method, 13 articles without exact data on recurrence rates, 11 articles without using F. nucleatum , B. fragilisor E. coli as the single biomarker and 6 articles with other chemotherapeutics included. Finally, 11 articles were included in this meta‑analysis, which involved 443 participants (200 CRC patients with high F. nucleatum abundance and 243 CRC patients with lowF. nucleatum abundance, 144 CRC patients with high B. fragilis abundance and 121 CRC patients with low B. fragilisabundance, 400 CRC patients with high E. coli abundance and 26 CRC patients with low E. coli abundance). The selection process is illustrated in Figure 1.
Basic characteristics and diagnostic performance of F. nucleatum ,B. fragilis or E. coli in these 11 studies are shown in Table 1. Fecal gut microbiota was evaluated by RNA in situhybridization (RNA-ISH), quantitative real-time PCR (RT-PCR), droplet digital PCR or multiplex PCR. All studies included in the meta-analysis used the histopathological examination as the gold standard. Figure 2 shows an overview of the methodological quality results. In general, the overall quality of the eligible studies was high.
Heterogeneity. It is known that the threshold effect, one source of heterogeneity, can be determined by calculating the Spearman correlation coefficient between sensitivity and specificity for all included studies. However, in our meta-analysis, Spearman’s rank correlation coefficient was -0.94, calculated by an equation using logarithm of sensitivity and 1-specificity. SROC distributed advisably (AUC: 0.79, 95% CI: 0.74-0.81; Figure 3), which indicated no statistical significance (P = 0.83). Then we performed the meta-regression based on the variables, including ethnicity, median age, detection methods, CRC patients’ number, sample size of 5-FU resistance to explain this heterogeneity (Table 1). Among the five factors, sample size of 5-FU resistance was identified as statistically significant (P = 0.012), indicating that sample size of CRC patients with 5-FU resistance was responsible for the relatively high heterogeneity.
Diagnostic performance in meta-regression analysis. As forest plot revealed significant heterogeneity between studies, single‑factor meta‑regression analysis was applied to screen the potential variables impacting on the pooled data. The performance of F. nucleatum for 5-FU resistance of CRC is shown in forest plot (Figure 4): pooled sensitivity: 0.65 (95% CI:0.60-0.69), specificity: 0.70 (95% CI:0.59-0.87), PLR: 2.57 (95% CI:1.47-3.21), NLR: 0.52 (95% CI:0.43-0.63) and DOR: 4.92 (95% CI:2.23-7.33). In contrast, the performance of B. fragilis for 5-FU resistance of CRC is lower than F. nucleatum (Figure 4): pooled sensitivity: 0.51 (95% CI:0.42-0.54), specificity: 0.36 (95% CI:0.21-0.53), PLR: 0.82 (95% CI:0.79-0.95), NLR: 1.55 (95% CI:1.01-1.62) and DOR: 0.53 (95% CI:0.31-0.82). The performance of E. coli for 5-FU resistance of CRC manifested remarkably higher sensitivity but poor specificity (Figure 4): pooled sensitivity: 0.93 (95% CI:0.90-0.95), specificity: 0.06 (95% CI:0.04-0.92), PLR: 0.99 (95% CI:0.94-1.05), NLR: 1.57 (95% CI:0.87-1.76) and DOR: 0.63 (95% CI:0.57-0.76).