3.1 Search and description of studies
The flow diagram for literature research processes was shown in Figure 1. All the diagnostic values of non-coding RNAs in BC patients were collected and evaluated. 2392 BC blood specimens and 1738 healthy specimens’ blood individuals from 32 studies (24 articles) published from January 2012 to November 2019 were contained in this meta-analysis (Table 1 and Supplement Table 2).
Among the 32 studies, 28 studies explore the association with 50 miRNA expression, 1 study investigated 1 circRNA, and the other 3 studies focused on 3 lncRNA. In terms of samples, blood specimens from whole blood, plasma, and serum and the sample size from 10 to 177. All studies used quantitative real time polymerase chain reaction (qPCR), excepted one study used absolute RT-qPCR. Histopathological grading information was provided in all studies. Seventeen studies were shown 29 miRNA up-regulated in breast cancers, and six studies were reported 7 miRNA down-regulated expression, and four studies were shown 1 lncRNA down-regulated, 2 lncRNA up-regulated, and 1 circRNA up-regulated in breast cancer (Table 2). The endogenous reference, quantitative method and cut-off value are not uniform in all studies. Three studies did not mention 14 miRNA expression changes.
The difference expression between ncRNAs and clinic-pathological factors were shown in Table 3, including tumor size, TNM stage, lymphatic metastasis, tumor invasion, ER/PgR, PR, HER2/c-erbB-2, and Ki67.
In addition, all the studies independently scored the included studies based on Quality assessment of diagnostic accuracy studies-2 (QUADAS-2) score system (Review Manager), including patient selection, index tests, reference standard, and flow and timing . Each domain is assessed in terms of the risk of bias, which should answer as “yes”, “no”, or “unclear”, and phrase such that “yes” indicates low risk of bias. And the first 3 domains are also assessed in terms of concerns about applicability, which should rate as “low”, “high”, or “unclear”, and the “unclear” category should be used only when insufficient data are reported.
Detailed results of the QUADAS-2 assessment are provided in supplement Figure 1. All studies showed risk of bias, because thresholds for index test positivity had been predefined, and all the patients had the same reference standard. Thirty-two documents included in the gold standard, but did not mention whether or not to use the blind method. All of them had relatively high quality in supplement Figure 1, indicating the relatively the reliable foundation of our analysis.
Thirty-two studies involving 3 types of RNA with 2392 BC patients investigated the diagnostic value of non-coding RNAs as the biomarkers of cancer, and a meta-analysis of the SEN, SPE, PLR, NLR, DOR, and AUC for non-coding RNAs were plotted in the breast cancer diagnosis. The random effect model was chosen to analyze the inconsistency (I2=100 > 50). The pooled SEN was 0.82(95%CI:0.76-0.86) (Figure 2A), SPE was 0.83(95%CI:0.75-0.88) (Figure 2B), PLR was 4.67(95% CI:3.19-6.84) (Figure 2C), NLR was 0.22(95%CI:0.17-0.29) (Figure 2D), DOR was 20.91(95%CI:12.01-36.40) (Figure 2E) and AUC values was 0.89 (95%CI:0.86-0.91) (Figure 2F). Then, the funnel plots implied that publication bias has no effect on the result (p=0.45) (Figure 2G). These results indicated a relatively moderate diagnostic accuracy of non-coding RNA in detection of cancer patients’ blood.
3.3 Meta-regression analyses
Using meta-regression to find the potential sources of heterogeneity, including TNM stage, lymphatic metastasis (tumor invasion), ER/PgR, PR and HER2/c-erbB-2. Meta-regression analysis was shown in Table 4, the result suggested that TNM stage, was the major cause of heterogeneity. There are only 15 studies reported association expression of non-coding RNA with clinicopathological features. The significant association for TNM stage, (p=0.0321) but not ER/PgR (p=0.3257), PR (p=0.2519), HER2/c-erbB-2 (p=0.5598) and lymphatic metastasis (tumor invasion) (p=0.1401) were shown. And p-value of TNM stage was less than 0.05, that means the TNM stage is the reason for the formation of heterogeneity.
We also show the construction of a bivariate boxplot which is useful tool for the detection of heterogeneity for each study (Figure 2H). There are 4 studies not located in the boxplot, including four studies (Eichelser C2, miR-93; Mishra S2, miR495; Zaleski M, miR34a; Swellam M2, miR222). But 3 of 4 studies (Eichelser C2, miR-93; Mishra S2, miR495; Zaleski M, miR34a) did not associate with TNM stage. That means TNM stage is the major causes of heterogeneity. We further excluded these 4 studies, found by influence analysis and detection in supplement Figure 2. After exclusion, the level of SEN decreased from 0.82 to 0.79, the NLR increased from 0.22 to 0.25, the DOR decreased from 21 to 19, and AUC increased from 0.89 to 0.87, and SPE and PLR not changed, showing minimal change with our overall analysis. The results confirmed that these 4 studies were not the cause for heterogeneity.
Furthermore, we used meta-regression (Meta-Disc) to analysis the source of samples which divided into blood (8 studies), plasma (13 studies), and serum (11 studies) three subgroups. The p-value of these three groups (p=0.4670 in blood group; p=0.5495 in plasma group; p=0.3517 in serum group) were more than 0.05, that means the sample source was not the main reason for the formation of heterogeneity (supplement table 3-5). We also test meta-analysis of the SEN, SPE, and AUC for these three groups (blood, plasma, and serum), the pooled SEN values were 0.84, 0.82, and 0.79, the pooled SPE values were 0.84, 0.84, and 0.80, and AUC values were 0.91, 0.90, and 0.85, respectively (supplement Figure 3). These results shown a little change, and confirmed that the samples source was not the reason for heterogeneity.
Therefore, we performed subgroup analysis for the expression of non-coding RNAs in different clinicopathological features. Eight studies showed significant differences among TNM stage. We conducted subgroup analysis on these studies. In addition, there was no significant effect in subtype of BC and lymphatic metastasis meta-regression, 5-6 studies showed significant difference in non-coding RNA expression in breast tumors and healthy people, and we also conducted subgroup analysis on these studies.
3.4 Subgroup analysis
Among all the studies, 8 studies show significant differences, and other 10 studies show no significant differences, the rest of 14 studies have no data. These eight studies including 814 patients evaluated the expression of non-coding RNAs as diagnostic biomarkers for BC. The meta-analysis of the SEN, SPE, PLR, NLR, DOR, and AUC for non-coding RNA association with TNM stage. The random effect model was used to analyze (I2>50%). The pooled SEN (Figure 3A), SPE (Figure 3B), PLR (Figure 3C), NLR (Figure 3D), DOR (Figure 3E), and AUC values (Figure 3F) were 0.87 (95%CI: 0.78–0.93), 0.84 (95%CI: 0.72–0.91), 5.5 (95%CI: 3.0–10.1), 0.15 (95%CI: 0.08–0.27), 37.00 (95%CI: 13–101), and 0.92 (95%CI:0.90–0.94), respectively. The funnel plot suggested that publication bias had no significant effect on diagnostic assessment (p=0.29) (Figure 3G).
3.4.2 Metastasis or invasion
Seven studies involving of 520 patients investigated the diagnostic values of non-coding RNAs as the biomarker association with metastasis or invasion, and the meta-analysis of the SEN, SPE, DOR, and AUC were plotted, using random effect model (I2>50%). The pooled SEN (Figure 4A), SPE (Figure 4B), PLR (Figure 4C), NLR (Figure 4D), DOR (Figure 4E), and AUC values (Figure 4F) were 0.78 (95%CI: 0.68–0.86), 0.85 (95%CI:0.77–0.91), 5.3(95%CI:3.5–8.0), 0.25 (95%CI: 0.17–0.38), 21 (95%CI: 12–37), and 0.89 (95%CI:0.86–0.92), respectively. In addition, the funnel plot suggested that the publication bias had no significant impact on assessment (p=0.07) (Figure 4G).
3.4.3 Subtypes of BC
Six studies including 581 patients evaluated the diagnostic value of non-coding RNA as a biomarker association with subtype of BC, including ER/PgR, PR. HER2/c-erbB-2. Using random effect models, the pooled SEN (Figure 5A), SPE (Figure 5B), PLR (Figure 5C), NLR (Figure 5D), DOR (Figure 5E), and AUC values (Figure 5F) were 0.89 (95%CI: 0.72–0.96), 0.91 (95%CI: 0.77–0.97), 10.1 (95%CI: 3.5–29.2), 0.13 (95%CI: 0.05–0.34), 80 (95%CI: 15–434), and 0.96 (95%CI: 0.94-0.97), respectively. The funnel plots indicated that the publication bias probably might have no effect on evaluation (p=0.82) (Figure 5G). Among these three subgroups, subtype of BC group shows high level of SEN, SPE and AUC value.