Diagnostic and prognostic value of circular RNA CDR1as/ciRS‐7 for solid tumours: A systematic review and meta‐analysis

Abstract The circular RNA, CDR1as/ciRS‐7, functions as a vital regulator in various cancers; however, the predictive value of CDR1as remains controversial. Therefore, a comprehensive analysis for clarifying the precise diagnostic and prognostic value of CDR1as in solid tumours is needed. A literature review of several databases was conducted for identifying potential studies. Pooled odds ratios (ORs) and hazard ratios (HRs) were used for evaluating the diagnostic accuracy variables and survival. Overall, 15 studies (1787 patients) and 11 studies (1578 patients) were included for diagnostic and prognostic outcome syntheses, respectively. Up‐regulated CDR1as expression was found to be correlated with worse clinicopathological characteristics, including the T status, N status, histological grade, TNM stage and distant metastasis. The synthesized sensitivity was 0.72 (95% confidence interval [CI], 0.65‐0.79), and the specificity was 0.80 (95% CI, 0.74‐0.86). The positive likelihood ratio (LR), negative LR and diagnostic odds ratio (DOR) were 3.70, 0.34 and 10.80, respectively. The area under the receiver operator characteristic curve was 0.84 (95% CI, 0.80‐0.87). In the pooled prognostic analysis, patients with high CDR1as expression had worse overall survival (HR = 2.40, P < 0.001) and disease‐free survival (HR = 1.74, P < 0.001). These results suggest that CDR1as is a reliable diagnostic and prognostic biomarker with high accuracy and efficiency, which may potentially facilitate clinical decisions on solid tumours in the future.

and widely used in clinical practice; however, they are inefficient with low accuracy. Therefore, it is imperative to establish a diagnostic biomarker with higher accuracy in the early stage of cancer and develop more robust therapeutic methods.
Recently, as a novel endogenous non-coding RNA, circular RNA (circRNA) has attracted the attention of many researchers and has rapidly become a heated topic in the field of biomedicine. 3 Derived from the back-splice of exons and/or introns of messenger RNAs, circRNAs are abundant and stable in mammalian tissues. 4 With no 5′ cap and 3′-poly-A tail, circRNAs are once considered useless by-products of incorrect splicing in cells. 5 With the progress made in high-throughput RNA sequencing and bioinformatic analysis, an increasing number of circRNAs have been captured and identified. 6,7 Scientists have discovered that circRNAs are versatile regulators of the process of multiple diseases, including diabetes mellitus, Alzheimer's disease, autism, heart failure and cancer. [8][9][10][11][12] circRNAs serve as mediators of intracellular biological activities by different mechanisms, including binding proteins, sponging miR-NAs and encoding short peptides. 13 For example, circFBXW7 is down-regulated in malignant tissues, which inhibits cell proliferation and invasion by encoding a 21kDa novel short peptide FBXW7-185aa and sponging miR-197-3p in glioma and triple-negative breast cancer. 14,15 A circRNA derived from CTNNB1 exons facilities cell proliferation and metastasis by encoding a novel 370-amino acid CTNNB1 isoform and activating the Wnt signalling pathway in hepatocellular carcinoma. 16 CircRAD18, circRNA FLI1 and circPLK1 were also identified as oncogenic drivers of breast cancer through different mechanisms, including reduction of apoptosis, maintenance of DNA methylation and activation of autophagy. [17][18][19] Originating from CDR1AS (cerebellar degeneration-related protein 1 antisense transcript), CDR1as is the most well-known circRNAs that has been studied in many disease models. CDR1as contains >70 conventional binding sites of miR-7 and serves as a negative regulator of miR-7 for altering the expression of multiple key target genes. 20 Because of its unique characteristic, CDR1as is also termed as ciRS-7, the circRNA sponge for miR-7. CDR1as has been widely studied and shown to be an oncogenic factor in various cancers. For example, CDR1as sponges miR-7 to facilitate colorectal cancer growth and invasion by regulating EGFR signalling pathway activity. 21 By inhibiting miR-876-5p, CDR1as promotes the growth and metastatic ability of oesophageal squamous cell carcinoma and up-regulates the expression of the MAGE-A family. 22 Additionally, CDR1as has the potential to regulate the tumour environment, which is negatively correlated with immune cell infiltration and immune response. 23 Thus, CDR1as has great potential as an accurate diagnostic biomarker and novel therapeutic target for multiple cancers.
In the current study, we conducted a systematic review and meta-analysis of reported studies for evaluating the diagnostic and prognostic value of CDR1as in solid tumours. Several indexes were analysed to assess whether CDR1as can be used as an ideal biomarker for diagnostic and prognostic prediction in solid tumours.

| Search strategy
A comprehensive search of PubMed, Embase, Cochrane Library and Web of Science online databases was performed. The website addresses of each database are presented in Table S5. The period of literature retrieval was from 1 January 1990 to 30 September 2019. The following keywords were used in the retrieval strategy: "CDR1as" or "ciRS-7" or "hsa_circ_0001946." All search strategies were conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. 24

| Inclusion and exclusion criteria
Studies concerning the prognostic or diagnostic value of CDR1as/ ciRS-7 expression in patients with cancer were eligible for quantitative synthesis. Only articles published in the English and Chinese languages were included in this study. The exclusion criteria were review, case reports, letters, and commentaries and studies with insufficient or ambiguous data or not relevant to ciRS-7 study in cancer.
An insufficient data study was defined as a study with neither diagnostic accuracy data nor prognostic outcomes of ciRS-7. An ambiguous data study was defined as a study with wrong statistics. Studies of other biomarkers (lncRNAs, other circRNAs, etc) or non-cancer diseases were also excluded in this research, which were defined as not relevant to ciRS-7 study in cancer.

| Study selection
All search results were independently examined by two authors (YTZ and XPD) with discrepancies consulted by a third reviewer (SQZ). The selection criteria were applied by reviewers after screening the potentially included studies. Duplicates were removed using Endnote X9 software or manually.

| Data extraction
The baseline characteristics of each study (authors, year of publication, cancer type, sample size, specimen sources, detective method and follow-up) were recorded independently by two reviewers. The correlation between CDR1as expression and clinicopathological characteristics, including age, sex, T status, N status, grade, distant metastasis and TNM stage, was also evaluated. Odds ratio (OR) and 95% confidence interval (CI) were used to describe the abovemen-

| Methodology quality assessment
Methodology quality was assessed using the Newcastle-Ottawa Scale (NOS) for prognostic cohort studies and the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) for diagnostic studies.
As for NOS, studies that scored 0-6 were regarded as low-quality research, while those that scored 7-9 were defined as high-quality evidence. A study with at least three unclear or high risk of bias was considered to have low quality, assessed using QUADAS-2.

| Data synthesis and analysis
HRs and ORs extracted from studies were synthesized using the random-effects model in Review Manager software (version 5.3).
Estimation of heterogeneity was performed using Cochran's Q test, which reported a P-value and I 2 statistic. A P-value < 0.1 or I 2 statistic > 50% indicated heterogeneity. In addition, publication bias was evaluated by inspecting the funnel plots and using Begg's test.

F I G U R E 1 PRISMA flow diagram of the article retrieval strategy in this meta-analysis
Pooled sensitivity, specificity, positive LR, negative LR, DOR and AUC were calculated using the Stata software (version 15.1). Deeks' funnel plot test and bivariate boxplot were employed to assess publication bias. Subgroup analysis was conducted to determine the potential sources of heterogeneity. Fagan's nomogram was used to predict post-test probabilities.  37 and cervical squamous cell carcinoma (CESC) in one study. 38 Additionally, real-time polymerase chain reaction (qRT-PCR) analysis was used to detect CDR1as expression. Detailed baseline information of the included studies is presented in Tables S1 and S2. The specific mechanisms and egulated pathways of CDR1as in each study are summarized in Table S3. Quality assessment results are shown in Table S4 using the Newcastle-Ottawa Scale for cohort prognostic studies and in Figures S1 and S2 using QUADAS-2 for diagnostic studies.

| Clinicopathological parameters of CDR1as in cancers
In the pooled analyses of clinicopathological characteristics of CDR1as in solid tumours, a significant association between CDR1as expression and clinicopathological parameters was revealed in our results (

| Diagnostic significance of CDR1as in cancers
The  (Table 2). Additionally, we found a higher diagnostic accuracy of CDR1as in plasma (AUC = 0.89) compared to tumour tissue (AUC = 0.66) in patients with ESCC despite a small sample size (Table 2). Deeks' funnel plot asymmetry test revealed no publication bias (P = 0.10), and the bivariate boxplot manifested heterogeneous statistics ( Figure 2G-H). A scatter plot of positive and negative LRs with combined summary points is shown in Figure 3A. Fagan's nomogram was constructed to calculate post-test probabilities of CDR1as, from which we found that the post-test probability increased to 48% with a positive LR of 4, and the post-test probability decreased to 8% with a negative LR of 0.34 ( Figure 3B). These results indicate that CDR1as is a reliable diagnostic biomarker with high accuracy and efficiency.  in all cancers included in this study and a prognostic factor for DFS in certain types of cancer ( Figure 4C-D). Moreover, subgroup analyses of the two groups based on sample size, univariate/multivariate analysis, follow-up duration, quality of evidence and publication year revealed a decrease in heterogeneity in each group (Table 3). Funnel plots and Begg's test (P = 0.651) manifested no potential publication bias in the pooled analysis ( Figure S3).

| D ISCUSS I ON
In this study, we conducted a systematic review and meta-analysis  when applying a threshold of 5 µg/L, as reported in a meta-analysis of 23 studies. 43 Another tumour marker, cancer antigen 125, is not able to reduce the mortality rate in screening for ovarian cancer according to four high-quality clinical trials. 44 Combined with ultrasound measurement in detecting early stage hepatocellular carcinoma, AFP is still inefficient with extremely low sensitivity (63%), which leads to a high misdiagnosis rate. 45 As newly discovered non-coding RNAs, an increasing number of circRNAs have been identified as promising biomarkers for cancer. 46,47 Compared to other biomarkers (protein, ln-cRNAs), circRNAs are more ubiquitous and stable with a closed-loop structure in mammalian tissues. 5 The predictive value of CDR1as as a biomarker for solid tumours was preliminarily explored in this study.
Compared to some existing biomarkers, CDR1as is universally and specifically expressed in cancer tissues, which lays a solid foundation for its future development as a biomarker.
Several limitations exist in this study. First, because of the lack of prospective and double-blind studies in diagnostic value investigation, bias was inevitable to some extent. Second, the small number of samples in each study contributed to the high heterogeneity in the analysis of diagnostic variables. Third, most samples are validated in Chinese cohorts, which results in population bias. Therefore, more multicentre and prospective diagnostic studies with high quality, large sample size and strict operation are required for further verification.
In conclusion, our study indicates that circRNA CDR1as has a remarkable association between its abnormal expression and clinicopathological, diagnostic and prognostic roles in patients with solid tumours. CDR1as is a promising diagnostic and predictive biomarker with high accuracy and efficiency which has potential to be an ideal indicator for solid tumours in the future.

CO N FLI C T O F I NTE R E S T
The authors declare no potential competing interests. Writing-review & editing (equal).

E TH I C A L A PPROVA L
This article does not contain any studies with human participants performed by any of the authors.