Circulating long noncoding RNAs as potential biomarkers for stomach cancer: a systematic review and meta-analysis

Background Recent researches have suggested that long noncoding RNA (lncRNA) is involved in the tumorigenesis and development of stomach cancer (SC). This meta-analysis aimed to identify the diagnostic performance of circulating lncRNAs in SC. Methods All relevant studies were systematically searched through PubMed, Web of Science, Cochrane Library, and EMBASE databases. The diagnostic values of lncRNAs were mainly assessed by pooled sensitivity, specificity, and summary receiver operating characteristic area under the curve (SROC AUC). Meta-DiSc 1.4, Review Manager 5.3, and STATA 12.0 were used for statistical analysis. The protocol for this systematic review was registered on INPLASY (INPLASY202120079) and is available in full on the inplasy.com (10.37766/inplasy2021.2.0079). Results A total of 42 eligible studies were included in this meta-analysis. The pooled sensitivity, specificity, and SROC AUC were 0.78 (95%CI 0.75–0.81), 0.75 (95%CI 0.71–0.78), and 0.83 (95%CI 0.80–0.86), respectively, suggesting that the lncRNAs test had a high accuracy for the diagnosis of SC. Obvious heterogeneity might come from the type of lncRNA through subgroup and meta-regression analysis. Fagan diagram shows the clinical value of lncRNAs test in SC. Conclusions Abnormal expression of circulating lncRNAs exhibits a high efficacy for diagnosing SC, which is promising in clinical application. Supplementary Information The online version contains supplementary material available at 10.1186/s12957-021-02194-6.

In recent years, the regulation of gene expression by noncoding RNAs has been studied thoroughly. Long noncoding RNAs (lncRNAs) are RNA molecules greater than 200 nucleotides that modulate gene expression at the levels of transcription, posttranscription, and translation, but are not able to encode proteins [6]. An increasing body of evidence has suggested that lncRNAs play a major role during the processes of tumorigenesis and development, which may offer new ideas for the early diagnosis of SC. For instance, for distinguishing SC patients from normal subjects, the lncRNAs PCGEM1 and LOC80054 have higher area under the curve (AUC) values than other conventional tumor markers (AFP, CEA, CA12-5, CA19-9, and CA72-4) [7,8]. Similarly, lncRNAs can also be detected in the blood, and circulating noncoding RNAs have become a new source of noninvasive cancer biomarkers [9], which can serve as new diagnostic biomarkers for SC.
However, considering the small sample size and limitations of the research design, there is insufficient evidence to confirm the diagnostic accuracy of circulating lncRNAs in SC patients. To address this shortcoming, a comprehensive systematic review and metaanalysis was conducted to explore the diagnostic accuracy of circulating lncRNAs in SC.

Search strategy
This meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [10]. The PubMed, Web of Science, Cochrane Library, and Embase databases were systematically searched for potentially relevant articles, which were independently screened by two authors (Cao F and Xu J). The reference lists of relevant metaanalyses and reviews were also searched to identify articles that were not included in the initial search. In addition, relevant articles in scientific congresses and conferences were reviewed. The search strategy and Participant, Index test, Comparison, Outcome, and Study (PICOS) design strategy are shown in Table 1. The publication search was updated regularly until July 9, 2020.

Selection criteria
The following inclusion criteria were used: (i) The expression of lncRNAs was determined in plasma or serum by quantitative reverse transcription-polymerase chain reaction or other molecular techniques; OR"Neoplasm, Stomach"OR"Stomach Neoplasm"OR"Neoplasms, Stomach"OR"Stomach Neoplasms"OR"Gastric Neoplasms"OR"Gastric Neoplasm"OR"Neoplasm, Gastric"OR"Neoplasms, Gastric"OR"Cancer of Stomach"OR"Stomach Cancers"OR"Gastric Cancer"OR"Cancer, Gastric"OR"Cancers, Gastric"OR"Gastric Cancers"OR"Stomach Cancer"OR"Cancer, Stomach"OR"Cancers, Stomach"OR"Cancer of the Stomach"OR"Gastric Cancer, Familial Diffuse") Index test #2 (RNA, Long Noncoding[MeSH Terms]OR"RNA, Long Noncoding"OR"Noncoding RNA, Long"OR"lncRNA"OR"Long ncRNA"OR"ncRNA, Long"OR"RNA, Long Non-Translated"OR"Long Non-Translated RNA"OR"Non-Translated RNA, Long"OR"RNA, Long Non Translated"OR"Long Non-Coding RNA"OR"Long Non Coding RNA"OR"Non-Coding RNA, Long"OR"RNA, Long Non-Coding"OR"Long Non-Protein-Coding RNA"OR"Long Non Protein Coding RNA"OR"Non-Protein-Coding RNA, Long"OR"RNA, Long Non-Protein-Coding"OR"Long Noncoding RNA"OR"RNA, Long Untranslated"OR"Long Untranslated RNA"OR"Untranslated RNA, Long"OR"Long ncRNAs"OR"ncRNAs, Long"OR"Long Intergenic Non-Protein Coding RNA"OR"Long Intergenic Non Protein Coding RNA"OR"LincRNAs"OR"LINC RNA")

Comparison None
(ii) Studies evaluated the diagnosis value of lncRNA for SC; (iii)Sufficient data to determine false negatives, true negatives, false positives, and true positives.
The exclusion criteria were as follows: (i) Duplicate publications; (ii) Meta-analysis, correspondence, single case reports, review articles, and animal model studies.

Data extraction
The two authors (Cao F and Xu J) reviewed the full texts and independently extracted data from all included studies. The following data were extracted: first author, year of publication, race of participants, pathological type of experimental group/control group, sample size, specimen type, lncRNA type, dysregulated state of lncRNAs, sensitivity, and specificity.

Quality assessment
Two authors (Xu J and Cao F) independently evaluated the quality of each diagnostic study. The methodological quality and applicability of the included studies were examined using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) [11] tool in using Review Manager software version 5.3. The QUADAS-2 tool is used to assess the quality of diagnostic accuracy studies [11]. The QUADAS-2 tool contains 4 main areas: process and timing, index testing, reference standards, and patient selection. The risk of prejudice and apprehension was classified as "low," "high," or "unclear." The differences were resolved through discussions among all the researchers. Details of the protocol for this systematic review were registered on INPLASY (INPLASY202120079) and are available in full on the inplasy.com (https://doi.org/10. 37766/inplasy2021.2.0079). This study is presented in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement.
For a meta-analysis of diagnostic accuracy, the sensitivity, specificity, negative likelihood ratio, positive likelihood ratio, diagnostic odds ratio, and the corresponding 95% CIs were used to determine the diagnostic value of lncRNAs. To quantitatively assess the accuracy of diagnosis, the area under the curves (AUCs) of summary receiver operating characteristic curves (SROCs) were determined. The SROC curve method is a meta-analysis of multiple different experiments of a certain detection index. According to the weight of their odds ratio, the diagnostic accuracy is comprehensively evaluated by fitting the SROC curve [13]. The hierarchical summary receiver operating characteristic (HSROC) model proposed by Rutter and Gatsonis in 2001 represents a general framework for the meta-analysis of diagnostic test studies that allows different parameters to be defined as random effects [14]. A HSROC model was adopted to extend the fixed-effects SROC model and evaluate the accuracy of multiple diagnostic tests.
The heterogeneity tests were carried out by the Q test and I 2 statistics. P values of < 0.05 were regarded as statistically significant. An I 2 value > 50% and a P value < 0.05 indicated significant heterogeneity between the included studies, and a random effects model was applied. Otherwise, if there was no obvious heterogeneity, the fixed effects model was applied to evaluate the aggregated results. The heterogeneity induced by the threshold effect was evaluated by the ROC plane. Galbraith Star charts and bivariate boxplots were employed to estimate the degree of heterogeneity. Subgroup analysis and meta-regression were used to assess the source of heterogeneity. Subgroup results were examined one at a time.
Sensitivity analysis was used to determine the stability of the results. Potential publication bias was examined by Deeks' funnel plot. A P value of > 0.1 indicates that there is no publication bias. Fagan's nomogram was applied to judge the clinical value of lncRNAs as a diagnostic method.

Literature searching and study screening
In total, 1867 articles were obtained from the four databases. After eliminating 639 duplicate articles, 1228 studies were further screened. After screening the titles, abstracts, and full texts, 42 eligible studies [2,7, were finally included based on the selection criteria ( Fig. 1).

Quality evaluation and main characteristics of the eligible studies
The diagnostic meta-analysis analyzed 42 eligible studies [2,7, published between 2013 and 2020. Thirty-seven studies detected lncRNA expression in Asian population, while 5 studies detected lncRNA expression in Caucasian populations. Sample types included plasma, serum, and plasma/serum exosomes. All SC patients were pathologically confirmed, and the control groups consisted of healthy donor individuals and benign stomach disease patients. A total of 49 different lncRNAs were examined across all included studies; most of the lncRNAs were upregulated in SC ( Table 2). The quality assessment is shown in Fig. 2.

Heterogeneity analysis
As illustrated in Fig. 3, obvious heterogeneity was found in the pooled sensitivity (I 2 = 88.93%, P < 0.01), specificity (I 2 = 88.14%, P < 0.01), positive likelihood ratio (I 2 = Fig. 2 Quality assessment of eligible studies for diagnostic meta-analysis 88.49%, P < 0.01), negative likelihood ratio (I 2 = 88.71%, P < 0.01), and diagnostic odds ratio (I 2 = 100.00%, P < 0.01). A nontypical shoulder arm appearance was observed in the ROC plane (Fig. 5a). Twenty out of the 63 studies of the Galbraith star chart and 10 out of 42 studies of the bivariate box plot fell outside the 95% CI (Fig. 5b, c). Figure 5d shows the meta-regression forest map. All studies were grouped according to race, pathological types of experimental groups, pathological types of control groups, sample size, specimen type, dysregulated state of lncRNAs, and lncRNA types. Table 3 shows the changes in sensitivity, specificity, and I 2 values after meta-regression and subgroup analysis.

Sensitivity analysis and publication bias
First, sensitivity analysis was carried out to determine the stability of our results. The removal of individual studies exhibited no noticeable changes in pooled results (Additional file 1: Supplementary Figure 1, Fig. S1A). The P value of Deeks' funnel plot asymmetry test was 0.12 (Additional file 1: Supplementary Figure 1, Fig. S1B).

Clinical values of lncRNAs for SC diagnosis
As shown in Fig. 6, Fagan's nomogram revealed that if the patient had a positive lncRNA test result, the actual probability of suffering from SC was 76%, while the probability was 22% if a negative test result was obtained.

Discussion
In recent years, lncRNAs have been recognized as potential diagnostic biomarkers for different cancers [55]. As a diagnostic biomarker for cancer, lncRNAs have the following special advantages. First, the abundance of lncRNAs is relatively high. In the human genome, the number of lncRNAs is four times greater than that of coding RNAs [56]. Second, lncRNAs are highly expressed in the plasma, tissue, and exosomes of cancer cases [57]. Third, lncRNAs have complex biological functions and are closely related to tumorigenesis and development. Therefore, lncRNAs may be promising biomarkers for the early detection and prognosis of various cancers [58].
In the present meta-analysis, a total of 42 eligible studies were screened. The aggregated results of sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio, and SROC AUC indicated that the abnormal expression of circulating lncRNAs exhibits a high accuracy for the diagnosis of SC. The βestimate in the HSROC model indicated that the SROC is symmetrical. Meanwhile, the estimate of lambda reflected the diagnostic accuracy of lncRNAs. Sensitivity analysis verified the stability of the results, and the Deeks funnel chart asymmetry test showed that there was no obvious publication bias. The Fagan diagram also shows its advantages in clinical application, which was mainly due to its moderately high positive and negative predictive value. For the obvious heterogeneity in the pooled estimates, many analyses have been applied to explore the source of heterogeneity. The ROC plane suggests the absence of a threshold effect, while the Galbraith star charts and bivariate boxplots suggest heterogeneity between studies. Meta-regression and subgroup analysis showed that the heterogeneity might come from the type of lncRNA: when lncRNA UCA1 was used as the grouping condition, the I 2 of sensitivity was reduced to 83.60%, and the I 2 of specificity was reduced to 45.20% (P = 0.161). In addition, the diagnostic value of lncRNA UCA1 was above average (AUC: 0.92 (95% CI 0.84-0.99) versus 0.83 (95% CI 0.80-0.86)). There was no evidence that race, pathological types of experimental groups, pathological types of control groups, sample size, specimen type, and dysregulated state of lncRNAs significantly affected the pooled results.
Although meta-analysis of lncRNAs in the diagnosis of SC has been reported before [59,60], most of them focus on lncRNAs in SC tissues. Although lncRNAs in tissue also have high diagnostic accuracy (AUC= 0.755 [59]; 0.80 [60]), their clinical application value is limited for the following reasons: first, the diagnosis of SC after surgery depends on the pathological morphology and immunohistochemical analysis, and the auxiliary role of lncRNAs is optional; second, in regard to endoscopic biopsy specimens, the diagnosis of SC still depends on the pathological morphology, and no extra tumor tissue can be used to extract lncRNAs. In contrast, circulating lncRNAs  are ideal biomarkers due to their convenience and low invasiveness. Therefore, the present study on the application of circulating lncRNAs in the diagnosis of SC has greater clinical significance. Nevertheless, this meta-analysis possessed some limitations. First, this systematic review and metaanalysis lacks eligible non-Asian studies. Second, almost every study focuses on different lncRNAs, and it was difficult to perform subgroup analysis based on lncRNA types to explain the possible sources of heterogeneity. Third, obvious heterogeneity was found in the included studies. Although diagnostic meta-analysis suggested that the type of lncRNA was a source of heterogeneity through metaregression and subgroup analysis, the heterogeneity of sensitivity and specificity were still high in each subgroup.

Conclusions
In conclusion, the findings of the diagnostic metaanalysis provide evidence that circulating lncRNA tests exhibit a high accuracy for diagnosing SC, which is promising in clinical application due to their high positive and negative predictive value. This study provides an important reference value for the application of circulating lncRNAs as biomarkers for the early diagnosis of SC. Due to potential limitations, further investigations are warranted to verify the diagnostic role of circulating lncRNAs in SC.