We search literature from Pubmed, Web of Science, and Embase databases. Search time is from the establishment of the database to January 2023. The details of search words are listed in Supplementary Table 1 . Then we manually searched for references of relevant articles to identify potentially eligible studies. Two researchers independently screened titles and abstracts. If there were any differences, we would discuss them until achieving a consensus.

The inclusion criteria were as follows (1): assessing the role of ncRNA in the prognosis of GIC (2); providing at least one prognostic outcome data (overall survival, disease-free survival, progression-free survival, and recurrence-free survival) (3); containing meta-analysis in the study.

The exclusion criteria were as follows (1): focus on genetics or experiments not in humans (2); full-text not available (3); lack critical information.

If two or more eligible studies evaluate the prognostic value of the same ncRNA for the same disease, we will include the largest number of original studies.

The first author, year of publication, journal, disease, kind of ncRNA, number of studies, total population, results, and tools for assessing the risk of bias were extracted. At the same time, we extract the number of studies, number of participants, relative risk, P value, I², effect model, and publication bias in each meta-analysis. If the study carries out subgroup analysis according to the types of ncRNA, we will extract the relevant results. Two researchers independently extracted the data and cross-verified it.

A Measurement Tool to Assess systematic Reviews 2 (AMSTAR2) is a questionnaire that asks reviewers to answer ‘yes,’ ‘partly yes,’ or ‘no’ (18). Among the total 16 items, seven items are considered the most important when assessing the quality of meta-analyses: registration in advance, reasonable search strategy, reasonable exclusion of literature, adoption of appropriate bias evaluation tools, selection of appropriate statistical methods, consideration of the impact of bias on results, and consideration of potential bias risks of articles. In the study, we viewed two ‘part yes’ as one ‘yes.’ According to the evaluation of 16 items, the final results are evaluated as high, moderate, low, or critically low. Two researchers independently evaluate the quality and cross-verified it. For visual display, we use Python to make forest maps to reveal the prognostic value of ncRNAs to GICs.

Grading of Recommendations Assessment, Development, and Evaluation (GRADE) methods propose five factors rating down certainty in the evidence (the risk of bias, inconsistency, indirectness, imprecision, and publication bias) and two factors rating up certainty in the evidence (large effect and dose-response) (19, 20). Based on GRADE, outcomes are evaluated as high, moderate, low, or very low quality. As for prognostic studies, the quality of evidence was high, and we downgraded and upgraded them by five rating down factors and two rating up factors, respectively (21). Only the standardized, systematic evaluation of research reports is suitable for grading the results, so we do not grade the research results rated as extremely low quality by ASMTAR2. Two researchers completed this step independently. Moreover, the results were shown by corresponding scales using Python.

Overall, 1218 articles were retrieved from three databases. After removing 742 duplicates and screening the titles and abstracts, 171 articles were identified. A further 113 articles were excluded during the full-text reading for the following reasons: 13 articles had no relevant outcome, 21 articles did not perform a meta-analysis, 4 studies have no full text, and 75 articles discussed the same topic. Ultimately, 58 studies were included in this umbrella review. As shown in Figure 1 .

CRC is a research hotspot. There are 25 studies to analyze the prognostic value of ncRNA for CRC (22–46). One study analyzed circRNAs, while Another study analyzed ciR-7 separately (36, 42). One study included 42 original studies for meta-analysis for lncRNA (25), and the remaining seven studies were analyzed for different lncRNA. In addition, 23 different miRNAs were analyzed in 15 studies. Details can be found in Table 1 , Supplementary Table 2 and Supplementary Figure 1 .

Of the 25 studies, ten were considered critically low methodological quality. The main defects are no protocol or guidance literature and insufficient consideration of the risk of bias included in the study. We further use GRADE to evaluate the meta-analysis of the remaining studies. Among them, lncRNA HOTAIR is considered highly credible, and it has no serious problems in the five degradation factors. It has an upgrade factor of a large magnitude of effect. While circRNA(up), ciRs-7, lncRNA CRNDE, lncRNA UCA1, miR-124, and miR-203 are considered moderate credibility, they have one or two problems in reducing factors. At the same time, circRNA(down), lncRNA MALAT1, and miR-133 are considered to have low credibility. Other studies have extremely low credibility. The detailed evaluation process is in the Supplementary Materials .

MiR-203 has no significant relationship with CRC in the medium or high-reliability meta-analysis. The other six analyses have a significant relationship, as shown in Supplementary Table 3 .

Seven studies have reported the prognostic value of ncRNA for EC, including four reports of lncRNA, two reports of miRNA, and one report of circRNA (25, 28, 47–51). Two study made subgroup analyses of lncRNA and miRNA, respectively, and analyzed their prognostic value according to the types of ncRNA (47, 49). Details can be found in Table 1 , Supplementary Table 2 and Supplementary Figure 2 .

According to ASMTAR2, one study was considered critically low quality because two important indicators have not been met (28). Other studies further evaluate the credibility of their meta-analysis according GRADE. Among them, circRNAs, lncRNA HOTAR, lncRNA AK001796, lncRNA Casc9, and lncRNA MEG3 are considered as moderate credibility. Although they have problems in continuity or directness, they are rated as medium because of large magnitude of effect. LncRNA MEG3 was negatively correlated with EC prognosis, and the other four were positively correlated with EC prognosis. In addition, lncRNA Linc00460, lncRNA PCAT-1 and lncRNA UCA1 are considered to be of low reliability. The other 25 meta-analyses are of extremely low reliability. The detailed evaluation process is in the Supplementary Materials .

As another research hotspot, 23 studies about SC were included in this study (25, 27, 36, 44–46, 52–68). Two studies related to circRNA (36, 53). One study conducted a detailed subgroup analysis according to the types of lncRNAs (55), and other eight studies also analyzed different lncRNAs. The remaining 11 studies are all related to miRNA. Fifteen studies used NOS to evaluate the quality of included literature, and eight studies used other methods or were not reported. Details can be found in Table 1 , Supplementary Table 2 and Supplementary Figure 3 .

Due to the failure to meet the requirements of some critical items, only 12 studies were rated as low quality or above. After carefully using GRADE, we found that circRNAs (up), circRNAs (down), ciRs-7, and lncRNA PVT1 were considered moderately credible. Among them, circRNAs (down) are negatively correlated with the prognosis of SC, while other ncRNAs are positively correlated with the prognosis of SC. Meta-analysis without high-level credibility. MiR-125a, miR-125b, and miR-145 are rated as extremely low credibility, and their evidence is insufficient. Other meta-analyses are low credibility. The detailed evaluation process is in the Supplementary Materials .

Seven studies each reported the prognostic value of miRNA and lncRNA for LC (27, 37, 44, 45, 60, 69–77), and one reported the circRNA (78). Among them, one study included the most original research, including 29 original studies for meta-analysis (75). Details can be found in Table 1 , Supplementary Table 2 and Supplementary Figure 4 .

Five studies were considered to be of critically low quality. One study did not meet one unimportant item but met the other 15 items, so it was considered high quality. The remaining nine studies were of low quality. In the reliability evaluation of meta-analysis, although there are some problems in precision, lncRNA SNHG16 is rated as high reliability because it meets the upgrade conditions. circRNAs(up) is considered as moderate credibility. The remaining ten meta-analyses are of low or extremely low reliability. The detailed evaluation process is in the Supplementary Materials .

Two studies have reported the prognostic value of lncRNA for PC respectively (27, 71). At the same time, four other studies reported the situation of miRNA (38, 45, 70, 79). At present, we have not found any research on circRNA. Details can be found in Table 1 , Supplementary Table 2 and Supplementary Figure 5 .

Four studies were rated as low quality and above. According to GRADE, only miR-21 was rated as medium quality, and the other 20 meta-analyses were rated as low quality or extremely low quality due to different degradation factors. The detailed evaluation process is in the Supplementary Materials .