Comparison of the diagnostic value of various microRNAs in blood for colorectal cancer: a systematic review and network meta-analysis

Background Despite the existence of numerous studies investigating the diagnostic potential of blood microRNAs for colorectal cancer, the microRNAs under consideration vary widely, and comparative analysis of their diagnostic value is lacking. Consequently, this systematic review aims to identify the most effective microRNA blood tumor markers to enhance clinical decision-making in colorectal cancer screening. Method A comprehensive search of databases, including PubMed, Embase, Web of Science, Scopus, and Cochrane, was conducted to identify case‒control or cohort studies that examined the diagnostic value of peripheral blood microRNAs in colorectal cancer. Studies were included if they provided sensitivity and specificity data, were published in English and were available between January 1, 2000, and February 10, 2023. The Critical Appraisal Skills Programme (CASP) checklist was employed for quality assessment. A Bayesian network meta-analysis was performed to estimate combined risk ratios (RRs) and 95% confidence intervals (CIs), with results presented via rankograms. This study is registered with the International Platform of Registered Systematic Review and Meta-analysis Protocols (INPLASY), 202,380,092. Results From an initial pool of 2254 records, 79 met the inclusion criteria, encompassing a total of 90 microRNAs. The seven most frequently studied microRNAs (43 records) were selected for inclusion, all of which demonstrated moderate to high quality. miR-23, miR-92, and miR-21 exhibited the highest sensitivity and accuracy, outperforming traditional tumor markers CA19-9 and CEA in terms of RR values and 95% CI for both sensitivity and accuracy. With the exception of miR-17, no significant difference was observed between each microRNA and CA19-9 and CEA in terms of specificity. Conclusions Among the most extensively researched blood microRNAs, miR-23, miR-92, and miR-21 demonstrated superior diagnostic value for colorectal cancer due to their exceptional sensitivity and accuracy. This systematic review and network meta-analysis may serve as a valuable reference for the clinical selection of microRNAs as tumor biomarkers. Supplementary Information The online version contains supplementary material available at 10.1186/s12885-024-12528-8.


Background
Colorectal cancer, which ranks third in global cancer prevalence and second in mortality, presents a significant health challenge [1].While developed regions such as Europe and the United States have observed a decline in incidence and mortality rates due to accessible screening and early treatment, the opposite trend is evident in certain low-and middle-income countries, where diagnosis and treatment are lacking [2].In addition, the multi-drug resistance of cancer also makes the treatment increasingly difficult [3].Thus, early detection and treatment are pivotal in enhancing colorectal cancer survival rates.
Peripheral blood tumor marker testing, a noninvasive procedure requiring only a patient's peripheral blood sample, offers a viable alternative to tissue biopsy and imaging.This method is not only easy and quick to administer but also eliminates the need for preoperative preparation and recovery time, making it highly generalizable.
MicroRNA, as a peripheral blood tumor marker, holds several competitive advantages over other tumor markers.First, microRNAs are highly specific, with distinct miRNA expression patterns associated with different cancer types [4].Second, they are stable, facilitating easy collection, storage, and transportation without degradation [5].Third, microRNAs are highly sensitive, potentially offering more accurate early tumor detection than traditional tumor markers [6].Finally, microRNA detection is versatile and useful not only for early cancer diagnosis but also for predicting treatment effects and prognosis assessment [7].Consequently, the potential of miRNAs as novel biomarkers in early colorectal cancer diagnosis has been extensively researched, providing a scientific foundation for their clinical application.
However, the practical application of miRNAs necessitates the identification of the most diagnostically valuable miRNAs by clinicians, given the wide variety of miRNAs studied.The lack of direct comparison among various miRNAs and the inability of existing systematic reviews to determine the most diagnostically valuable miRNA necessitates an indirect comparison of the diagnostic value of different miRNAs through Bayesian network meta-analysis.
This study aims to identify the most diagnostically valuable microRNAs as blood tumor markers for colorectal cancer detection through network meta-analysis.

Methods
The systematic reviews of observational studies were executed by the PRISMA guidelines (Preferred Reporting Items for Systematic Reviews and Meta-analysis) [8], with the study protocol registered under INPLASY202380092 [9].

Eligibility criteria and PICO definition
Participants: The diagnostic test population was bifurcated into two groups, namely, patients diagnosed with colorectal cancer and healthy individuals.
Intervention: Pretreatment levels of microRNA in the peripheral blood of patients.
Comparison: Clinical pathological results as the gold standard test.
Study Design: Cohort or case-control studies.Outcome: Sensitivity and specificity.Exclusion Criteria: Studies were excluded based on the following criteria: (1) To reflect current clinical practice, the timeliness of research is included in the exclusion criteria.Studies published before 2000 will be excluded [10,11].(2) Studies that were not published in the English language.(3) Manuscripts and conference abstracts that remained unpublished.(4) Studies that failed to report on either sensitivity or specificity.

Information sources
On April 14, 2024, a subsequent 'snowball' search was conducted.This involved scrutinizing the reference lists of publications that were eligible for a full-text review and utilizing Google Scholar to discover and scrutinize studies that cited them, with the aim of identifying further studies.

Search strategy
The following key words and MeSH terms (medical subject heading) in PubMed were used to find the related articles: (1)Search: ("Colorectal Neoplasms"[Mesh]) OR (Colorectal cancer).

Selection process
The selection process involved a rigorous review of titles and abstracts by three independent researchers (XJH, PLF, WD).Discrepancies were resolved through discussion until consensus was reached.The researchers then worked in pairs to independently screen the titles and abstracts of all retrieved articles.In cases of disagreement, a third researcher was consulted to make the final decision.The researchers also selected 7 microRNAs with a high number of cases for the net meta-analysis from a total of 106 microRNAs obtained by the nadir criteria screening.

Data collection process and data items
Two review authors (XS and DZY) independently extracted data from eligible studies using a customdesigned data extraction table.The extracted data were compared, and any inconsistencies were resolved through discussion.When any of the above information was unclear, we contacted the report authors to provide further details.Eligible results included sensitivity, specificity, accuracy, ROC (receiver operating characteristics) curve and area under the curve for peripheral blood microRNA diagnosis.For some multi-arm studies that do not directly provide the sensitivity and specificity of a miRNA, the decision threshold of the prediction model is achieved by the commonly used method of "maximizing the Jorden index" [12].Measurements were taken at the time point for all patients whose samples were collected before any treatment.

Study risk of bias assessment
The potential bias in the studies was evaluated using a scoring system grounded on the Critical Appraisal Skills Programme (CASP) checklist, designed explicitly for diagnostic studies [13].The CASP checklist has 12 questions to help understand a cohort or case-control study, which was independently applied to each of the included studies by the two review authors (XJH and PLF), who documented supporting information and the reasoning behind their bias risk judgment for each domain (low; high; some concerns).Any discrepancies in bias risk judgments or the reasoning behind these judgments were resolved through discussion until a consensus was reached between the two review authors.If necessary, a third review author acted as an arbitrator.

Effect measures and synthesis methods
This review assesses the diagnostic value of microRNAs across three dimensions: sensitivity, specificity, and accuracy.Studies that satisfied the inclusion and exclusion criteria and provided data for all three dimensions were deemed suitable for synthesis.The primary steps of the analysis in this paper are as follows: Firstly, sensitivity, specificity, and accuracy were compared with the estimated odds ratios (ORs) and 95% confidence intervals (CI) using a random-effects model.Also, I-square tests were run to detect the amount of heterogeneity for each pairwise meta-analysis.Secondly, a network graph was constructed to elucidate the interconnections among the microRNAs being scrutinized.This graphical representation was instrumental in enabling comparisons, both direct and indirect.The scale of nodes and edges within the network was determined by the sample size and the number of studies contributing to each comparison.The Bayesian network meta-analysis was applied to estimate the combined effect sizes of these comparisons, harnessing Markov chain Monte Carlo (MCMC) simulations.Noninformative priors underpinned these simulations to estimate the magnitude and precision of effects.The convergence of the model was confirmed after executing four separate chains and a preliminary burn-in phase consisting of 10,000 simulations.The probability distributions were derived from a subsequent series of 50,000 simulations [14].The RRs and 95% CIs were calculated to articulate our results, interpreting intervals that did not encompass the value of one as statistically significant.The heterogeneity was assessed using the I 2 statistic, calculated from the MCMC samples, considering values above 50% indicative of considerable heterogeneity among the cohorts compared [15].The relative efficacy of each microRNA was assessed through nanograms, which depicted the cumulative probability of each microRNA's effectiveness, ranging from the most to the least effective.The node-splitting methods were further employed to test the network meta-analysis's underlying assumption of consistency between direct and indirect evidence [16].
All computations were executed using R-4.0.3, with the "gemtc" and "netmeta" packages for network meta-analysis and the "mada" package for traditional meta-analysis.Furthermore, the mvmeta package was employed to plot inconsistency analyses and publication bias.

Selection and characteristics of the study
The PRISMA 2020 flow diagram for new systematic reviews, as depicted in Fig. 1, illustrates the process of the current systematic review.Out of 80 studies that met the inclusion criteria, 106 microRNAs were identified as potential hematologic tumor markers.However, due to the limited number of studies on most microRNAs, a comprehensive comparison via a systematic review and network meta-analysis was not feasible.Consequently, seven microRNAs with a substantial number of studies were selected for the network meta-analysis.The final selection comprised 43 studies  and seven prevalent microRNAs, namely, miR-150, 17, 20, 21, 23, 29, and 92.
The 28 case-control studies and 15 cohort studies included in this review, as detailed in Table 1, were published between 2002 and 2022.The studies involved a total of 6008 CRC patients and 5341 healthy controls, with sample sizes ranging from 41 to 1302.The geographical distribution of the studies included ten from Europe and America, 25 from Asia, and eight from Africa.The quality assessments, also presented in Table 1, indicate a moderate to high overall quality for all studies, with CASP check scores exceeding nine for each.

Pairwise meta-analysis
A traditional pairwise meta-analysis was conducted to estimate the odds ratio (OR) and 95% confidence interval (CI) for sensitivity and specificity.The summary results are presented in Table 2.The heterogeneity analysis results indicated that all I 2 values exceeded 50%, necessitating the use of a random effects model for the Bayesian network meta-analysis.

Network geometry
Figure 2 illustrates the network structure, reflecting the relationships between the different marker studies.The size of the nodes corresponds to the number of studies included in the final analysis.Direct comparisons are indicated by solid lines between nodes, while the thickness and depth of the colors represent the number of studies compared between the two methods.

Sensitivity
Figure 3A illustrates that when benchmarked against postoperative pathological results, miR-29 exhibited the lowest diagnostic sensitivity among the seven peripheral circulating microRNA indicators, with a relative risk (RR) of 0.35 and a 95% confidence interval (CI) of 0.26-0.47.Conversely, miR-23 demonstrated the highest diagnostic sensitivity, with an RR of 0.87 and a 95% CI of 0.75-1.00,closely mirroring the postoperative pathological results.
Table 3 reveals that, in comparison to CA199 and CEA, commonly used clinical tumor indicators, all seven microRNA indicators displayed superior sensitivity to CA199, with miR-21, miR-23, and miR-92 outperforming CEA.miR-20 and miR-29 were less sensitive than miR-21, miR-23, and miR-92.A node-splitting analysis was conducted to evaluate the discrepancy between indirect and direct comparisons across all modalities.This analysis revealed inconsistencies in some comparisons  (p value < 0.05), including those between miR-23 and CA199, CEA and miR-17, and miR-21 and CEA (Supplementary Fig. 1).These controversial comparisons are labeled in Table 3.
The diagnostic sensitivities of the seven microRNAs, in descending order, were miR-23, 92, 21, 17, 150, 20, and 29, as depicted in the rankograms in Fig. 4A and the ranking table in Table 4.

Specificity
Figure 3B indicates that miR-17 had the lowest diagnostic specificity among the seven peripheral blood circulating microRNA indicators, with an RR of 0.72 and a 95% CI of 0.60-0.86.miR-29 demonstrated the highest specificity, with an RR of 0.86 and a 95% CI of 0.75-0.99.
Compared to CA199 and CEA, miR-17 was less specific than CEA, while the remaining microRNAs did not exhibit statistically significant differences in specificity.Table 3 shows no significant difference in diagnostic specificity among the microRNAs.Node-split analysis revealed inconsistency between indirect and direct comparisons between miR-21 and CA199 (Supplementary Fig. 2).
The diagnostic specificity of the seven microRNAs, in descending order, were miR-29, 20, 23, 21, 92, 150, and 17, as depicted in the rankograms in Fig. 4B and the ranking table in Table 4.

Accuracy
Figure 3C shows that miR-17 had the lowest diagnostic accuracy among the seven peripheral blood circulating microRNA indicators, with an RR of 0.71 and a 95% CI of 0.62-0.81.miR-23 demonstrated the highest accuracy, with an RR of 0.87 and a 95% CI of 0.78-0.97.
The diagnostic accuracy of the seven microRNAs, in descending order, were miR-23, 92, 21, 20, 29, 150, and 17, as depicted in the rankograms in Fig. 4C and the ranking table in Table 4.

Publication bias
To assess the publication bias of this study, a funnel plot was constructed.Figure 5a and b present the funnel plots for sensitivity and specificity, respectively.The relative symmetry of the plots suggests a minimal publication bias, which can be disregarded.Subsequently, Egger's test (p = 0.241 for sensitivity and 0.188 for specificity) also support the view.The gray line symbolizes the null hypothesis that the study-specific effect sizes are not different from the respective comparison-specific pooled effect estimates.The green line represents the regression line, with different colors corresponding to different comparisons.

Discussion
Multidrug resistance in cancer makes the treatment increasingly difficult [3], so early identification of colorectal tumors is crucial in mitigating the mortality rates associated with colorectal cancer.Blood tumor markers are considered as straightforward, noninvasive, and readily available among the various diagnostic methods for colorectal cancer.Circulating miRNAs, capable of withstanding adverse physiological conditions such as extreme pH and temperature fluctuations and multiple freeze-thaw cycles, have recently emerged as a promising tool for early colorectal tumor screening [60].Given the broad spectrum of circulating miRNAs, each with differing sensitivity and specificity, a network meta-analysis was conducted to compare each circulating miRNA's diagnostic pros and cons, correlating them with the commonly employed blood-based colorectal cancer biomarkers CA-199 and CEA.miR-23 exhibited the highest sensitivity, and higher sensitivity reduces the likelihood of false negatives, thereby saving time in conducting more definitive diagnostic tests [61].miR-29 demonstrated the highest specificity, and high specificity (high true-negative rate) can prevent significant psychological stress or additional diagnostic costs for the patient due to false-positive tests [61].The accuracy of an assay, its ability to correctly distinguish between patients and healthy cases, is generally judged by its sensitivity and specificity.Our findings indicate that miR-23 has the highest accuracy and overall function compared to other markers.miR-23 is an oncomiR that inhibits the expression of pyruvate dehydrogenase kinase 4, which activates pyruvate dehydrogenase and oxidative phosphorylation to produce sufficient ATP for cell proliferation [62].Wang et al. found that miR-23a promotes the migration and invasion of CRC cells and tumor stem cells by targeting the metastasis suppressor 1 (MTSS1) gene [63].The expression of miR-23a was reported to be positively correlated with the clinical stage and infiltration depth of the tumor, and high expression levels of miR-23a in tissues were found to be a poor prognostic marker of cancer [64].
Given that a single tumor marker indicator cannot fulfill the screening needs in terms of detection sensitivity and specificity, the concurrent use of multiple blood tumor markers as a panel can enhance sensitivity and specificity.Referring to this network meta-study, the more studied circulating miRNAs with high diagnostic accuracy, such as miR-23, 21, 92, can be selected for panel composition, and pairing with traditional blood CEA assays to improve the panel's specificity.
Next, how to integrate the newly identified miRNAs into the existing colorectal cancer screening program?Firstly, it is necessary to conduct multicenter studies with independent samples to validate the sensitivity, specificity, and predictive value of these miRNAs.After confirming their effectiveness, the screened miRNA tests can be used as standalone tests or in combination with existing tests (such as fecal occult blood test and colonoscopy).When used as an independent screening tool, miRNA tests can be used as a preliminary screening tool to identify high-risk individuals for further colonoscopy examination.When used as a combined screening tool, miRNA tests can be used in conjunction with existing tests like fecal occult blood test to improve the accuracy and coverage of screening.
Additionally, we have designed methods to compare the cost-effectiveness of miRNA testing with traditional screening methods to evaluate its economic feasibility.First, cost analysis will calculate the direct costs of miRNA testing (including reagents, equipment, and  labor) compared to the costs of traditional screening methods.Next, benefit analysis will estimate the longterm benefits such as medical cost savings and improved quality of life due to early detection of colorectal cancer.Lastly, a cost-effectiveness model will be established using decision tree models [65] or Markov models [66] to compare the long-term health economic outcomes of different screening strategies.Factors to consider include cancer detection rates, treatment success rates, and changes in quality of life caused by screening.Finally, after confirming the effectiveness of the miRNA screening method, specific implementation plans will be formulated and this new strategy will be promoted through appropriate channels, including education and training, public awareness, and policy support.Our analysis was constrained by the data of the included studies and the structure of the reported data.Initially, 79 articles were screened with a total of 105 miR-NAs, but not all of them could be used in our analysis, so the 7 most studied miRNA metrics were streamlined based on the number of studies.In some articles where sensitivity and/or specificity were unavailable in the original article, sensitivity and specificity were indirectly derive using AUC plots (principle of maximum area under the curve), which may result in observed heterogeneity in pairwise meta-analyses and potentially affect the accuracy of network meta-analyses.Additionally, due   to the limitations in the data of the included studies, this article did not perform subgroup analysis on colorectal cancer at different stages of development.Whether the diagnostic value of miRNA is universally applicable in early and advanced colorectal cancer requires further investigation.Last, although the sensitivity and specificity of miR-23 were high, the number of articles studying miR-23 was still needed to be improved, thus potentially leading to inconsistencies between the results of direct and indirect comparisons shown in the nodal analysis.

Conclusions
In conclusion, circulating microRNAs have high diagnostic value for colorectal cancer, which is not inferior to traditional CEA and CA19-9.miR-23, 92, and 21 had high diagnostic value in terms of sensitivity, with sensitivities of 87%, 82%, and 79%, respectively, when combined.In terms of specificity, miR-29, 23, and 20 had high diagnostic value, and the specificity after combination was 86%, 86%, and 84%, respectively.Combining sensitivity and specificity, miR-23, 92, and 21 had high accuracies of 87%, 81%, and 79%, respectively.This systematic review and network meta-analysis may provide a reference basis for the clinical selection of circulating miRNAs as tumor biomarkers for the early detection of CRC and improved survival of CRC patients.

Fig. 1
Fig. 1 PRISMA 2020 flow diagram for new systematic reviews, which included searches of databases, registers and other sources

Fig. 2
Fig. 2 Evidence network plot of the diagnostic value of pathology and 7 different blood biomarker tests

Fig. 5
Fig. 5 Comparative adjusted funnel plot for publication bias

Table 1
Characteristics of articles included in the network meta-analysis

Table 2
Traditional paired meta-analysis of the sensitivity and specificity of individual microRNAs OR: odds ratio; CI: confidence interval CRC: colorectal cancer; EVs: serum extracellular vesicles; NR: not reported.The scoring system based on the CASP checklist (specified for diagnostic studies) was applied for all studies.Table 1 (continued)

Table 3
Relative effects and 95% confidence intervals of all pairwise panels for sensitivity, specificity and accuracy based on the Bayesian network meta-analysis method

Table 4
Ranking of sensitivity, specificity and accuracy of different miRNAs in diagnosing CRC