Malignant tumor has become the number one killer to human health and early diagnosis of these malignancies remains a compelling challenge for the clinicians. For the serological diagnosis in cancer thus far, classical blood-based tumor markers like carcinoembryonic antigen (CEA), prostate specific antigen (PSA), and carbohydrate antigen (CA) have gained a lot of recognition in the diagnosis or prediction of variety malignant tumors. Nevertheless, the utility of these available tumor markers is limited by disappointing diagnostic accuracies, especially with respect to their applications in diagnosing early phase carcinomas or incapability in distinguishing aggressive tumors from the indolent ones [
Analysis of molecular genetic markers in biological fluids or tissues has been proposed as a useful tool for cancer diagnosis. MicroRNAs (miRNAs) are a class of short, single stranded, approximately 18–25 nucleotide noncoding RNAs and are cleaved from 70 to 100 nucleotide hairpin precursors by a complex protein system that involves the ribonucleases (RNases) III Drosha and Dicer, Pol-II-dependent transcription, and members of the argonaute family [
The current meta-analysis followed the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) Statement and methods [
All eligible studies satisfying the following criteria were firstly included in our analysis: (1) miR-155 was assessed in cancer diagnostic studies; and (2) studies mentioned the sample number, sensitivity, specificity, AUC, and their 95% confidence intervals (CIs) or other more detailed information. Studies were excluded based on the following criteria: (1) studies that failed to explicitly state the control groups; and (2) review articles, meta-analysis, letters, commentaries, abstracts presented in conferences, and studies without complete data.
Two reviewers independently evaluated the final set of selected articles. The extracted data elements of this study for diagnosis included the first author, year of publication, country of origin, number of patients, control sources, sample types, miRNA profiles, test method, diagnostic parameters, and other substantial information. In studies containing both a training and a validation group, data of each group was regarded as a single study in the meta-analysis. Any disagreement was resolved by consensus.
The evidence-based and critical review checklist of quality assessment of diagnostic accuracy studies (QUADAS) tool was used to assess each study’s quality [
Statistical analysis was conducted utilizing Stata 12.0 (Stata Corporation, College Station, TX, USA) and Meta-disc 1.4 (XI Cochrane Colloquium, Barcelona, Spain) software. The bivariate meta-analysis model was employed to summarize the sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) and generate the bivariate SROC curve with their corresponding 95% CIs. For the heterogeneity analysis, Spearman correlation coefficient was performed to analyze the threshold effect, while Cochran-
Flow chart for study selection is shown in Figure
Flow diagram of the studies identification and selection.
In this meta-analysis, the final set of 17 diagnostic studies included a total of 886 patients with various cancers and 670 healthy control individuals. The sample size of cancer patients in each study varied from 20 to 103, and control numbers varied from 6 to 92. All the cancer patients were diagnosed based on the histopathological examination. Among the 17 studies, 6 studies had an ethnicity of Caucasian, 11 studies had an ethnicity of Asian, and 13 studies conducted single miR-155 assay for cancer detection. Besides, all the 17 studies performed reverse transcription quantitative PCR (RT-qPCR) for the miRNAs detection, and the specimen type included serum [
The main features of the included studies for miR-155 in the diagnosis of cancers. DLBCL: diffuse large B cell lymphoma; AML: acute myeloid leukemia; QUADAS: quality assessment for studies of diagnostic accuracy.
Author | Year | Ethnicity | Patients |
Cancer |
Sample | Test method | MicroRNA profiling | QUADAS |
---|---|---|---|---|---|---|---|---|
Mar-Aguilar et al. [ |
2013 | Caucasian | 61 (10) | Breast cancer | Serum | RT-qPCR | miR-155 | 12 |
|
||||||||
Sun et al. [ |
2012 | Asian | 103 (55) | Breast cancer | Serum | RT-qPCR | miR-155 | 13 |
|
||||||||
Zhao et al. [ |
2012 | Asian | 20 (10) | Breast cancer | Serum | RT-qPCR | miR-155 | 12 |
|
||||||||
Fang et al. [ |
2012 | Asian | 75 (77) | DLBCL | Serum | RT-qPCR | miR-155 | 12 |
|
||||||||
Xie et al. [ |
2012 | Asian | 45 (30) | AML | Serum | RT-qPCR | miR-155 | 8 |
|
||||||||
Pan et al. [ |
2014 | Asian | 30 (26) | Pancreatic cancer | Plasma | RT-qPCR | miR-155 | 8 |
|
||||||||
Wang et al. [ |
2009 | Asian | 30 (29) | Pancreatic cancer | Plasma | RT-qPCR | miR-155 | 12 |
|
||||||||
Abd-El-Fattah et al. [ |
2013 | Caucasian | 65 (37) | Lung cancer | Serum | RT-qPCR | miR-155 | 11 |
|
||||||||
Gao et al. [ |
2014 | Asian | 36 (32) | Lung cancer | Serum | RT-qPCR | miR-155 | 9 |
|
||||||||
Tang et al. [ |
2013 | Asian | 96 (92) | Lung cancer | Plasma | RT-qPCR | Single miR-155 and paneled miRNAs | 12 |
|
||||||||
Liu et al. [ |
2012 | Asian | 60 (60) | Esophageal cancer | Plasma | RT-qPCR | miR-155 | 9 |
|
||||||||
Yang et al. [ |
2015 | Asian | 20 (20) | Rectal cancer | Tissue | RT-qPCR | miR-155 | 12 |
|
||||||||
Gombos et al. [ |
2013 | Caucasian | 40 (40) | Oral squamous cell cancer | Tissue | RT-qPCR | miR-155 | 9 |
|
||||||||
Häusler et al. [ |
2010 | Caucasian | 24 (15) | Ovarian cancer | Whole blood | RT-qPCR | More than thirty paneled miRNAs involved miR-155 | 11 |
|
||||||||
Heneghan et al. [ |
2010 | Caucasian | 83 (63) | Breast cancer | Whole blood | RT-qPCR | Three paneled miRNAs |
12 |
|
||||||||
Roa et al. [ |
2012 | Caucasian | 24 (6) | Lung cancer | Sputum | RT-qPCR | Five paneled miRNAs |
12 |
|
||||||||
Zheng et al. [ |
2011 | Asian | 74 (68) | Lung cancer | Plasma | RT-qPCR | Three paneled miRNAs involved miR-155 | 12 |
We estimated the quality of the 17 included publications according to the 14-item QUADAS assessment tool [
Summary of assessment of the included studies analyzed using the quality assessment for studies of diagnostic accuracy (QUADAS) tool: proportion of studies with low (Yes), mediate (Unclear), and high risk of bias (No).
As displayed in Table
Heterogeneity assessment of the individual pooled analysis.
Analyses | Spearman correlation coefficient |
Cochran’s |
|
Heterogeneity | |
---|---|---|---|---|---|
Threshold effect | Nonthreshold effect | ||||
Single miR-155 | −0.363a |
46.73b |
72.2 | No | Yes |
Panel miR-155 | −0.606a |
35.98b |
69.4 | Yes | Yes |
Serum-based | −0.571a |
16.06b |
62.6 | No | Yes |
Plasma-based | 0.359a |
10.42b |
61.6 | No | Yes |
Asian | −0.409a |
25.23b |
60.4 | No | Yes |
Caucasian | 0.500a |
2.16b |
7.4 | No | No |
Overall | −0.385a |
85.38b |
70.7 | No | Yes |
Outliers excluded | −0.418a |
46.76b |
70.3 | No | Yes |
aThe value of sensitivity and (1 − specificity); b
Since there existed significant heterogeneity between studies, the random-effects model was applied in the meta-analysis. Forest plots of the sensitivity and specificity for miR-155 validation in diagnosing cancers are shown in Table
Summary table of the diagnostic accuracy of miR-155 for various cancers.
Analyses | Sensitivity |
Specificity |
DOR |
Positive LR |
Negative LR |
AUC (95% CI) |
---|---|---|---|---|---|---|
MicroRNA profile | ||||||
Single miR-155 | 0.82 (0.73–0.88) | 0.77 (0.70–0.83) | 15.34 (7.32–32.17) | 3.56 (2.55–4.95) | 0.23 (0.14–0.37) | 0.85 |
Paneled miR-155 | 0.76 (0.68–0.82) | 0.82 (0.77–0.86) | 14.27 (7.98–25.53) | 4.23 (3.12–5.75) | 0.30 (0.22–0.40) | 0.86 |
Sample types | ||||||
Plasma-based | 0.74 (0.58–0.71) | 0.70 (0.63–0.76) | 5.54 (2.79–11.03) | 2.02 (1.65–2.48) | 0.41 (0.27–0.63) | 0.73 |
Serum-based | 0.78 (0.74–0.82) | 0.77 (0.71–0.82) | 15.61 (7.03–34.67) | 3.47 (2.40–5.02) | 0.23 (0.13–0.41) | 0.87 |
Ethnicity | ||||||
Asian | 0.72 (0.68–0.76) | 0.72 (0.68–0.76) | 8.03 (4.82–13.38) | 2.68 (2.07–3.46) | 0.38 (0.29–0.50) | 0.81 |
Caucasian | 0.94 (0.89–0.97) | 0.79 (0.69–0.87) | 61.93 (23.00–166.75) | 5.79 (1.51–22.24) | 0.08 (0.04–0.15) | 0.96 |
Overall | 0.79 (0.73–0.84) | 0.79 (0.75–0.84) | 14.62 (9.10–23.48) | 3.86 (3.05–4.90) | 0.26 (0.20–0.35) | 0.86 |
Outliers excluded | 0.84 (0.74–0.90) | 0.76 (0.68–0.83) | 16.41 (3.32–36.77) | 3.54 (2.49–5.05) | 0.21 (0.13–0.36) | 0.86 |
CI: confidence interval; LR: likelihood ratio; DOR: diagnostic odds ratio; AUC: area under the curve.
Forest plots of the pooled sensitivity and specificity for single miR-155 in detecting cancers. (a) Sensitivity; (b) specificity. Only the first author of each study is given. Sensitivity and specificity were given with confidence intervals (CIs).
Pre- and posttest probabilities of single and paneled miR-155 analyses. (a) Single miR-155 test; (b) paneled miR-155 test.
The SROC curves of the pooled individual analyses. Sample size is indicated by the size of the square. The regression SROC curve indicates the overall diagnostic accuracy. (a) Single miR-155 test; (b) paneled miR-155 test; (c) serum-based miR-155 test; (d) Asian population-based miR-155 test. AUC: area under curve,
The results of our subgroup analyses are summarized in Table
As shown in Figure
Influence and outlier detection analyses of the overall pooled study: the intermediate variable of RR (a) and outlier detection analysis (b). RR: relative risk.
Therefore, we further conducted meta-regression analysis by adding a total of 7 prespecified covariates (design type, study quality, specimen type, ethnicity, cut-off value setting, number of cases, and number of controls) to the bivariate model to assess their impacts on sensitivity and specificity. In consideration of the small study size of our analysis, a permute metaregression module was employed with a check value of 10,000, and each time only two covariates were estimated by Stata 12.0 software. Our data exhibited that ethnicity (
Publication bias of the included studies was checked by Deeks’ funnel plot asymmetry test. The slope coefficient was associated with a
Funnel plot test for the assessment of potential bias for the analyses. (a) Deeks’ funnel plot asymmetry test for the single miR-155 analysis,
Sensitive and specific tumor biomarkers are essential to early cancer detection and diagnosis and for undertaking novel therapeutic trials and prevention strategies in clinic. In recent years, aberrant expression of miRNAs has been widely reported in various carcinomas [
Our data showed promising accuracy for single miR-155 detection in diagnosing tumors, in which the overall pooled sensitivity was 0.82 and specificity was 0.77, with an AUC of 0.85, suggesting that miR-155 achieved a relatively high accuracy for cancer detection. In a meta-analysis study, the diagnostic odds ratio (DOR), defined as the ratio of the odds of a true positive to the odds of a false-positive, is an important indicator of diagnostic accuracy: a DOR value less than 1.0 often indicates a low discriminating ability in the diagnostic test [
Interestingly, the parallel testing of miR-155 seemed to achieve a high diagnostic accuracy for the differentiation between cancer patients and healthy people, with a sensitivity of 0.76 and specificity of 0.82, displaying an AUC of 0.86. Strikingly, the pooled specificity was presented as 0.82, indicating a more specific discriminatory performance of miR-155 panel than single miR-155 test in cancers. Moreover, the pooled PLR of 4.23 also suggested an increased diagnostic performance for the combined miR-155 test. Research from Gombos et al. [
Furthermore, we conducted subgroup analyses based on the following variables like sample type and ethnicity. Notably, our analysis based on sample type showed that using serum miR-155 as biomarker for detecting cancers yielded an overall sensitivity of 0.78 and an overall specificity of 0.77. The AUC of 0.87 and DOR of 15.61 also indicated a relatively high level of the diagnostic accuracy. Data from a newly published meta-analysis revealed that serum-based miRNA assay seemed to undergo a higher combined DOR, NLR, and AUC than that of plasma-based test, suggesting that serum may be a better matrix for diagnostic profiling of miRNAs in breast cancer [
In this meta-analysis, heterogeneity from threshold effect existed in the single miR-155 test. The threshold effect was mainly generated by the different cut-off value setting or thresholds used in different studies. The cut-off values for miR-155 test were not uniformed among studies, which may further contribute the heterogeneity from threshold effect. On the other hand, the pooled DOR is often used to discuss the heterogeneity caused by nonthreshold effects [
Our data demonstrated that miR-155 has a potential of being promising biomarker of cancers. However, several points should be addressed in their interpretation. First, for the researchers, how to select an appropriate cut-off value for the test is a vital point. All the enrolled studies in this meta-analysis varied for their cut-off values setting, and most of the studies used median or mean value in their laboratory or hospital as the cut-off value thus far. Second, as we obtained different diagnostic accuracy for the matrix-based studies, which matrix should be used for the test, plasma, serum, whole blood, tissue, or other bodily substances? Serum, plasma, and blood are easy to obtain and are convenient to be monitored routinely, whereas the tissues are widely utilized resources for miRNA study currently, especially in some research laboratories. Last, other than single miR-155 test, the miR-155 panel similarly revealed promising accuracy for the cancer detection, so which should be conducted for the diagnosis test, single or paneled miR-155 test, still warrants further investigations. Last, difficulty still remains for miR-155 as a new diagnostic marker for various carcinomas: aberrant miR-155 signature was depictured in various cancers instead of a particular one, which has compromised the diagnostic specificity when used in the practice. In this aspect, the combination of miR-155 with the circulating protein-biomarkers may be a novel potential tool for cancer detection. A new proof has shown that parallel testing of miR-155 and serum CEA level preoperatively can afford more accurate information for colon cancer diagnosis [
In summary, our findings clearly demonstrated that miR-155 confers high diagnostic accuracy for cancer detection, and combined sequential testing of miR-155 achieves an improved specificity compared to single miR-155 assay. Despite the promising results, the current study does have limitations involving the small study size as well as the substantial heterogeneity from nonthreshold effect existing among studies. In consequence, the combined diagnostic indices of miR-155 in this study are unable to completely mirror its actual diagnostic value for cancers, and, further, large cohort studies are still warranted.
The authors declare that they have no conflict of interests.