Recently, several studies have shown that blood-based microRNAs in patients with pancreatic cancer (PC) could be aberrantly expressed. The purpose of this meta-analysis was to evaluate blood-based microRNAs as novel biomarkers for diagnosis of PC.
Eligible studies which had evaluated the diagnostic performance of blood-based microRNAs and had been published from February 2004 to February 2014 were retrieved. The quality of the studies was evaluated with the QUADAS-2 tool. The performance characteristics were pooled using random-effects models. Statistical analysis was performed with STATA and Meta-Disc1.4 software.
The global meta-analysis included 12 studies from 8 articles, which contained 1,060 blood-based samples of PC patients and 935 blood-based samples of non-PC patients. Summary results suggested pooled sensitivity of 0.87 (95% confidence interval [95% CI], 0.85-0.89), specificity 0.92 (95% CI, 0.90-0.94), positive likelihood ratio 11.18 (95% CI, 5.57-22.46), negative likelihood ratio 0.16 (95% CI, 0.11-0.23), diagnostic odds ratio 88.98 (95% CI, 39.85-198.69) and the area under the summary receiver operating characteristic (SROC) curve 0.96.
This meta-analysis demonstrated blood-based microRNA expression profiles with the potential to discriminate PC patients from non-PC patients, which have moderate diagnostic accuracy. However, further validation studies are needed for their clinical significance in the diagnosis of PC to be established.
Tumori 2015; 101(2): 199 - 205
Article Type: ORIGINAL RESEARCH ARTICLE
AuthorsQing-Cai Meng, Wang-Wang Qiu, Zi-Qiang Wen, Hong-Cheng Wang, Ze-Zhi Shan, Zhou Yuan, Xin-Yu Huang
- • Accepted on 23/10/2014
- • Available online on 20/03/2015
- • Published in print on 28/04/2015
This article is available as full text PDF.
Pancreatic cancer (PC) is a common gastrointestinal malignancy and the fourth leading cause of cancer-related deaths in world, with a 1- and 5-year survival rates of about 20% and 6% (1-2-3). Nowadays, surgery is still the only effective treatment, while a majority of patients are diagnosed with locally advanced or metastatic lesions and lose the chance for an operation due to the difficulty of early diagnosis of PC. In a clinical setting, carbohydrate antigen 19-9 (CA19-9) and carcinoembryonic antigen (CEA) as conventional serum biomarkers, have been used for diagnosis and detection of PC (4, 5), but they lack sufficient sensitivity and specificity for the early diagnosis of PC. Therefore, to improve the survival rate of PC, effective diagnosis methods for early-stage tumors are urgently needed.
MicroRNAs (miRNAs) are small noncoding RNAs included 17 to 25 nucleotides, regulating the function of the target gene by inhibiting translation of mRNA or degrading mRNA. The aberrant expression of miRNAs is closely correlated with initiation and progression of tumors as well as the prognosis. Today, there are more than 2,500 human miRNAs being reported gradually, and miRNA expression profiles originating from samples of PC tissue have been described (6-7-8-9). Human miRNAs stably present in the blood by resisting degradation of Rnase and which show a difference between peripheral blood miRNA in patients with tumors and the normal population have been reported (10-11-12-13-14-15). To date, several clinical studies have identified aberrant miRNA expression panels from blood-based samples in PC patients, such as plasma, serum, whole blood, etc. (11-12-13-14-15-16-17-18), which suggest that the blood-based miRNA expression pattern = novel biomarkers for PC diagnosis.
Therefore, the present meta-analysis was performed to establish the overall diagnostic accuracy of miRNAs, to provide more reliable evidence that may help surgeons in the diagnosis of PC.
Methods and Study design
Search strategy and study selection
We carefully searched online PubMed/Medline, Web of Science and Embase from February 2004 to February 2014 to identify relevant studies, using the terms “pancreatic cancer/neoplasm” and “microRNAs,” in combination with keywords such as “diagnosis,” “plasma,” “serum” and “biomarker.” Eligible studies had to meet the following criteria: (i) study of miRNA expression profiling in PC patients; (ii) blood-based samples obtained from PC patients before treatment; (iii) each study included more than 10 patients; and (iv) data for diagnostic accuracy, such as sensitivity and specificity. Studies were excluded according to the following criteria: (i) non-English-language publication; (ii) samples from animals; and (iii) reviews, case reports and letters. Each study was assessed independently by 2 reviewers (Q.-C. M. and W.-W. Q.).
According to the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2), we systematically assessed the quality of all studies included. Four key domains of the tool include the following: (i) patient selection; (ii) index test; (iii) reference standard; (iv) flow and timing. Each domain is assessed by the risk of bias, and the first 3 are also assessed by applicability. Assessment results are presented as low, high or unclear risk of bias or concerns regarding applicability for each domain.
In accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (19), 2 reviewers (Q.-C. M. and W.-W. Q.) independently carried out the data extraction, and disagreements between reviewers were resolved by a third reviewer (Z.-Q. W.). The following data elements were collected from each study: author, year of publication, origin of population, test method, number of samples, origin of sample, sensitivity and specificity data, cutoff criteria and miRNA expression profiles.
We used the statistical methods recommended for diagnostic accuracy meta-analysis (20). Indexes of test accuracy were analyzed and calculated from the extracted data of each study, including sensitivity, specificity, true positive rate (TP), false positive rate (FP), true negative rate (TN), false negative rate (FN), positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR) and the area under the curve (AUC). The AUC is a comprehensive, representative test accuracy, combining sensitivity with specificity. An AUC of 1.0 (100%) demonstrates perfect ability to distinguish cases from noncases. The pooled indexes from all studies were used in random-effects models or fixed-effects (weighted with inverse variance) models. A random-effects model was chosen if heterogeneity was >50%, whereas a fixed-effects model was chosen if heterogeneity was <50% (21). Heterogeneity analysis was carried out using chi-square test and I-square test. Publication bias was evaluated using a funnel plot with the Egger’s regression test (22). All statistical analyses were performed using Meta-Disc 1.4 (XI Cochrane Colloquium, Barcelona, Spain) and Stata Version 12.0 software (Stata Corp., College Station, TX, USA). The significance level of all statistical tests was 2-sided and set at a p value <0.05.
By means of the above search strategy, a total of 268 potentially relevant articles were identified. After prudential screening of the titles, abstracts and key words, 259 articles were excluded because they were non-English articles or were reviews, letters, case reports, samples of nonblood origin or studies irrelevant to the current analysis. For the remaining 10 articles, we reviewed the full text again in detail, and 2 papers were excluded because they lacked data associated with sensitivity and specificity. Finally 12 studies from 8 articles were included in this meta-analysis (11-12-13-14-15-16-17-18). The process of study selection is shown in a flow diagram (
Flow chart showing the process of study selection.
The main characteristics of the eligible studies are listed in
Characteristics of the included studies
|Study||Author (ref.)||Year||Origin of population||Test method||Sample size||Sample origin||TP (%)||FP (%)||FN (%)||TN (%)||Cutoff criteria||MicroRNA profiles|
|FN = false negative rate; FP = false positive rate; TN = true negative rate; TP = true positive rate; qRT-PCR = quantitative real-time polymerase chain reaction; “-” indicates that the study does not mention cutoff criteria. *from the opposite arm of the precursor(miR126).|
|1||Wang et al (18)||2009||USA||qRT-PCR||47||Plasma||38.3||4.3||21.3||36.1||2-fold||miR-21, 155, 196a, 210|
|2||Morimura et al (16)||2011||Japan||qRT-PCR||66||Plasma||34.8||0||19.7||45.5||Mean||miR-18a|
|3||Liu et al (14)||2012||China||qRT-PCR||245||Plasma||49.4||1.6||6.9||42.1||-||miR-16, 196a|
|4||Liu et al (14)||2012||China||qRT-PCR||206||Plasma||61.7||1.5||5.3||31.5||-||miR-16, 196a|
|5||Liu et al (15)||2012||China||qRT-PCR||202||Serum||59.4||1.5||0||39.1||2-fold||miR-20a, 21, 24, 25, 99a, 185, 191|
|6||Liu et al (15 )||2012||China||qRT-PCR||69||Serum||46.4||5.8||7.2||40.6||2-fold||miR-20a, 21, 24, 25, 99a, 185, 191|
|7||Li et al (13)||2013||America||qRT-PCR||120||Serum||59.2||5.0||8.3||27.5||2-fold||miR-1290|
|8||Li et al (13)||2013||America||qRT-PCR||213||Serum||31.5||15.9||6.6||46.0||2-fold||miR-1290|
|9||Kawaguchi et al (11)||2013||Japan||qRT-PCR||47||plasma||61.7||8.5||2.1||27.7||Mean||miR-221|
|10||Ganepola et al (12)||2014||America||qRT-PCR||22||Plasma||45.5||4.5||4.5||45.5||-||miR-22.3p, 642b.3p, 885.5p|
|11||Schultz et al (17)||2014||Denmark||qRT-PCR||379||Whole blood||40.4||2.9||7.1||49.6||Mean||miR-145, 150, 223, 636|
|12||Schultz et al (17)||2014||Denmark||qRT-PCR||379||Whole blood||40.4||1.1||7.1||51.4||Mean||miR-26b, 34a, 122, 126*, 145, 150, 223, 505, 636, 885.5p|
QUADAS-2 is designed to assess the quality of primary diagnostic accuracy studies.
Methodological quality graph: judgments about overall methodological quality across 11 included studies, using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2).
Diagnostic accuracy assessment
Forest plots of sensitivity, specificity, PLR and NLR for miRNA expression profiles are shown in
Pooled diagnostic accuracy
|Sensitivity (95% CI)||Specificity (95% CI)||PLR (95% CI)||NLR (95% CI)||DOR (95% CI)||AUC (SEM)|
|AUC = the area under the curve; CI = confidence interval; DOR = diagnostic odds ratio; NLR = negative likelihood ratio; PC = pancreatic cancer; PLR = positive likelihood ratio.|
|PC vs. non-PC||0.87 (0.85-0.89)||0.92 (0.90-0.94)||11.18 (5.57-22.46)||0.16 (0.11-0.23)||88.98 (39.85-198.69)||0.96 (0.01)|
|Heterogeneity (p value)||64.67 (0.00)||72.16 (0.00)||89.24 (0.00)||48.45 (0.00)||48.04 (0.00)||-|
Forest plots of sensitivity (A), specificity (B), PLR (C) and NLR (D) for miRNA assays in the diagnosis of PC. Each solid circle represents an eligible study. The size of the solid circle reflects the sample size of each eligible study. Error bars represent 95% confidence interval (CI). LR = likelihood ratio.
The pooled DOR for the diagnostic value showed a significant difference between PC with non-PC (DOR = 88.98; 95% CI, 39.85-198.69). The chi-squared value for the heterogeneity test was 48.04%, so the pooled DOR for the diagnostic value was calculated using the random-effects model. The summary receiver operating characteristic (SROC) curve for the miRNA expression profiles was drawn based on a series of 2 different classification methods, including TP (sensitivity) for the vertical axis and FP (1-specificity) for the horizontal axis (
Summary receiver operating characteristic (SROC) curve for miRNA assays in the diagnosis of pancreatic cancer (PC). Each solid circle represents an eligible study. The size of solid circles reflects the sample size of each eligible study. The regression SROC curve summarizes the overall diagnostic accuracy. AUC = the area under the curve; SE = standard error.
In this meta-analysis, strict inclusion or exclusion criteria and the careful design of the data analysis were used to minimize the potential publication bias. Finally, publication bias of the included studies was assessed by funnel plots and Egger’s tests. There was no evidence for significant publication bias in the meta-analysis if the p value was not <0.05. Hence, publication bias was not found according to this funnel plot, which is almost symmetric, and Egger’s test (p = 0.324). The funnel plot is shown in
Funnel graph for the assessment of potential publication bias of the 12 included studies. The funnel graph plots the log of diagnostic odds ratio (DOR) against the standard error of the log of the DOR. Each solid circle represents an eligible study.
Although surgical resection shows promise as an effective treatment, PC remains a malignant disease with poor prognosis because it is generally asymptomatic, and there are no effective biomarkers for diagnosis at the early stage. Currently, many serum biomarkers have been reported for PC diagnosis, such as CA19-9, CEA-CAM1, MUC-1, TIMP-1, MMP-9, lectins, REG4 and K-ras mutations (23-24-25-26-27-28). Serum CA19-9 is the most widely available diagnostic biomarker, with a sensitivity of 70%-80% in PC patients but a specificity of less than 50% (29). Thus, the conventional diagnostic biomarkers in PC patients have certain limitations.
Aberrant miRNA expression profiles between PC and non-PC were previously reported in several studies, with miRNAs derived from tumor tissues and cell lines (8, 9, 30-31-32). A study of measurement of miRNA abundance in PC and chronic pancreatitis (CP) was conducted by Bauer et al, enunciating the suggestion that miRNAs derived from tissues or blood could have an immediate diagnostic value for the evaluation of tumor reoccurrence in patients who have undergone curative surgical resection, and proposing the hypothesis that inflammation, such as CP, could be required as 1 step toward the development of PC (8). However, although it can be used as a new tumor marker, tumor tissue miRNA is obtained by invasive surgery or biopsy, which still can not meet the needs of diagnosis. In contrast, blood-based miRNAs are stable, easily detectable and significantly disease-specific (33), and have been reported as potential biomarkers for early diagnosis of lung and liver cancer (34, 35). Similarly, this present meta-analysis showed the diagnostic value of blood-based miRNAs in PC.
The present meta-analysis included 12 studies, each of which obtained miRNA quantitative expression data using qRT-PCR. Ultimately, there are a total of 26 miRNAs (miRNA-16, 18a, 20a, 21, 22-3p, 24, 25, 26b, 34a, 99a, 122, 126*, 145, 150, 155, 185, 191, 196a, 210, 221, 223, 505, 636, 642b-3p, 885-5p, 1290) included in this analysis, some of which were also compared with CA19-9. From the summary results, the pooled sensitivity and specificity of miRNA biomarkers were 0.87 and 0.92, indicating a better diagnostic result compared with CA19-9 alone. The PLR and NLR are relatively independent, clinically meaningful indicators to assess diagnostic test effects. When the PLR value was >10 or the NLR value was <0.1, possibility of a disease diagnosis was markedly increased. Hence, the pooled PLR value was 11.18, which indicates PC patients can have an 11 times higher possibility of a positive miRNA test compared with non-PC. Similarly, the NLR value was 0.16, which suggests that the likelihood of a patient having PC is only 16% if the miRNA test is negative.
As an evaluation index of overall performance in diagnostic testing, the DOR has a significant advantage, which is not affected by the morbidity, and also takes into account the sensitivity and specificity. Diagnostic accuracy is better with a higher value of the DOR. For instance, a diagnostic test has almost no ability to distinguish case from noncase if the DOR of the test is 1.0. The pooled DOR in this meta-analysis was 88.98, indicating a high diagnostic accuracy.
AUC is also a very important indicator of diagnostic accuracy similar to DOR. The area under the SROC curve was 0.96, which demonstrates potential diagnostic capability. Thus, these indicators reflect the promising clinical value of blood-based miRNA as a diagnostic biomarkers.
The origins of blood-based samples include plasma, serum and whole blood. There is almost no difference in miRNA extraction methods between plasma- and serum-based samples, while miRNAs extracted from whole blood collected in PAXgene blood RNA tubes avoid the interference caused by centrifugation (17, 36, 37). Liu et al have identified a 7 miRNAs–based serum panel for accurately discriminating PC cases from cancer-free controls (15). Another study has indicated that the combination of plasma miRNA-16, miRNA-196a and CA19-9 was more effective for PC diagnosis (14). The diagnostic panels of miRNA expression in whole blood were identified by Schultz et al, with the potential to distinguish patients with PC from healthy controls, with sensitivity and specificity superior to those of CA19-9 (17). As a risk factor for PC, CP has similar clinical symptoms, making it difficult to distinguish between the 2 diseases (38, 39). Several studies have shown that blood-based miRNAs could serve as a specific biomarker to distinguish PC from CP (14, 15, 17, 40). Furthermore, to achieve an early diagnosis, Schultz et al used samples from patients with only stage I/II PC and observed that a panel of miRNA expression profiles also had significant diagnostic value (17). Although miRNAs were derived from blood-based samples, their expression profiles have varied in different studies, which means they could not form a unified diagnostic panel.
Several limitations have to be considered in the current meta-analysis. Firstly, the heterogeneity of the population was increased due to the difference in the characteristics of patients (age, race, country, etc.), the cutoff criteria and miRNA expression profiles. Secondly, significant differences in the number of samples in different studies may have had some impact on the results. Finally, to further validate a novel miRNA biomarkers in the diagnosis of PC, we need more multicenter prospective clinical studies.
In summary, the present meta-analysis, representing a quantified synthesis of published studies, has shown that blood-based miRNA expression profiles have the potential to discriminate PC patients from non-PC patients. As a novel method, blood-based miRNA biomarkers have diagnostic advantages compared with other indicators. However, more validation studies should be conducted before these can be implemented as part of the routine clinical diagnosis of PC.
We thank all clinical investigators who were involved in this meta-analysis.
- Meng, Qing-Cai [PubMed] [Google Scholar]
- Qiu, Wang-Wang [PubMed] [Google Scholar]
- Wen, Zi-Qiang [PubMed] [Google Scholar]
- Wang, Hong-Cheng [PubMed] [Google Scholar]
- Shan, Ze-Zhi [PubMed] [Google Scholar]
- Yuan, Zhou [PubMed] [Google Scholar]
- Huang, Xin-Yu [PubMed] [Google Scholar] , * Corresponding Author (firstname.lastname@example.org)
Department of General Surgery, Sixth People’s Hospital Affiliated With Shanghai Jiao Tong University, Shanghai - PR China