Tumor Biol. (2014) 35:10789–10798 DOI 10.1007/s13277-014-2382-2
RESEARCH ARTICLE
Clinical significance of microRNA expressions in diagnosing uterine cancer and predicting lymph node metastasis Changlong Hou & Guosheng Tan & Shiting Feng
Received: 24 June 2014 / Accepted: 23 July 2014 / Published online: 31 July 2014 # International Society of Oncology and BioMarkers (ISOBM) 2014
Abstract Recently, accumulating lines of evidence have demonstrated the association between microRNA (miRNAs) expression and uterine cancer, indicating that they may serve as promising novel biomarkers for uterine cancer. Therefore, we conducted this study to systematically evaluate the diagnostic accuracy of miRNAs in discriminating the uterine cancer patients from controls and further to determine their diagnostic values in lymph node metastasis (LNM) prediction. The pooled sensitivity, specificity, and other parameters, together with summary receiver operating characteristic (SROC) curve were used to assess the overall test performance. All statistical analyses were conducted using STATA 12.0 software. A total of nine articles were included in this meta-analysis. As for the accuracy of miRNAs in differentiating uterine cancer from controls, the pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under curve (AUC) were 0.84, 0.83, 4.8, 0.19, 25, and 0.90, respectively. As for the diagnostic accuracy of miRNAs in differentiating patients with LNM from those without LNM, the pooled sensitivity, specificity, PLR, NLR, DOR, and AUC
Electronic supplementary material The online version of this article (doi:10.1007/s13277-014-2382-2) contains supplementary material, which is available to authorized users. C. Hou Department of Interventional Radiology, The Affiliated Provincial Hospital of Anhui Medical University, 230001 Hefei, China G. Tan (*) Department of Interventional Radiology, The First Affiliated Hospital of Sun Yat-sen University, Guangzhou 510080, China e-mail: [email protected] S. Feng Department of Radiology, The First Affiliated Hospital of Sun Yat-sen University, Guangzhou 510080, China
were 0.75, 0.78, 3.5, 0.32, 011, and 0.83, respectively. In addition, subgroup analyses based on miRNA profiles suggested that multiple-miRNA assay displayed much better accuracy than single-miRNA assay, with an excellent AUC of 0.98 (92 % sensitivity and 96 % specificity). The high accuracy of multiple-miRNA assay, together with the application of miRNAs in LNM prediction, suggested that miRNAs may serve as non-invasive diagnostic markers of uterine cancer and further improve the comprehensive management of patients with uterine cancer. However, further larger studies are needed to confirm our findings. Keywords MicroRNAs . Uterine cancer . Lymph node metastasis . Diagnosis . Accuracy . Meta-analysis
Introduction Uterine cancer, which consists of cervical cancer and endometrial cancer, has a high incidence and mortality among women worldwide [1]. Due to the advanced healthcare systems, widespread and uptake of screening for the prevention, incidence, and mortality rates of uterine cancer for women in more developed areas were 2.5 and 0.6 %, lower than those in less developed areas with 2.5 and 1.3 % [2, 3]. However, an estimated global incidence of cervical cancer is still up to 529,800 new cases and approximately 275,100 deaths per year, and endometrial cancer results in approximately 300,000 deaths annually [3, 4]. What is worse, uterine cancer is characterized by a high incidence of distant metastases, and various sites including the liver, lung, brain, bone, pancreas, and lymph nodes can be targeted via bloodstream, which impedes the treatment in most cases [5–7]. For cervical cancer patients, the 5-year survival rate with negative lymph nodes is approximately 80–95 %, versus only 50–65 % in patients with lymph node metastasis (LNM) [8–10]. Even though without
10790
LNM, the 5-year survival rate remains low when it comes to the advanced stages, with only 10–29 % [11]. Thus, improvement in uterine cancer diagnosis at early stage, together with the management of LNM, is the main challenge nowadays. Currently, surgery, imaging, and pathological analysis are the main methods to identify uterine cancer [9]. However, these techniques have their limitations. As to the radical surgery or (chemo)radiotherapy, they can control majority of cervical cancer patients at early stage. However, recurrence or failure to survive still occurs after surgically treatment among patients with LNM. Additionally, the imaging methods, such as CT, MRI, and PET-CT, all have low sensitivity and specificity. Small cell carcinoma of the cervix antigen (SCC-Ag) is currently the most widely used cervical cancer marker with sensitivity of 61 % and specificity of 70 % in LNM prediction [12]. Unfortunately, although high SCC-Ag level may indicate LNM or extra-cervical spread, normal level cannot exclude the presence of LNM [10, 13–15]. Other cervical cancer markers, such as CA125, tissue polypeptide antigen (TPA), tissue polypeptide-specific antigen (TPS), and carcinoembryonic antigen (CEA), are not effective enough to predict LNM in cervical cancer patients [16, 17]. Hence, with the increase in the incidence of uterine cancer, due to the helplessness of existing diagnostic tools and biomarkers and low survival rate of patients with LNM, there is a great need to identify novel biomarkers with high accuracy for uterine cancer and the presence of LNM. MicroRNAs (miRNAs) are small non-coding RNAs of 18– 25 nucleotides that implicated in tumor development [18]. The discovery of miRNAs put a new sight in molecular diagnosis of human cancers [19, 20]. In general, each miRNA can target hundreds of genes, thereby inactivate tumor suppressor genes or activate oncogenes in tumorigenesis [21–24]. Further, it has been reported that aberrant expression of miRNAs is linked to tumor clinical features, which makes miRNAs potential valuable biomarkers [25–27]. For example, Lee et al. has found six up-regulation miRNAs (miR-21, miR-182, miR-183, miR200a, miR-200c, and miR-205) in endometrial cancer tissues compared with healthy individual tissues [28]. Moreover, circulating miRNA levels have been specifically linked to LNM, and Chen et al. first reported that high serum miR-21 levels were associated with LNM in cervical cancer, suggesting that miRNAs are useful in predicting the presence of LNM [29]. What is more, miRNAs have similar signatures between men and women, so as to individuals of different age [30]. Combined with characteristics of highly stability and abundant quantity in body fluids or tissues [31–34], miRNAs have a promising future in uterine detection and predicting LNM. In the past decades, numbers of studies have been focused on the diagnostic performance of miRNAs in uterine cancer, as well as in predicting LNM. However, individual studies have their limitations, and utilization of sample sets with various histological types and grade, different platforms, and
Tumor Biol. (2014) 35:10789–10798
different normalization strategies resulted in their conflicting conclusions. As only a few miRNA molecules have been consistently reported by more than two research groups and there are still many questions unknown [35–37], we conducted this meta-analysis to estimate the overall diagnosis value of miRNAs in uterine cancer.
Methods Literature search Several international databases (PubMed, Web of Science, the Cochrane Library, and Embase) and two Chinese databases (Wan Fang DATA and Chinese National Knowledge Infrastructure [CNKI]) were searched for relevant publications up to May 24, 2014. We used (“uterine cancer” or “cervical cancer” or “endometrial cancer” or “lymph node metastasis”) and (“microRNA” or “miRNA”) and (“serum” or “plasma” or “blood” or “tissue”) and (“diagnostic” or “diagnosis” or “ROC curve” or “sensitivity” or “specificity”) as keywords without language restrictions. Additional records were also identified through manual search.
Inclusion and exclusion criteria Studies included in this meta-analysis should meet the following selection criteria: (1) clinical studies evaluating the diagnostic performance of miRNAs in uterine cancer; (2) studies about human; and (3) studies providing information for truepositive (TP), false-positive (FP), true-negative (TN), and false-negative (FN) (or the data of sensitivity and specificity). Studies were excluded if they had any of the following items: (1) duplicate publications; (2) reviews, letters, meta-analysis, or meeting reports; and (3) studies without sufficient data.
Data extraction and quality assessment Relevant data were extracted from each included study as follows: basic features (the first author, publish year, and country), descriptions about study population (ethnicity and number of the group), diagnostic test data (sensitivity, specificity, TP, FP, TN, and FN), and specimens and miRNA profiles. Based on the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool [38], the quality of each study was scored independently by two reviewers. The QUADAS-2 tool consists of four domains (patient’s selection, the index test, the reference standard, and flow and timing) to analyze the risk of bias and applicability concerns of the included studies.
Tumor Biol. (2014) 35:10789–10798
Statistical analysis All statistical analyses were performed using STATA 12.0 statistical software. Sensitivity and specificity were presented as forest plots to calculate the inconsistency between studies. The I2, which describes the percentage of total variation across studies that is due to heterogeneity rather than chance, was higher than 50 %; thereby, the random effect model was chosen for our analysis. Additionally, sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were summarized to evaluate the diagnostic accuracy, so as to the summary receiver operating characteristic (SROC) curve and the area under curve (AUC) value [39]. The subgroup and metaregression analyses were performed to explore the heterogeneity between studies. What is more, we conducted influence analysis and outlier detection test to confirm the robustness for our work. For publication bias, all eligible studies were assessed by Deek’s test; the P value with less than 0.10 shows a result of statistical significance.
Results Characteristics and quality of the included studies The search strategy used in this study is showed in Fig. 1. The literature search yielded 260 records through database searching and 7 through a manual search, of which, 83 duplicate records among databases were excluded. Of the remaining 184 records, 161 studies were excluded for the departure from inclusion criteria: 124 of them were reviews or meta-analyses, 14 were not human studies, and 23 were not relevant to uterine cancer. Then, 23 records were fulltext assessed for eligibility, and 14 articles were excluded because they were not relevant to diagnosis (n = 7), not relevant to miRNAs (n = 3), and without sufficient data (n=4). Finally, nine articles were included in this metaanalysis, among which, four focused on the diagnosis value of miRNAs between uterine cancer patients and healthy controls [40–43] and the other five explored the performance in cancer patients with LNM or without LNM [12, 29, 40, 42, 44]. All the included studies published from 2012 to 2014 and used quantitative reverse transcription-PCR (qRTPCR) method to measure the expression of miRNAs. A total of 59 studies from four articles involving 395 participants (738 uterine cancer patients and 576 healthy controls) assessed the accuracy of miRNAs for the diagnosis of uterine cancer from healthy individuals. Among
10791
the 59 diagnostic studies, 50 studies focused on the performance of a single miRNA, while the other nine studies evaluated diagnosis value of multiple miRNAs for detecting uterine cancer. The categories of specimens included blood (n=15) and tissue (n=44). In the patients with LNM vs. those without LNM group, 26 studies from five articles involving 316 participants (128 patients with LNM and 188 without LNM) were analyzed. The categories of specimens included blood (n=11) and tissue (n=15). QUADAS-2 was used to assess the quality of included studies. All nine articles got scores ≥4, four of which achieved scores higher than 5, indicating a relatively high quality of the included studies and high reliability of this meta-analysis. The main clinical characteristics of the included studies, along with QUADAS2 scores, are presented in Table 1. Diagnostic performance of miRNAs in discriminating uterine cancer from controls Table 2 presents the summary results of the diagnostic performance of miRNAs in differentiating uterine cancer from healthy controls. The pooled sensitivity was 0.84 (95 % confidence interval (CI) 0.81–0.87), and specificity was 0.83 (95 % CI 0.78–0.86). The overall PLR was 4.8 (95 % CI 3.8–6.2), NLR was 0.19 (95 % CI 0.16–0.23), and pooled DOR was 25 (95 % CI 18–36), respectively. The AUC value, which is a parameter to estimate the overall accuracy, was 0.90 (95 % CI 0.87–0.92). Subgroup analyses based on miRNA profiles and specimen types used in the studies were also conducted. The subgroup analyses results are also listed in Table 2. As to the miRNA profiling subgroups, compared with singlemiRNA assay, the summary sensitivity increased from 0.82 to 0.92 when using multiple-miRNA assay, specificity increased from 0.78 to 0.96, PLR increased from 0.78 to 0.96, and NLR decreased from 0.23 to 0.09. What is more, the DOR made a significant improvement from 18 to 259, and AUC increased from a moderate accuracy of 0.87 (Fig. 2a) to a high accuracy of 0.98 (Fig. 2b), suggesting an outstanding diagnostic value of multiplemiRNA assay in discriminating uterine cancer from healthy controls. As to the specimen type subgroups, the pooled sensitivity, specificity, PLR, NLR, and DOR of tissue-based test were 0.86, 0.79, 4.1, 0.17, and 24, respectively. The AUC of the tissue-based test was 0.89 (95 % CI 0.86–0.91) (Fig. 2c). As for blood-based test, the pooled sensitivity, specificity, PLR, NLR, and DOR were 0.74, 0.91, 8.5, 0.28, and 30, respectively. The AUC of blood-based test was 0.92 (95 % CI 0.89– 0.94) (Fig. 2d). The meta-regression in Fig. S1 was conducted to explore possible sources of heterogeneity across the studies. The
Fig. 1 Flow chart of study selection based on the inclusion and exclusion criteria
Tumor Biol. (2014) 35:10789–10798 Identification
10792 Records identified through database searching (N = 260)
Additional records identified through a manual search (N = 7)
Records reviewed for duplicates (N = 267)
Screening
(n = 83) Duplicate records were excluded
Records after duplicates removed (N = 184) 161 studies were excluded: (n = 124) Reviews, meta-analysis (n = 14) Not human studies (n = 23) Not relevant to uterine cancer
Full-text articles assessed for eligibility Eligibility
(N = 23) 14 studies were excluded: (n = 7) Not relevant to diagnosis (n = 3) Not relevant to miRNAs (n = 4) Without sufficient data
Included
Nine articles were included in this meta-analysis (N = 9)
Cancer vs. Control
LNM vs. non-LNM
(N = 4)
(N = 5)
results suggested that the miRNA profiles (P
RESEARCH ARTICLE
Clinical significance of microRNA expressions in diagnosing uterine cancer and predicting lymph node metastasis Changlong Hou & Guosheng Tan & Shiting Feng
Received: 24 June 2014 / Accepted: 23 July 2014 / Published online: 31 July 2014 # International Society of Oncology and BioMarkers (ISOBM) 2014
Abstract Recently, accumulating lines of evidence have demonstrated the association between microRNA (miRNAs) expression and uterine cancer, indicating that they may serve as promising novel biomarkers for uterine cancer. Therefore, we conducted this study to systematically evaluate the diagnostic accuracy of miRNAs in discriminating the uterine cancer patients from controls and further to determine their diagnostic values in lymph node metastasis (LNM) prediction. The pooled sensitivity, specificity, and other parameters, together with summary receiver operating characteristic (SROC) curve were used to assess the overall test performance. All statistical analyses were conducted using STATA 12.0 software. A total of nine articles were included in this meta-analysis. As for the accuracy of miRNAs in differentiating uterine cancer from controls, the pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under curve (AUC) were 0.84, 0.83, 4.8, 0.19, 25, and 0.90, respectively. As for the diagnostic accuracy of miRNAs in differentiating patients with LNM from those without LNM, the pooled sensitivity, specificity, PLR, NLR, DOR, and AUC
Electronic supplementary material The online version of this article (doi:10.1007/s13277-014-2382-2) contains supplementary material, which is available to authorized users. C. Hou Department of Interventional Radiology, The Affiliated Provincial Hospital of Anhui Medical University, 230001 Hefei, China G. Tan (*) Department of Interventional Radiology, The First Affiliated Hospital of Sun Yat-sen University, Guangzhou 510080, China e-mail: [email protected] S. Feng Department of Radiology, The First Affiliated Hospital of Sun Yat-sen University, Guangzhou 510080, China
were 0.75, 0.78, 3.5, 0.32, 011, and 0.83, respectively. In addition, subgroup analyses based on miRNA profiles suggested that multiple-miRNA assay displayed much better accuracy than single-miRNA assay, with an excellent AUC of 0.98 (92 % sensitivity and 96 % specificity). The high accuracy of multiple-miRNA assay, together with the application of miRNAs in LNM prediction, suggested that miRNAs may serve as non-invasive diagnostic markers of uterine cancer and further improve the comprehensive management of patients with uterine cancer. However, further larger studies are needed to confirm our findings. Keywords MicroRNAs . Uterine cancer . Lymph node metastasis . Diagnosis . Accuracy . Meta-analysis
Introduction Uterine cancer, which consists of cervical cancer and endometrial cancer, has a high incidence and mortality among women worldwide [1]. Due to the advanced healthcare systems, widespread and uptake of screening for the prevention, incidence, and mortality rates of uterine cancer for women in more developed areas were 2.5 and 0.6 %, lower than those in less developed areas with 2.5 and 1.3 % [2, 3]. However, an estimated global incidence of cervical cancer is still up to 529,800 new cases and approximately 275,100 deaths per year, and endometrial cancer results in approximately 300,000 deaths annually [3, 4]. What is worse, uterine cancer is characterized by a high incidence of distant metastases, and various sites including the liver, lung, brain, bone, pancreas, and lymph nodes can be targeted via bloodstream, which impedes the treatment in most cases [5–7]. For cervical cancer patients, the 5-year survival rate with negative lymph nodes is approximately 80–95 %, versus only 50–65 % in patients with lymph node metastasis (LNM) [8–10]. Even though without
10790
LNM, the 5-year survival rate remains low when it comes to the advanced stages, with only 10–29 % [11]. Thus, improvement in uterine cancer diagnosis at early stage, together with the management of LNM, is the main challenge nowadays. Currently, surgery, imaging, and pathological analysis are the main methods to identify uterine cancer [9]. However, these techniques have their limitations. As to the radical surgery or (chemo)radiotherapy, they can control majority of cervical cancer patients at early stage. However, recurrence or failure to survive still occurs after surgically treatment among patients with LNM. Additionally, the imaging methods, such as CT, MRI, and PET-CT, all have low sensitivity and specificity. Small cell carcinoma of the cervix antigen (SCC-Ag) is currently the most widely used cervical cancer marker with sensitivity of 61 % and specificity of 70 % in LNM prediction [12]. Unfortunately, although high SCC-Ag level may indicate LNM or extra-cervical spread, normal level cannot exclude the presence of LNM [10, 13–15]. Other cervical cancer markers, such as CA125, tissue polypeptide antigen (TPA), tissue polypeptide-specific antigen (TPS), and carcinoembryonic antigen (CEA), are not effective enough to predict LNM in cervical cancer patients [16, 17]. Hence, with the increase in the incidence of uterine cancer, due to the helplessness of existing diagnostic tools and biomarkers and low survival rate of patients with LNM, there is a great need to identify novel biomarkers with high accuracy for uterine cancer and the presence of LNM. MicroRNAs (miRNAs) are small non-coding RNAs of 18– 25 nucleotides that implicated in tumor development [18]. The discovery of miRNAs put a new sight in molecular diagnosis of human cancers [19, 20]. In general, each miRNA can target hundreds of genes, thereby inactivate tumor suppressor genes or activate oncogenes in tumorigenesis [21–24]. Further, it has been reported that aberrant expression of miRNAs is linked to tumor clinical features, which makes miRNAs potential valuable biomarkers [25–27]. For example, Lee et al. has found six up-regulation miRNAs (miR-21, miR-182, miR-183, miR200a, miR-200c, and miR-205) in endometrial cancer tissues compared with healthy individual tissues [28]. Moreover, circulating miRNA levels have been specifically linked to LNM, and Chen et al. first reported that high serum miR-21 levels were associated with LNM in cervical cancer, suggesting that miRNAs are useful in predicting the presence of LNM [29]. What is more, miRNAs have similar signatures between men and women, so as to individuals of different age [30]. Combined with characteristics of highly stability and abundant quantity in body fluids or tissues [31–34], miRNAs have a promising future in uterine detection and predicting LNM. In the past decades, numbers of studies have been focused on the diagnostic performance of miRNAs in uterine cancer, as well as in predicting LNM. However, individual studies have their limitations, and utilization of sample sets with various histological types and grade, different platforms, and
Tumor Biol. (2014) 35:10789–10798
different normalization strategies resulted in their conflicting conclusions. As only a few miRNA molecules have been consistently reported by more than two research groups and there are still many questions unknown [35–37], we conducted this meta-analysis to estimate the overall diagnosis value of miRNAs in uterine cancer.
Methods Literature search Several international databases (PubMed, Web of Science, the Cochrane Library, and Embase) and two Chinese databases (Wan Fang DATA and Chinese National Knowledge Infrastructure [CNKI]) were searched for relevant publications up to May 24, 2014. We used (“uterine cancer” or “cervical cancer” or “endometrial cancer” or “lymph node metastasis”) and (“microRNA” or “miRNA”) and (“serum” or “plasma” or “blood” or “tissue”) and (“diagnostic” or “diagnosis” or “ROC curve” or “sensitivity” or “specificity”) as keywords without language restrictions. Additional records were also identified through manual search.
Inclusion and exclusion criteria Studies included in this meta-analysis should meet the following selection criteria: (1) clinical studies evaluating the diagnostic performance of miRNAs in uterine cancer; (2) studies about human; and (3) studies providing information for truepositive (TP), false-positive (FP), true-negative (TN), and false-negative (FN) (or the data of sensitivity and specificity). Studies were excluded if they had any of the following items: (1) duplicate publications; (2) reviews, letters, meta-analysis, or meeting reports; and (3) studies without sufficient data.
Data extraction and quality assessment Relevant data were extracted from each included study as follows: basic features (the first author, publish year, and country), descriptions about study population (ethnicity and number of the group), diagnostic test data (sensitivity, specificity, TP, FP, TN, and FN), and specimens and miRNA profiles. Based on the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool [38], the quality of each study was scored independently by two reviewers. The QUADAS-2 tool consists of four domains (patient’s selection, the index test, the reference standard, and flow and timing) to analyze the risk of bias and applicability concerns of the included studies.
Tumor Biol. (2014) 35:10789–10798
Statistical analysis All statistical analyses were performed using STATA 12.0 statistical software. Sensitivity and specificity were presented as forest plots to calculate the inconsistency between studies. The I2, which describes the percentage of total variation across studies that is due to heterogeneity rather than chance, was higher than 50 %; thereby, the random effect model was chosen for our analysis. Additionally, sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were summarized to evaluate the diagnostic accuracy, so as to the summary receiver operating characteristic (SROC) curve and the area under curve (AUC) value [39]. The subgroup and metaregression analyses were performed to explore the heterogeneity between studies. What is more, we conducted influence analysis and outlier detection test to confirm the robustness for our work. For publication bias, all eligible studies were assessed by Deek’s test; the P value with less than 0.10 shows a result of statistical significance.
Results Characteristics and quality of the included studies The search strategy used in this study is showed in Fig. 1. The literature search yielded 260 records through database searching and 7 through a manual search, of which, 83 duplicate records among databases were excluded. Of the remaining 184 records, 161 studies were excluded for the departure from inclusion criteria: 124 of them were reviews or meta-analyses, 14 were not human studies, and 23 were not relevant to uterine cancer. Then, 23 records were fulltext assessed for eligibility, and 14 articles were excluded because they were not relevant to diagnosis (n = 7), not relevant to miRNAs (n = 3), and without sufficient data (n=4). Finally, nine articles were included in this metaanalysis, among which, four focused on the diagnosis value of miRNAs between uterine cancer patients and healthy controls [40–43] and the other five explored the performance in cancer patients with LNM or without LNM [12, 29, 40, 42, 44]. All the included studies published from 2012 to 2014 and used quantitative reverse transcription-PCR (qRTPCR) method to measure the expression of miRNAs. A total of 59 studies from four articles involving 395 participants (738 uterine cancer patients and 576 healthy controls) assessed the accuracy of miRNAs for the diagnosis of uterine cancer from healthy individuals. Among
10791
the 59 diagnostic studies, 50 studies focused on the performance of a single miRNA, while the other nine studies evaluated diagnosis value of multiple miRNAs for detecting uterine cancer. The categories of specimens included blood (n=15) and tissue (n=44). In the patients with LNM vs. those without LNM group, 26 studies from five articles involving 316 participants (128 patients with LNM and 188 without LNM) were analyzed. The categories of specimens included blood (n=11) and tissue (n=15). QUADAS-2 was used to assess the quality of included studies. All nine articles got scores ≥4, four of which achieved scores higher than 5, indicating a relatively high quality of the included studies and high reliability of this meta-analysis. The main clinical characteristics of the included studies, along with QUADAS2 scores, are presented in Table 1. Diagnostic performance of miRNAs in discriminating uterine cancer from controls Table 2 presents the summary results of the diagnostic performance of miRNAs in differentiating uterine cancer from healthy controls. The pooled sensitivity was 0.84 (95 % confidence interval (CI) 0.81–0.87), and specificity was 0.83 (95 % CI 0.78–0.86). The overall PLR was 4.8 (95 % CI 3.8–6.2), NLR was 0.19 (95 % CI 0.16–0.23), and pooled DOR was 25 (95 % CI 18–36), respectively. The AUC value, which is a parameter to estimate the overall accuracy, was 0.90 (95 % CI 0.87–0.92). Subgroup analyses based on miRNA profiles and specimen types used in the studies were also conducted. The subgroup analyses results are also listed in Table 2. As to the miRNA profiling subgroups, compared with singlemiRNA assay, the summary sensitivity increased from 0.82 to 0.92 when using multiple-miRNA assay, specificity increased from 0.78 to 0.96, PLR increased from 0.78 to 0.96, and NLR decreased from 0.23 to 0.09. What is more, the DOR made a significant improvement from 18 to 259, and AUC increased from a moderate accuracy of 0.87 (Fig. 2a) to a high accuracy of 0.98 (Fig. 2b), suggesting an outstanding diagnostic value of multiplemiRNA assay in discriminating uterine cancer from healthy controls. As to the specimen type subgroups, the pooled sensitivity, specificity, PLR, NLR, and DOR of tissue-based test were 0.86, 0.79, 4.1, 0.17, and 24, respectively. The AUC of the tissue-based test was 0.89 (95 % CI 0.86–0.91) (Fig. 2c). As for blood-based test, the pooled sensitivity, specificity, PLR, NLR, and DOR were 0.74, 0.91, 8.5, 0.28, and 30, respectively. The AUC of blood-based test was 0.92 (95 % CI 0.89– 0.94) (Fig. 2d). The meta-regression in Fig. S1 was conducted to explore possible sources of heterogeneity across the studies. The
Fig. 1 Flow chart of study selection based on the inclusion and exclusion criteria
Tumor Biol. (2014) 35:10789–10798 Identification
10792 Records identified through database searching (N = 260)
Additional records identified through a manual search (N = 7)
Records reviewed for duplicates (N = 267)
Screening
(n = 83) Duplicate records were excluded
Records after duplicates removed (N = 184) 161 studies were excluded: (n = 124) Reviews, meta-analysis (n = 14) Not human studies (n = 23) Not relevant to uterine cancer
Full-text articles assessed for eligibility Eligibility
(N = 23) 14 studies were excluded: (n = 7) Not relevant to diagnosis (n = 3) Not relevant to miRNAs (n = 4) Without sufficient data
Included
Nine articles were included in this meta-analysis (N = 9)
Cancer vs. Control
LNM vs. non-LNM
(N = 4)
(N = 5)
results suggested that the miRNA profiles (P
