Tumor Biol. DOI 10.1007/s13277-014-2209-1
RESEARCH ARTICLE
Diagnostic value of microRNAs in discriminating malignant thyroid nodules from benign ones on fine-needle aspiration samples Yang Zhang & Qi Zhong & Xiaohong Chen & Jugao Fang & Zhigang Huang
Received: 1 May 2014 / Accepted: 6 June 2014 # International Society of Oncology and BioMarkers (ISOBM) 2014
Abstract Many studies have suggested that microRNAs (miRNAs) might serve as novel diagnostic indicators of thyroid cancer (TC). However, inconsistent results have also been reported. This meta-analysis was conducted to assess the diagnostic value of miRNAs in discriminating malignant thyroid nodules from benign ones on fine-needle aspiration samples. A systematic literature search for relevant literature published up to April 5, 2014 was conducted in PubMed, Embase, Chinese National Knowledge Infrastructure (CNKI), and Chinese Biological Medicine (CBM) databases. Data from different studies were pooled to estimate the summary sensitivity (SEN), specificity (SPE), positive likelihood ratios (PLR), negative likelihood ratios (NLR), diagnostic odds ratio (DOR), using the random-effect model. Summary receiver operator characteristic curves (SROCs) were plotted and areas under the SROC curve (AUC) were calculated to evaluate the overall test performance. Between-study heterogeneity was tested using the Q tests and the I2 statistics. Potential sources of heterogeneity were analyzed through subgroup analyses and meta-regression. Deeks’ funnel plot asymmetry test was performed to evaluate publication bias. All analyses were performed using STATA 12.0 software. Eighteen studies from 7 articles, including 543 patients with malignant thyroid nodules (n=266) and benign ones (n=277), were included in this meta-analysis. The pooled SEN was 0.77 (95 % CI: 0.70–0.83), SPN was 0.75 (95 % CI: 0.68–0.81), PLR was 3.1 (95 % CI: 2.4–4.0), NLR was 0.30 (95 % CI: 0.23–0.39), DOR was 10 (95 % CI: 7–16), and AUC was 0.83 (95 %CI: 0.79–0.86). Subgroup analyses indicated that Y. Zhang : Q. Zhong : X. Chen : J. Fang : Z. Huang (*) Department of Otolaryngology-Head and Neck Surgery, Beijing Tongren Hospital, Key Laboratory of Otolaryngology Head and Neck Surgery, Capital Medical University, No.1 Dongjiaominxiang Street, Dongcheng District, Beijing 100730, China e-mail: [email protected]
multiple miRNAs assays showed a higher diagnostic accuracy than single miRNA assays. In conclusion, this meta-analysis suggests that miRNAs analysis can significantly improve diagnostic accuracy for differentiating malignant thyroid nodules from benign indeterminate ones on fine-needle aspiration (FNA) samples. With further confirmation, multiple miRNAs assays may play a critical role as a complement to fine-needle aspiration biopsy (FNAB). Keywords MicroRNAs . Thyroid cancer . Fine-needle aspiration . Diagnostic value . Meta-analysis
Introduction Thyroid cancer (TC) is the most common endocrine malignancy (95 % of all endocrine malignancies) and its incidence has increased by 2.3-fold in the last three decades [1]. Fortunately, TC only accounts for 1 % of all malignancies worldwide and the majority of patients that are found to have thyroid malignancies have well-differentiated carcinoma associated with favorable prognosis [2, 3]. However, a vexing diagnostic problem still exists. TC typically presents as a thyroid nodule, which are very common and are diagnosed in over 7 % of the adult population [4]. However, only 5 % of thyroid nodule cases are malignant and most are benign [5]. Therefore, preoperative differentiation between malignant thyroid nodules and benign ones is imperative. Currently, fine-needle aspiration biopsy (FNAB) is the gold standard diagnostic method for thyroid nodules with a relatively high sensitivity and specificity (65–98 % and 72– 100 %, respectively) [6]. Nevertheless, up to 10–25 % of the detections are interpreted as indeterminate without a definitive cytological diagnosis [7, 8]. Further, as only 20–30 % of indeterminate nodules are malignant, up to 80 % of patients
Tumor Biol.
undergo unnecessary thyroid resections, which harbor risks like hypoparathyroidism, recurrent laryngeal nerve injury, and the need for lifelong thyroid replacement [9]. Therefore, it is of utter importance to develop non-invasive and accurate method to use along with FNAB for distinguishing between malignant thyroid nodules and benign ones. However, considering that possible improvement in the accuracy of standard cytology analysis is very limited, the incorporation of novel biomarkers for improving FNAB accuracy seems to be more feasible. In the field of molecular marker research, several studies have been conducted to evaluate the potential of microRNAs (miRNAs) as diagnostic biomarkers for TC [10, 11]. miRNAs are a family of small noncoding single-stranded RNAs (18–24 nucleotides in length) [12]. They are an abundant class of gene regulatory molecules that modulate the expression of multiple protein-coding genes. Previous researches clearly indicate that they play a critical role in a wide variety of physiologic cellular processes, including differentiation, proliferation, apoptosis, and tumorigenesis [13–15]. Moreover, miRNAs are remarkably stable in vitro and can be extracted from the residual cells obtained by FNAB, which set the foundation for miRNAs as clinical biomarkers in the diagnosis of TC. Emerging evidence has shown the feasibility of testing dysregulated miRNAs in FNAB samples to improve the diagnostic accuracy for indeterminate nodules. Sheu et al. studied the diagnostic utility of miR-146b, miR-181b, miR-21, miR-221, and miR-222 in a follicular variant of papillary thyroid cancer (FPTC) and follicular thyroid carcinoma (FTC) and found that all five miRNAs could distinguish common variants of papillary thyroid cancer from follicular adenoma (FA) and multinodular goiter [16]. In another study, Weber et al. found two miRNAs (miR-197 and miR-346) were significantly overexpressed in FTC tissues compared with their benign counterparts [11]. Several recent studies have also suggested the possible utility of miRNAs in improving TC diagnostic accuracy. However, these studies have arrived at inconsistent results. Shen et al. developed another set of 4 miRNAs (miR-146b, miR-221, miR-187, and miR30d) and suggested a high sensitivity (100.0 %) but an unexpectedly low specificity (29.0 %) [17]. Whereas Keutgen et al. reported overall good diagnostic characteristics: 86.0 % sensitivity and 85 % specificity for a set of four miRNAs (miR-328, miR-222, miR-21, and miR-197) [18]. Interestingly, the study by Mazeh et al. on a single miRNA (miR-221) also yields an excellent diagnostic accuracy with 95.0 % sensitivity and 100.0 % specificity [19]. Differences in types of patients, study designs, sample sizes, miRNAs profiling and sample types may be the sources of these studies’ inconsistent conclusions. In order to further explore the clinical applicability of miRNAs for distinguishing malignant thyroid nodules from benign ones on fine-needle aspiration samples, the current meta-analysis was conducted to provide a comprehensive evaluation.
Materials and methods Search strategy We conducted a search for relevant literature published up to April 5, 2014 in PubMed, Embase, Chinese National Knowledge Infrastructure (CNKI), and Chinese Biological Medicine (CBM) databases. The following medical subject headings (MeSH) and keywords were used: (“thyroid cancer” or “thyroid tumor” or “thyroid neoplasm”) and (“microRNA” or “miRNA”) and (“diagnosis” or “ROC” or “sensitivity” or “specificity”). We also manually searched the reference lists of eligible studies to identify additional studies. Inclusion and exclusion criteria All included studies had to comply with the following inclusion criteria: (1) studies addressed the use of miRNAs for TC diagnosis; (2) diagnoses of TC were based on standard histopathology; (3) all controls with benign thyroid nodules were confirmed to be free of malignancy; and (4) studies presented sufficient data to allow for the construction of two-by-two tables, including true positive (TP), false positive (FP), true negative (TN) and false negative (FN). Studies that met the following criteria were excluded: (1) reports of duplicate data published in other studies; (2) letters, editorials, case reports or reviews; (3) studies focusing on the survival or prognosis of TC; (4) studies without qualified data. Data extraction Two investigators independently reviewed the full texts of included articles and recorded the following data: study details (first author, year and country of publication), description of study population (ethnicity, sample size, and source of control) and data for diagnostic meta-analysis (specimen, detection method, miRNA profiles, sensitivity, specificity, TP, FP, FN, and TN). Quality assessment The qualities of included studies were scored independently by two reviewers using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) criteria, which gives a maximum score of 7 [20]. This tool, containing a seven-item checklist (each of which is defined as yes, unclear, no), was specifically developed to assess studies’ applicability and risk of bias. We gave a score of 1 for each “yes” (low risk/high concern), a score of 0.5 for each “unclear” (unclear risk/unclear concern), and a score of 0 for each ‘no’ (high risk/low concern). Any scoring discrepancies were resolved through discussion.
Tumor Biol.
Results
The sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) together with their 95 % confidence intervals (95 % CIs) were summarized using a bivariate model in the current meta-analysis [21]. The sensitivity and specificity of each included study were used to plot the summary receiver operator characteristic (SROC) curve and calculate the area under the SROC curve (AUC) [22]. The Q tests and I2 statistics were used to assess the degree of heterogeneity between studies. A P value less than 0.1 for the Q test and an I2 higher than 50 % indicated the existence of significant heterogeneity [23, 24]. In addition, to explore the sources of between-study heterogeneity, subgroup analyses and meta-regression were performed according to the characteristics of the included studies. As publication bias is a concern for meta-analyses of diagnostic studies, we tested for it using Deeks’ funnel plot [25]. All analyses were performed using STATA 12.0 software.
Literature screening process The results of our literature research are presented in Fig. 1. The initial search returned a total of 65 manuscripts (61 from database searches, 4 from other sources), of which 7 were excluded as duplications. The remaining 58 research articles were subject to further evaluation. After titles and abstracts were reviewed, 48 were excluded (23 were reviews and letters, 14 were not about TC and 11 were not about miRNAs), leaving 10 articles available for full text review, following which 3 manuscripts were excluded because they lacked relevant data. Finally, our meta-analysis was performed using the remaining 7 publications [17–19, 26–29]. Basic characteristics of included studies The present study analyzed 18 studies from 7 articles, which covered 18 types of miRNAs and included 543 patients with
Articles identified through electronic database searching (N = 61)
Additional articles identified through a manual search (N = 4)
Articles reviewed for duplicates (N = 65)
Screening
7 studies were excluded, due to: (N = 7) Duplicate studies with same data
Articles after duplicates removed (N = 58) 48 studies were excluded, due to: (N = 23) Letters, reviews, meta-analysis (N = 14) Not related to Thyroid cancer (N = 11) Not related to MicroRNAs
Full-text articles assessed for eligibility Eligibility
(N = 10) 3 studies were excluded, due to: (N = 3) No available data
18 studies from 7 articles were included (N = 7 / n = 18)
Included
Fig. 1 Flow diagram of the study selection process
Identification
Statistical methods
Single miRNA assays
Multiple miRNAs assays
(n = 14)
(n = 4)
Tumor Biol.
malignant (n=266) and benign (n=277) thyroid diseases. The publication years of the included articles ranged from 2011 to 2013. In all included studies, TC was diagnosed based on standard histopathology, which is considered the gold standard reference for diagnosis. Among the 18 diagnostic studies, 11 studies were conducted in USA and 7 in Israel. All 18 studies used the quantitative real-time reverse transcriptionPCR (qRT-PCR) method to detect the expression of miRNAs on fine-needle aspiration (FNA) samples of thyroid tissues. All of the included studies examined different expression levels of miRNAs between TC patients with malignant thyroid nodules and controls with benign ones. Fourteen studies focused on a single miRNA while the remaining 4 studies researched a panel of miRNAs. The main characteristics of the included studies are listed in Table 1 by order of publication year. Quality assessments are shown in a bar graph of
QUADAS-2 scores in Fig. 2. The graph indicates that all included studies were of moderately high quality. Diagnostic accuracy of microRNAs for thyroid cancer Since we found significant heterogeneity between studies in sensitivity and specificity data (I2 =68.83 % and I2 =63.91 %, respectively), the random-effect model was used to calculate the pool estimates in this study. Forest plots of data from the 18 studies on the sensitivity and specificity of miRNAs assays in diagnosing TC are shown in Fig. 3. The pooled sensitivity and specificity were 0.77 (95 % CI: 0.70–0.83) and 0.75 (95 % CI: 0.68–0.81), respectively. The pooled PLR, NLR, and DOR with their corresponding 95 % CIs were 3.1 (95 % CI: 2.4–4.0), 0.30 (95 % CI: 0.23–0.39) and 10 (95 % CI: 7– 16), respectively. Figure 4a shows the corresponding overall
Table 1 Main characteristics of seven studies included in meta-analysis First author
Year
Country Sample size
Patient spectrum
Source miRNAs profiling of control
Sensitivity Specificity QUADAS-2 (%) (%)
PTC, FPTC, FTC, HCC
BTD
miR-7
81.0
62.5
miR-126 miR-221 miR-222 miR-21 miR-31 miR-146b miR-187 miR-21
92.0 95.0 90.0 50.0 45.0 80.0 50.0 69.0
52.0 100.0 100.0 100.0 100.0 100.0 100.0 86.0
miR-21 and miR-328 miR-328, miR-222, miR-21, and miR-197 miR-7 miR-146b, miR-221, miR-187, and miR-30d miR-146b, miR-221, miR-187, and -30d miR-100
71.0 86.0
85.0 85.0
82.0 100.0
73.0 29.0
93.2
93.8
88.9
78.3
5
62.0 78.0 75.0 78.0
67.0 67.0 67.0 70.0
4
Cases Controls Kitano et al. 2011
Mazeh et al. 2011
Keutgen et al.
USA
21
Israel
20
7
PTC
BTD
2012 USA
22
50
PTC, FPTC, FTC, HCTC
BTD
Kitano et al. 2012 USA Shen et al. 2012 USA
Vriens et al.
26
2012 USA
Mazeh et al. 2013 Israel
45 44
109 16
PTC, FPTC PTC, FPTC, FTC, ATC
44
24
PTC, FPTC, FTC BTD
56
48
PTC, FPTC, FTC, HCTC, ATC
9
2
PTC, FPTC
BTD BTD
BTD
BTD
miR-125b miR-138 miR-768-3p miR-21, miR-31, miR-146b, miR-187, miR-221, miR-222
4
4
5
6 7
NA not available, QUADAS quality assessment of diagnostic accuracy studies, PTC papillary thyroid carcinoma, HCC Hurthle cell carcinoma, FPTC follicular variant of papillary thyroid carcinoma, FTC follicular thyroid carcinoma, ATC anaplastic thyroid carcinoma, HCTC Hurthle cell thyroid carcinomas, BTD benign thyroid disease
Tumor Biol. Fig. 2 Overall quality assessment of included studies (QUADAS-2 tool): proportion of studies with low, high, or unclear risk of bias (a), proportion of studies with low, high, or unclear concerns regarding applicability (b)
(a) Risk of Bias Reference Standard
Patient Selection Bias
Index Test
Flow and Timing
0
20
40
60
80
100
80
100
percent No
Unclear
Yes
(b) Applicability Reference Standard
Patient Selection Bias
Index Test
0
20
40
60 percent
SROC curve with an AUC of 0.83 (95 % CI: 0.79-0.86). In order to explore potential sources of inter-study heterogeneity in the overall estimates, subgroup analyses were conducted. Subgroup analyses and meta-regression Subtotal analyses stratified by miRNA profiling (single miRNA vs. multiple miRNAs) and ethnicity (USA vs. Israel). The pooled sensitivity, specificity, PLR, NLR, and DOR for each subgroup are listed in Table 2. For studies conducted in the USA, the pooled sensitivity was 0.80, specificity was 0.72, PLR was 2.9, NLR was 0.28, DOR was 10, and AUC was 0.83 (Fig. 4c). For studies in Israel, the pooled sensitivity was 0.72, specificity was 0.84, PLR was 4.4, NLR was 0.33, DOR was 13 and the AUC was 0.86 (Fig. 4d). These results suggested similar accuracy outcomes for these two groups. However, the subgroup analysis
based on miRNAs profiling indicated that multiple miRNAs assays were significantly more accurate than single miRNA ones with a sensitivity of 0.87 versus 0.74, specificity of 0.85 versus 0.71, PLR of 5.9 versus 2.5, NLR of 0.16 versus 0.36, DOR of 37 versus 7 and AUC of 0.87 versus 0.78, respectively (Fig. 4e, f). We also performed meta-regression to further explore potential sources of heterogeneity and to confirm the results of subgroup analyses (Fig. 5). Our results suggest that ethnicity (USA or not), and miRNA profiles (single miRNA or not) may be potential sources of inter-study heterogeneity, especially for specificity. Sensitivity analyses and publication bias Goodness-of-fit and bivariate normality analyses (Fig. 6a, b) suggested that the random-effect bivariate model was robust
Tumor Biol.
(a) Forest Plot of Sensitivity
(b) Forest Plot of Specificity
Included studies
Sensitivity (95% CI)
Included studies
Specificity (95% CI)
2013 Mazeh et al
0.89 [0.52 − 1.00]
2013 Mazeh et al
0.67 [0.09 − 0.99]
2012 Vriens et al
0.79 [0.66 − 0.88]
2012 Vriens et al
0.71 [0.56 − 0.83]
2012 Vriens et al
0.75 [0.62 − 0.86]
2012 Vriens et al
0.67 [0.52 − 0.80]
2012 Vriens et al
0.79 [0.66 − 0.88]
2012 Vriens et al
0.67 [0.52 − 0.80]
2012 Vriens et al
0.63 [0.49 − 0.75]
2012 Vriens et al
0.67 [0.52 − 0.80]
2012 Shen et al
0.91 [0.83 − 0.96]
2012 Shen et al
0.85 [0.70 − 0.94]
2012 Kitano et al
0.89 [0.76 − 0.96]
2012 Kitano et al
0.56 [0.46 − 0.65]
2012 Keutgen et al
0.86 [0.65 − 0.97]
2012 Keutgen et al
0.86 [0.73 − 0.94]
2012 Keutgen et al
0.73 [0.50 − 0.89]
2012 Keutgen et al
0.86 [0.73 − 0.94]
2012 Keutgen et al
0.68 [0.45 − 0.86]
2012 Keutgen et al
0.86 [0.73 − 0.94]
2011 Mazeh et al
0.50 [0.27 − 0.73]
2011 Mazeh et al
0.88 [0.47 − 1.00]
2011 Mazeh et al
0.80 [0.56 − 0.94]
2011 Mazeh et al
0.88 [0.47 − 1.00]
2011 Mazeh et al
0.45 [0.23 − 0.68]
2011 Mazeh et al
0.88 [0.47 − 1.00]
2011 Mazeh et al
0.55 [0.32 − 0.77]
2011 Mazeh et al
0.86 [0.42 − 1.00]
2011 Mazeh et al
0.85 [0.62 − 0.97]
2011 Mazeh et al
0.86 [0.42 − 1.00]
2011 Mazeh et al
0.90 [0.68 − 0.99]
2011 Mazeh et al
0.75 [0.35 − 0.97]
2011 Kitano et al
0.88 [0.70 − 0.98]
2011 Kitano et al
0.48 [0.26 − 0.70]
2011 Kitano et al
0.77 [0.56 − 0.91]
2011 Kitano et al
0.57 [0.34 − 0.78]
Combined
0.77 [0.70 − 0.83]
Combined
0.75 [0.68 − 0.81]
0.20
0.77
Q = 54.54, df = 17.00, P < 0.001
Q = 47.11, df = 17.00, P < 0.001
I2 = 68.83 [53.68 − 83.98]
I2 = 63.91 [45.72 − 82.11]
1.00
0.10
0.75
1.00
Fig. 3 Forest plot showing study-specific (right-axis) and mean sensitivity and specificity with corresponding heterogeneity statistics
for the calculation of the pooled estimates. Influence analysis and outlier detection (Fig. 6c, d) identified only one outlier study. After the exclusion of the outlier study, the I2 value for heterogeneity was decreased: from 68.83 to 61.50 % for sensitivity and from 63.91 to 62.05 % for specificity. The pooled estimates (SEN, SPE, PLR, NLR, DOR, and AUC) showed only minimal changes (Table 2) (Fig. 4b), which did not significantly affect the overall estimates. In order to evaluate the final set of studies for potential publication bias, the Deeks’ funnel plot asymmetry test was used. The slope coefficient was associated with a P value of 0.67 (Fig. 7), suggesting symmetry in the data and a low likelihood of publication bias.
Discussion As the morbidity of thyroid cancer increases worldwide, more accurate methods for its diagnosis are urgently needed [1, 30]. While FNAB is the current standard method for characterizing thyroid nodules preoperatively, due to its limited ability in
differentially diagnosing indeterminate lesions, numerous studies have been conducted on the application of biomarkers to improve the diagnostic accuracy of indeterminate thyroid nodules [31, 32]. These efforts include the mutation analysis of FNA samples, identification of gene panels and miRNA analysis. Although some studies have indicated that the mutation analysis of FNA samples could improve the diagnostic accuracy for evaluation of indeterminate FNA lesions, several reports have shown that it does not detect a large number of malignant lesions and that many benign lesions also have mutations [33, 34]. As for identification of gene panels, several studies have identified some gene panels that are highly sensitive and specific in differentiating benign from malignant thyroid nodules samples [35, 36]. However, while these were validated on small cohorts of patients, the predictive accuracy of these gene panels on FNA samples in clinical contexts remains limited. Increasing attention has also been paid to the aberrant expression of miRNAs in TC and altered miRNAs expression in thyroid neoplasm has been found in several studies. Tetzlaffet al. extracted miRNAs from thyroid tissues and showed that the expressions of miR-21, 31, 221,
Tumor Biol.
(a) Overall
(b) Outlier excluded
1.0
1.0 13
3
18
12
3
2
11 4
17
7
17
7
15
16
1 15 10
9
9 14
13
Sensitivity
Sensitivity
2
14
1 16 10
5
0.5
12
11 4
8 6
5
0.5
8 6
Observed Data
Observed Data
Summary Operating Point SENS = 0.77 [0.70 − 0.83] SPEC = 0.75 [0.68 − 0.81]
Summary Operating Point SENS = 0.75 [0.68 − 0.81] SPEC = 0.74 [0.67 − 0.81]
SROC Curve AUC = 0.83 [0.79 − 0.86]
SROC Curve AUC = 0.81 [0.78 − 0.84]
95% Confidence Contour
95% Confidence Contour
95% Prediction Contour
95% Prediction Contour
0.0
0.0 1.0
0.5
0.0
1.0
0.5
Specificity
0.0
Specificity
(c) USA
(d) Israel 1.0
1.0 7 6
1
2
5
7
2
11
5
9 1 10
4 3
Sensitivity
Sensitivity
8
0.5
3
0.5
6 4
Observed Data
Observed Data
Summary Operating Point SENS = 0.80 [0.74 − 0.85] SPEC = 0.72 [0.64 − 0.79]
Summary Operating Point SENS = 0.72 [0.56 − 0.84] SPEC = 0.84 [0.70 − 0.92]
SROC Curve AUC = 0.83 [0.80 − 0.86]
SROC Curve AUC = 0.86 [0.83 − 0.89]
95% Confidence Contour
95% Confidence Contour 95% Prediction Contour
95% Prediction Contour
0.0
0.0 1.0
0.5
1.0
0.0
0.5
0.0
Specificity
Specificity
(e) Single miRNA
(f) Multiple miRNAs
1.0
1.0 3
3
10
4
2 2
4 7
14
12 1 13 1
9
Sensitivity
Sensitivity
11
5
0.5
8 6
0.5
Observed Data
Observed Data
Summary Operating Point SENS = 0.74 [0.66 − 0.81] SPEC = 0.71 [0.63 − 0.77]
Summary Operating Point SENS = 0.87 [0.77 − 0.92] SPEC = 0.85 [0.78 − 0.90]
SROC Curve AUC = 0.78 [0.75 − 0.82]
SROC Curve AUC = 0.87 [0.84 − 0.90]
95% Confidence Contour
95% Confidence Contour
95% Prediction Contour
95% Prediction Contour
0.0
0.0 1.0
0.5
Specificity
0.0
1.0
0.5
Specificity
Fig. 4 Summary ROC curve with confidence and prediction regions around mean operating sensitivity and specificity point
0.0
Tumor Biol. Table 2 Summary estimates of diagnostic criteria and the 95 % confidence intervals Sensitivity (95 %CI) Specificity (95 %CI) Positive LR (95 %CI) Negative LR (95 %CI) DOR (95 %CI) AUC (95 %CI) Ethnicity USA Israel MiRNA profiling Single miRNA Multiple miRNA Overall Outliers excluded
0.80 (0.74–0.85) 0.72 (0.56–0.84)
0.72 (0.64–0.79) 0.84 (0.70–0.92)
2.9 (2.2–3.8) 4.4 (2.3–8.5)
0.28 (0.21–0.37) 0.33 (0.20–0.56)
10 (6–17) 13 (5–35)
0.83 (0.80–0.86) 0.86 (0.83–0.89)
0.74 (0.66–0.81) 0.87 (0.77–0.92) 0.77 (0.70–0.83) 0.75 (0.68–0.81)
0.71 (0.63–0.77) 0.85 (0.78–0.90) 0.75 (0.68–0.81) 0.74 (0.67–0.81)
2.5 (2.1–3.1) 5.9 (3.9–8.8) 3.1 (2.4–4.0) 2.9 (2.3–3.7)
0.36 (0.29–0.46) 0.16 (0.09–0.27) 0.30 (0.23–0.39) 0.33 (0.27–0.42)
7 (5–10) 37 (17–81) 10 (7–16) 9 (6–12)
0.78 (0.75–0.82) 0.87 (0.84–0.90) 0.83 (0.79–0.86) 0.81 (0.78–0.84)
CI confidence interval, LR likelihood ratio, DOR diagnostic odds ratio, AUC area under the curve
malignant from benign indeterminate FNA thyroid lesions, which may obviate the risk of hypoparathyroidism and the need for unnecessary thyroid surgeries. Moreover, based on our subgroup analyses, we found that multiple miRNAs assays yielded significantly better diagnostic performance than single miRNA assays in all diagnostic parameters. These results suggest that a combination of multiple miRNAs could offer a more accurate diagnosis than single miRNA. With regard to the country origin of studied populations, there were no significant difference between the effectiveness of miRNA assays on patients from the USA and Israel. Although the miRNA assays yield excellent diagnostic characteristics, the exact role of miRNAs in the carcinogenesis of thyroid tumors is still unclear. MiRNAs may regulate gene expression at the post-transcriptional level and may thereby control cellular processes, such as developmental transitions, organ morphology, cell proliferation, and apoptosis [40]. It is
and 222 were increased in papillary tumors of the thyroid [37]. Chen et al. found that miR-222 and 146b were significantly differentially expressed between 20 benign and 20 malignant ex vivo FNA specimens [38]. Nikiforova et al. also identified 10 miRNA fragments differentially expressed in thyroid malignancies [39]. In addition, miRNAs have the unique merits of being stable, easy to be detected and non-invasive. Therefore, miRNAs have potential to serve as revolutionary biomarkers for TC to improve the accuracy of FNAB. In this meta-analysis, we found statistically significant difference in the expression levels of miRNAs between TC patients and controls with benign thyroid nodules. Overall, miRNA assays yielded an AUC of 0.83 (95 % CI: 0.79–0.86), sensitivity of 0.77 (95 % CI: 0.70–0.83) and specificity of 0.75 (95 %CI: 0.68–0.81). Although our results did not show ideal diagnostic characteristics, we found that miRNAs analysis can significantly improve diagnostic accuracy for differentiating
Fig. 5 Forest plots of multivariable meta-regression for sensitivity and specificity
(a) Meta-Regression of Sensitivity
(b) Meta-Regression of Specificity
USA Yes
**USA Yes
No
No
No. of cases
No. of cases
No. of controls
No. of controls
**Single miRNA Yes
***Single miRNA Yes
No
No 0.59 0.96 0.77 Sensitivity (95% CI)
*P
RESEARCH ARTICLE
Diagnostic value of microRNAs in discriminating malignant thyroid nodules from benign ones on fine-needle aspiration samples Yang Zhang & Qi Zhong & Xiaohong Chen & Jugao Fang & Zhigang Huang
Received: 1 May 2014 / Accepted: 6 June 2014 # International Society of Oncology and BioMarkers (ISOBM) 2014
Abstract Many studies have suggested that microRNAs (miRNAs) might serve as novel diagnostic indicators of thyroid cancer (TC). However, inconsistent results have also been reported. This meta-analysis was conducted to assess the diagnostic value of miRNAs in discriminating malignant thyroid nodules from benign ones on fine-needle aspiration samples. A systematic literature search for relevant literature published up to April 5, 2014 was conducted in PubMed, Embase, Chinese National Knowledge Infrastructure (CNKI), and Chinese Biological Medicine (CBM) databases. Data from different studies were pooled to estimate the summary sensitivity (SEN), specificity (SPE), positive likelihood ratios (PLR), negative likelihood ratios (NLR), diagnostic odds ratio (DOR), using the random-effect model. Summary receiver operator characteristic curves (SROCs) were plotted and areas under the SROC curve (AUC) were calculated to evaluate the overall test performance. Between-study heterogeneity was tested using the Q tests and the I2 statistics. Potential sources of heterogeneity were analyzed through subgroup analyses and meta-regression. Deeks’ funnel plot asymmetry test was performed to evaluate publication bias. All analyses were performed using STATA 12.0 software. Eighteen studies from 7 articles, including 543 patients with malignant thyroid nodules (n=266) and benign ones (n=277), were included in this meta-analysis. The pooled SEN was 0.77 (95 % CI: 0.70–0.83), SPN was 0.75 (95 % CI: 0.68–0.81), PLR was 3.1 (95 % CI: 2.4–4.0), NLR was 0.30 (95 % CI: 0.23–0.39), DOR was 10 (95 % CI: 7–16), and AUC was 0.83 (95 %CI: 0.79–0.86). Subgroup analyses indicated that Y. Zhang : Q. Zhong : X. Chen : J. Fang : Z. Huang (*) Department of Otolaryngology-Head and Neck Surgery, Beijing Tongren Hospital, Key Laboratory of Otolaryngology Head and Neck Surgery, Capital Medical University, No.1 Dongjiaominxiang Street, Dongcheng District, Beijing 100730, China e-mail: [email protected]
multiple miRNAs assays showed a higher diagnostic accuracy than single miRNA assays. In conclusion, this meta-analysis suggests that miRNAs analysis can significantly improve diagnostic accuracy for differentiating malignant thyroid nodules from benign indeterminate ones on fine-needle aspiration (FNA) samples. With further confirmation, multiple miRNAs assays may play a critical role as a complement to fine-needle aspiration biopsy (FNAB). Keywords MicroRNAs . Thyroid cancer . Fine-needle aspiration . Diagnostic value . Meta-analysis
Introduction Thyroid cancer (TC) is the most common endocrine malignancy (95 % of all endocrine malignancies) and its incidence has increased by 2.3-fold in the last three decades [1]. Fortunately, TC only accounts for 1 % of all malignancies worldwide and the majority of patients that are found to have thyroid malignancies have well-differentiated carcinoma associated with favorable prognosis [2, 3]. However, a vexing diagnostic problem still exists. TC typically presents as a thyroid nodule, which are very common and are diagnosed in over 7 % of the adult population [4]. However, only 5 % of thyroid nodule cases are malignant and most are benign [5]. Therefore, preoperative differentiation between malignant thyroid nodules and benign ones is imperative. Currently, fine-needle aspiration biopsy (FNAB) is the gold standard diagnostic method for thyroid nodules with a relatively high sensitivity and specificity (65–98 % and 72– 100 %, respectively) [6]. Nevertheless, up to 10–25 % of the detections are interpreted as indeterminate without a definitive cytological diagnosis [7, 8]. Further, as only 20–30 % of indeterminate nodules are malignant, up to 80 % of patients
Tumor Biol.
undergo unnecessary thyroid resections, which harbor risks like hypoparathyroidism, recurrent laryngeal nerve injury, and the need for lifelong thyroid replacement [9]. Therefore, it is of utter importance to develop non-invasive and accurate method to use along with FNAB for distinguishing between malignant thyroid nodules and benign ones. However, considering that possible improvement in the accuracy of standard cytology analysis is very limited, the incorporation of novel biomarkers for improving FNAB accuracy seems to be more feasible. In the field of molecular marker research, several studies have been conducted to evaluate the potential of microRNAs (miRNAs) as diagnostic biomarkers for TC [10, 11]. miRNAs are a family of small noncoding single-stranded RNAs (18–24 nucleotides in length) [12]. They are an abundant class of gene regulatory molecules that modulate the expression of multiple protein-coding genes. Previous researches clearly indicate that they play a critical role in a wide variety of physiologic cellular processes, including differentiation, proliferation, apoptosis, and tumorigenesis [13–15]. Moreover, miRNAs are remarkably stable in vitro and can be extracted from the residual cells obtained by FNAB, which set the foundation for miRNAs as clinical biomarkers in the diagnosis of TC. Emerging evidence has shown the feasibility of testing dysregulated miRNAs in FNAB samples to improve the diagnostic accuracy for indeterminate nodules. Sheu et al. studied the diagnostic utility of miR-146b, miR-181b, miR-21, miR-221, and miR-222 in a follicular variant of papillary thyroid cancer (FPTC) and follicular thyroid carcinoma (FTC) and found that all five miRNAs could distinguish common variants of papillary thyroid cancer from follicular adenoma (FA) and multinodular goiter [16]. In another study, Weber et al. found two miRNAs (miR-197 and miR-346) were significantly overexpressed in FTC tissues compared with their benign counterparts [11]. Several recent studies have also suggested the possible utility of miRNAs in improving TC diagnostic accuracy. However, these studies have arrived at inconsistent results. Shen et al. developed another set of 4 miRNAs (miR-146b, miR-221, miR-187, and miR30d) and suggested a high sensitivity (100.0 %) but an unexpectedly low specificity (29.0 %) [17]. Whereas Keutgen et al. reported overall good diagnostic characteristics: 86.0 % sensitivity and 85 % specificity for a set of four miRNAs (miR-328, miR-222, miR-21, and miR-197) [18]. Interestingly, the study by Mazeh et al. on a single miRNA (miR-221) also yields an excellent diagnostic accuracy with 95.0 % sensitivity and 100.0 % specificity [19]. Differences in types of patients, study designs, sample sizes, miRNAs profiling and sample types may be the sources of these studies’ inconsistent conclusions. In order to further explore the clinical applicability of miRNAs for distinguishing malignant thyroid nodules from benign ones on fine-needle aspiration samples, the current meta-analysis was conducted to provide a comprehensive evaluation.
Materials and methods Search strategy We conducted a search for relevant literature published up to April 5, 2014 in PubMed, Embase, Chinese National Knowledge Infrastructure (CNKI), and Chinese Biological Medicine (CBM) databases. The following medical subject headings (MeSH) and keywords were used: (“thyroid cancer” or “thyroid tumor” or “thyroid neoplasm”) and (“microRNA” or “miRNA”) and (“diagnosis” or “ROC” or “sensitivity” or “specificity”). We also manually searched the reference lists of eligible studies to identify additional studies. Inclusion and exclusion criteria All included studies had to comply with the following inclusion criteria: (1) studies addressed the use of miRNAs for TC diagnosis; (2) diagnoses of TC were based on standard histopathology; (3) all controls with benign thyroid nodules were confirmed to be free of malignancy; and (4) studies presented sufficient data to allow for the construction of two-by-two tables, including true positive (TP), false positive (FP), true negative (TN) and false negative (FN). Studies that met the following criteria were excluded: (1) reports of duplicate data published in other studies; (2) letters, editorials, case reports or reviews; (3) studies focusing on the survival or prognosis of TC; (4) studies without qualified data. Data extraction Two investigators independently reviewed the full texts of included articles and recorded the following data: study details (first author, year and country of publication), description of study population (ethnicity, sample size, and source of control) and data for diagnostic meta-analysis (specimen, detection method, miRNA profiles, sensitivity, specificity, TP, FP, FN, and TN). Quality assessment The qualities of included studies were scored independently by two reviewers using the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) criteria, which gives a maximum score of 7 [20]. This tool, containing a seven-item checklist (each of which is defined as yes, unclear, no), was specifically developed to assess studies’ applicability and risk of bias. We gave a score of 1 for each “yes” (low risk/high concern), a score of 0.5 for each “unclear” (unclear risk/unclear concern), and a score of 0 for each ‘no’ (high risk/low concern). Any scoring discrepancies were resolved through discussion.
Tumor Biol.
Results
The sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) together with their 95 % confidence intervals (95 % CIs) were summarized using a bivariate model in the current meta-analysis [21]. The sensitivity and specificity of each included study were used to plot the summary receiver operator characteristic (SROC) curve and calculate the area under the SROC curve (AUC) [22]. The Q tests and I2 statistics were used to assess the degree of heterogeneity between studies. A P value less than 0.1 for the Q test and an I2 higher than 50 % indicated the existence of significant heterogeneity [23, 24]. In addition, to explore the sources of between-study heterogeneity, subgroup analyses and meta-regression were performed according to the characteristics of the included studies. As publication bias is a concern for meta-analyses of diagnostic studies, we tested for it using Deeks’ funnel plot [25]. All analyses were performed using STATA 12.0 software.
Literature screening process The results of our literature research are presented in Fig. 1. The initial search returned a total of 65 manuscripts (61 from database searches, 4 from other sources), of which 7 were excluded as duplications. The remaining 58 research articles were subject to further evaluation. After titles and abstracts were reviewed, 48 were excluded (23 were reviews and letters, 14 were not about TC and 11 were not about miRNAs), leaving 10 articles available for full text review, following which 3 manuscripts were excluded because they lacked relevant data. Finally, our meta-analysis was performed using the remaining 7 publications [17–19, 26–29]. Basic characteristics of included studies The present study analyzed 18 studies from 7 articles, which covered 18 types of miRNAs and included 543 patients with
Articles identified through electronic database searching (N = 61)
Additional articles identified through a manual search (N = 4)
Articles reviewed for duplicates (N = 65)
Screening
7 studies were excluded, due to: (N = 7) Duplicate studies with same data
Articles after duplicates removed (N = 58) 48 studies were excluded, due to: (N = 23) Letters, reviews, meta-analysis (N = 14) Not related to Thyroid cancer (N = 11) Not related to MicroRNAs
Full-text articles assessed for eligibility Eligibility
(N = 10) 3 studies were excluded, due to: (N = 3) No available data
18 studies from 7 articles were included (N = 7 / n = 18)
Included
Fig. 1 Flow diagram of the study selection process
Identification
Statistical methods
Single miRNA assays
Multiple miRNAs assays
(n = 14)
(n = 4)
Tumor Biol.
malignant (n=266) and benign (n=277) thyroid diseases. The publication years of the included articles ranged from 2011 to 2013. In all included studies, TC was diagnosed based on standard histopathology, which is considered the gold standard reference for diagnosis. Among the 18 diagnostic studies, 11 studies were conducted in USA and 7 in Israel. All 18 studies used the quantitative real-time reverse transcriptionPCR (qRT-PCR) method to detect the expression of miRNAs on fine-needle aspiration (FNA) samples of thyroid tissues. All of the included studies examined different expression levels of miRNAs between TC patients with malignant thyroid nodules and controls with benign ones. Fourteen studies focused on a single miRNA while the remaining 4 studies researched a panel of miRNAs. The main characteristics of the included studies are listed in Table 1 by order of publication year. Quality assessments are shown in a bar graph of
QUADAS-2 scores in Fig. 2. The graph indicates that all included studies were of moderately high quality. Diagnostic accuracy of microRNAs for thyroid cancer Since we found significant heterogeneity between studies in sensitivity and specificity data (I2 =68.83 % and I2 =63.91 %, respectively), the random-effect model was used to calculate the pool estimates in this study. Forest plots of data from the 18 studies on the sensitivity and specificity of miRNAs assays in diagnosing TC are shown in Fig. 3. The pooled sensitivity and specificity were 0.77 (95 % CI: 0.70–0.83) and 0.75 (95 % CI: 0.68–0.81), respectively. The pooled PLR, NLR, and DOR with their corresponding 95 % CIs were 3.1 (95 % CI: 2.4–4.0), 0.30 (95 % CI: 0.23–0.39) and 10 (95 % CI: 7– 16), respectively. Figure 4a shows the corresponding overall
Table 1 Main characteristics of seven studies included in meta-analysis First author
Year
Country Sample size
Patient spectrum
Source miRNAs profiling of control
Sensitivity Specificity QUADAS-2 (%) (%)
PTC, FPTC, FTC, HCC
BTD
miR-7
81.0
62.5
miR-126 miR-221 miR-222 miR-21 miR-31 miR-146b miR-187 miR-21
92.0 95.0 90.0 50.0 45.0 80.0 50.0 69.0
52.0 100.0 100.0 100.0 100.0 100.0 100.0 86.0
miR-21 and miR-328 miR-328, miR-222, miR-21, and miR-197 miR-7 miR-146b, miR-221, miR-187, and miR-30d miR-146b, miR-221, miR-187, and -30d miR-100
71.0 86.0
85.0 85.0
82.0 100.0
73.0 29.0
93.2
93.8
88.9
78.3
5
62.0 78.0 75.0 78.0
67.0 67.0 67.0 70.0
4
Cases Controls Kitano et al. 2011
Mazeh et al. 2011
Keutgen et al.
USA
21
Israel
20
7
PTC
BTD
2012 USA
22
50
PTC, FPTC, FTC, HCTC
BTD
Kitano et al. 2012 USA Shen et al. 2012 USA
Vriens et al.
26
2012 USA
Mazeh et al. 2013 Israel
45 44
109 16
PTC, FPTC PTC, FPTC, FTC, ATC
44
24
PTC, FPTC, FTC BTD
56
48
PTC, FPTC, FTC, HCTC, ATC
9
2
PTC, FPTC
BTD BTD
BTD
BTD
miR-125b miR-138 miR-768-3p miR-21, miR-31, miR-146b, miR-187, miR-221, miR-222
4
4
5
6 7
NA not available, QUADAS quality assessment of diagnostic accuracy studies, PTC papillary thyroid carcinoma, HCC Hurthle cell carcinoma, FPTC follicular variant of papillary thyroid carcinoma, FTC follicular thyroid carcinoma, ATC anaplastic thyroid carcinoma, HCTC Hurthle cell thyroid carcinomas, BTD benign thyroid disease
Tumor Biol. Fig. 2 Overall quality assessment of included studies (QUADAS-2 tool): proportion of studies with low, high, or unclear risk of bias (a), proportion of studies with low, high, or unclear concerns regarding applicability (b)
(a) Risk of Bias Reference Standard
Patient Selection Bias
Index Test
Flow and Timing
0
20
40
60
80
100
80
100
percent No
Unclear
Yes
(b) Applicability Reference Standard
Patient Selection Bias
Index Test
0
20
40
60 percent
SROC curve with an AUC of 0.83 (95 % CI: 0.79-0.86). In order to explore potential sources of inter-study heterogeneity in the overall estimates, subgroup analyses were conducted. Subgroup analyses and meta-regression Subtotal analyses stratified by miRNA profiling (single miRNA vs. multiple miRNAs) and ethnicity (USA vs. Israel). The pooled sensitivity, specificity, PLR, NLR, and DOR for each subgroup are listed in Table 2. For studies conducted in the USA, the pooled sensitivity was 0.80, specificity was 0.72, PLR was 2.9, NLR was 0.28, DOR was 10, and AUC was 0.83 (Fig. 4c). For studies in Israel, the pooled sensitivity was 0.72, specificity was 0.84, PLR was 4.4, NLR was 0.33, DOR was 13 and the AUC was 0.86 (Fig. 4d). These results suggested similar accuracy outcomes for these two groups. However, the subgroup analysis
based on miRNAs profiling indicated that multiple miRNAs assays were significantly more accurate than single miRNA ones with a sensitivity of 0.87 versus 0.74, specificity of 0.85 versus 0.71, PLR of 5.9 versus 2.5, NLR of 0.16 versus 0.36, DOR of 37 versus 7 and AUC of 0.87 versus 0.78, respectively (Fig. 4e, f). We also performed meta-regression to further explore potential sources of heterogeneity and to confirm the results of subgroup analyses (Fig. 5). Our results suggest that ethnicity (USA or not), and miRNA profiles (single miRNA or not) may be potential sources of inter-study heterogeneity, especially for specificity. Sensitivity analyses and publication bias Goodness-of-fit and bivariate normality analyses (Fig. 6a, b) suggested that the random-effect bivariate model was robust
Tumor Biol.
(a) Forest Plot of Sensitivity
(b) Forest Plot of Specificity
Included studies
Sensitivity (95% CI)
Included studies
Specificity (95% CI)
2013 Mazeh et al
0.89 [0.52 − 1.00]
2013 Mazeh et al
0.67 [0.09 − 0.99]
2012 Vriens et al
0.79 [0.66 − 0.88]
2012 Vriens et al
0.71 [0.56 − 0.83]
2012 Vriens et al
0.75 [0.62 − 0.86]
2012 Vriens et al
0.67 [0.52 − 0.80]
2012 Vriens et al
0.79 [0.66 − 0.88]
2012 Vriens et al
0.67 [0.52 − 0.80]
2012 Vriens et al
0.63 [0.49 − 0.75]
2012 Vriens et al
0.67 [0.52 − 0.80]
2012 Shen et al
0.91 [0.83 − 0.96]
2012 Shen et al
0.85 [0.70 − 0.94]
2012 Kitano et al
0.89 [0.76 − 0.96]
2012 Kitano et al
0.56 [0.46 − 0.65]
2012 Keutgen et al
0.86 [0.65 − 0.97]
2012 Keutgen et al
0.86 [0.73 − 0.94]
2012 Keutgen et al
0.73 [0.50 − 0.89]
2012 Keutgen et al
0.86 [0.73 − 0.94]
2012 Keutgen et al
0.68 [0.45 − 0.86]
2012 Keutgen et al
0.86 [0.73 − 0.94]
2011 Mazeh et al
0.50 [0.27 − 0.73]
2011 Mazeh et al
0.88 [0.47 − 1.00]
2011 Mazeh et al
0.80 [0.56 − 0.94]
2011 Mazeh et al
0.88 [0.47 − 1.00]
2011 Mazeh et al
0.45 [0.23 − 0.68]
2011 Mazeh et al
0.88 [0.47 − 1.00]
2011 Mazeh et al
0.55 [0.32 − 0.77]
2011 Mazeh et al
0.86 [0.42 − 1.00]
2011 Mazeh et al
0.85 [0.62 − 0.97]
2011 Mazeh et al
0.86 [0.42 − 1.00]
2011 Mazeh et al
0.90 [0.68 − 0.99]
2011 Mazeh et al
0.75 [0.35 − 0.97]
2011 Kitano et al
0.88 [0.70 − 0.98]
2011 Kitano et al
0.48 [0.26 − 0.70]
2011 Kitano et al
0.77 [0.56 − 0.91]
2011 Kitano et al
0.57 [0.34 − 0.78]
Combined
0.77 [0.70 − 0.83]
Combined
0.75 [0.68 − 0.81]
0.20
0.77
Q = 54.54, df = 17.00, P < 0.001
Q = 47.11, df = 17.00, P < 0.001
I2 = 68.83 [53.68 − 83.98]
I2 = 63.91 [45.72 − 82.11]
1.00
0.10
0.75
1.00
Fig. 3 Forest plot showing study-specific (right-axis) and mean sensitivity and specificity with corresponding heterogeneity statistics
for the calculation of the pooled estimates. Influence analysis and outlier detection (Fig. 6c, d) identified only one outlier study. After the exclusion of the outlier study, the I2 value for heterogeneity was decreased: from 68.83 to 61.50 % for sensitivity and from 63.91 to 62.05 % for specificity. The pooled estimates (SEN, SPE, PLR, NLR, DOR, and AUC) showed only minimal changes (Table 2) (Fig. 4b), which did not significantly affect the overall estimates. In order to evaluate the final set of studies for potential publication bias, the Deeks’ funnel plot asymmetry test was used. The slope coefficient was associated with a P value of 0.67 (Fig. 7), suggesting symmetry in the data and a low likelihood of publication bias.
Discussion As the morbidity of thyroid cancer increases worldwide, more accurate methods for its diagnosis are urgently needed [1, 30]. While FNAB is the current standard method for characterizing thyroid nodules preoperatively, due to its limited ability in
differentially diagnosing indeterminate lesions, numerous studies have been conducted on the application of biomarkers to improve the diagnostic accuracy of indeterminate thyroid nodules [31, 32]. These efforts include the mutation analysis of FNA samples, identification of gene panels and miRNA analysis. Although some studies have indicated that the mutation analysis of FNA samples could improve the diagnostic accuracy for evaluation of indeterminate FNA lesions, several reports have shown that it does not detect a large number of malignant lesions and that many benign lesions also have mutations [33, 34]. As for identification of gene panels, several studies have identified some gene panels that are highly sensitive and specific in differentiating benign from malignant thyroid nodules samples [35, 36]. However, while these were validated on small cohorts of patients, the predictive accuracy of these gene panels on FNA samples in clinical contexts remains limited. Increasing attention has also been paid to the aberrant expression of miRNAs in TC and altered miRNAs expression in thyroid neoplasm has been found in several studies. Tetzlaffet al. extracted miRNAs from thyroid tissues and showed that the expressions of miR-21, 31, 221,
Tumor Biol.
(a) Overall
(b) Outlier excluded
1.0
1.0 13
3
18
12
3
2
11 4
17
7
17
7
15
16
1 15 10
9
9 14
13
Sensitivity
Sensitivity
2
14
1 16 10
5
0.5
12
11 4
8 6
5
0.5
8 6
Observed Data
Observed Data
Summary Operating Point SENS = 0.77 [0.70 − 0.83] SPEC = 0.75 [0.68 − 0.81]
Summary Operating Point SENS = 0.75 [0.68 − 0.81] SPEC = 0.74 [0.67 − 0.81]
SROC Curve AUC = 0.83 [0.79 − 0.86]
SROC Curve AUC = 0.81 [0.78 − 0.84]
95% Confidence Contour
95% Confidence Contour
95% Prediction Contour
95% Prediction Contour
0.0
0.0 1.0
0.5
0.0
1.0
0.5
Specificity
0.0
Specificity
(c) USA
(d) Israel 1.0
1.0 7 6
1
2
5
7
2
11
5
9 1 10
4 3
Sensitivity
Sensitivity
8
0.5
3
0.5
6 4
Observed Data
Observed Data
Summary Operating Point SENS = 0.80 [0.74 − 0.85] SPEC = 0.72 [0.64 − 0.79]
Summary Operating Point SENS = 0.72 [0.56 − 0.84] SPEC = 0.84 [0.70 − 0.92]
SROC Curve AUC = 0.83 [0.80 − 0.86]
SROC Curve AUC = 0.86 [0.83 − 0.89]
95% Confidence Contour
95% Confidence Contour 95% Prediction Contour
95% Prediction Contour
0.0
0.0 1.0
0.5
1.0
0.0
0.5
0.0
Specificity
Specificity
(e) Single miRNA
(f) Multiple miRNAs
1.0
1.0 3
3
10
4
2 2
4 7
14
12 1 13 1
9
Sensitivity
Sensitivity
11
5
0.5
8 6
0.5
Observed Data
Observed Data
Summary Operating Point SENS = 0.74 [0.66 − 0.81] SPEC = 0.71 [0.63 − 0.77]
Summary Operating Point SENS = 0.87 [0.77 − 0.92] SPEC = 0.85 [0.78 − 0.90]
SROC Curve AUC = 0.78 [0.75 − 0.82]
SROC Curve AUC = 0.87 [0.84 − 0.90]
95% Confidence Contour
95% Confidence Contour
95% Prediction Contour
95% Prediction Contour
0.0
0.0 1.0
0.5
Specificity
0.0
1.0
0.5
Specificity
Fig. 4 Summary ROC curve with confidence and prediction regions around mean operating sensitivity and specificity point
0.0
Tumor Biol. Table 2 Summary estimates of diagnostic criteria and the 95 % confidence intervals Sensitivity (95 %CI) Specificity (95 %CI) Positive LR (95 %CI) Negative LR (95 %CI) DOR (95 %CI) AUC (95 %CI) Ethnicity USA Israel MiRNA profiling Single miRNA Multiple miRNA Overall Outliers excluded
0.80 (0.74–0.85) 0.72 (0.56–0.84)
0.72 (0.64–0.79) 0.84 (0.70–0.92)
2.9 (2.2–3.8) 4.4 (2.3–8.5)
0.28 (0.21–0.37) 0.33 (0.20–0.56)
10 (6–17) 13 (5–35)
0.83 (0.80–0.86) 0.86 (0.83–0.89)
0.74 (0.66–0.81) 0.87 (0.77–0.92) 0.77 (0.70–0.83) 0.75 (0.68–0.81)
0.71 (0.63–0.77) 0.85 (0.78–0.90) 0.75 (0.68–0.81) 0.74 (0.67–0.81)
2.5 (2.1–3.1) 5.9 (3.9–8.8) 3.1 (2.4–4.0) 2.9 (2.3–3.7)
0.36 (0.29–0.46) 0.16 (0.09–0.27) 0.30 (0.23–0.39) 0.33 (0.27–0.42)
7 (5–10) 37 (17–81) 10 (7–16) 9 (6–12)
0.78 (0.75–0.82) 0.87 (0.84–0.90) 0.83 (0.79–0.86) 0.81 (0.78–0.84)
CI confidence interval, LR likelihood ratio, DOR diagnostic odds ratio, AUC area under the curve
malignant from benign indeterminate FNA thyroid lesions, which may obviate the risk of hypoparathyroidism and the need for unnecessary thyroid surgeries. Moreover, based on our subgroup analyses, we found that multiple miRNAs assays yielded significantly better diagnostic performance than single miRNA assays in all diagnostic parameters. These results suggest that a combination of multiple miRNAs could offer a more accurate diagnosis than single miRNA. With regard to the country origin of studied populations, there were no significant difference between the effectiveness of miRNA assays on patients from the USA and Israel. Although the miRNA assays yield excellent diagnostic characteristics, the exact role of miRNAs in the carcinogenesis of thyroid tumors is still unclear. MiRNAs may regulate gene expression at the post-transcriptional level and may thereby control cellular processes, such as developmental transitions, organ morphology, cell proliferation, and apoptosis [40]. It is
and 222 were increased in papillary tumors of the thyroid [37]. Chen et al. found that miR-222 and 146b were significantly differentially expressed between 20 benign and 20 malignant ex vivo FNA specimens [38]. Nikiforova et al. also identified 10 miRNA fragments differentially expressed in thyroid malignancies [39]. In addition, miRNAs have the unique merits of being stable, easy to be detected and non-invasive. Therefore, miRNAs have potential to serve as revolutionary biomarkers for TC to improve the accuracy of FNAB. In this meta-analysis, we found statistically significant difference in the expression levels of miRNAs between TC patients and controls with benign thyroid nodules. Overall, miRNA assays yielded an AUC of 0.83 (95 % CI: 0.79–0.86), sensitivity of 0.77 (95 % CI: 0.70–0.83) and specificity of 0.75 (95 %CI: 0.68–0.81). Although our results did not show ideal diagnostic characteristics, we found that miRNAs analysis can significantly improve diagnostic accuracy for differentiating
Fig. 5 Forest plots of multivariable meta-regression for sensitivity and specificity
(a) Meta-Regression of Sensitivity
(b) Meta-Regression of Specificity
USA Yes
**USA Yes
No
No
No. of cases
No. of cases
No. of controls
No. of controls
**Single miRNA Yes
***Single miRNA Yes
No
No 0.59 0.96 0.77 Sensitivity (95% CI)
*P
