Characteristics and quality of the included studies
Figure 1 depicts the flowchart of the literature search, and Table 1 presents the detailed characteristics of the studies. Initially, 937 relevant studies were identified in the systematic literature search. A flowchart of the detailed selection steps was provided in Fig. 1. By checking the titles and abstracts, 662 studies were excluded and 99 potential studies were retrieved. An additional 71 studies were then excluded after they were fully reviewed for the following reasons: narrative review/editorial/ comment. Finally, 7 studies were yielded as meeting our inclusion criteria and were eligible for our study (Fig. 1). The 7 included studies were published between 2014 and 2018. All studies detected tumor cells from peripheral blood with the molecular detection method of immunofluorescence or CellSearch. In addition, all of the studies were prospective design. The methodological quality of the studies assessed by the QUADAS-2 tool is depicted in Fig. 2, and the overall methodological quality was better.
Table 1
Main characters of the studies included in the meta-analysis [25–31]
Authors | Year | District | Study design | participants | Reference standard | Detection method | Biomarkers | TP | FP | TN | FN |
Lin et al. | 2016 | Taiwan | Prospective | 48 | Cytopathologically | Immunofluorescence | EpCAM、TSHR | 22 | 2 | 4 | 20 |
Xu et al. | 2016 | USA | Prospective | 42 | Medical image | CellSearch | EpCAM | 13 | 0 | 19 | 10 |
Tseng et al. | 2017 | Taiwan | Prospective | 77 | Medical image | immunofluorescence | EpCAM | 27 | 2 | 5 | 43 |
Li et al. | 2018 | Taiwan | Prospective | 25 | Medical image | immunofluorescence | EpCAM、TSHR | 6 | 0 | 1 | 18 |
Qiu et al. | 2018 | China | Prospective | 72 | Pathological、medical image | immunofluorescence | EpCAM | 19 | 7 | 11 | 35 |
Winken et al. | 2014 | Germany | Prospective | 14 | Tg、Medical image | immunofluorescence | EpCAM | 1 | 4 | 1 | 8 |
Lin et al. | 2015 | Taiwan | Prospective | 29 | Biochemical、medical image | immunofluorescence | EpCAM、TSHR | 16 | 2 | 2 | 9 |
TP: true positive, FP༚false positive༌ TN༚true negative༌ FN༚false negative |
Overall analysis
An overview of the sensitivity and specificity of EpCAM in the detection of recurrence or metastases among TC patients is displayed in Fig. 3. Seven studies investigated the diagnostic accuracy of EpCAM in 307 participants. There was heterogeneity shown in meta-analysis of EpCAM, as revealed by the results of sensitivity (p = 0.0005, I2 = 74.9%) and specificity (p = 0.02, I2 = 60.1%), respectively. Thus, random effects models were adopted for analysis and the DOR was 26.75 (95%CI: 9.11–78.53) (Fig. 5). The pooled positive likelihood ratio was 6.19 (95% CI: 3.05–12.57), and the pooled negative likelihood ratio was 0.30 (95% CI: 0.15–0.57). Of note, the AUC of the SROC plot was employed to assess the overall accuracy of a diagnostic test. The SROC curve of EpCAM produced an AUC of 0.91(Fig. 6), which showed that the overall performance of EpCAM was suitable.
Three studies investigated the diagnostic accuracy of TSHR in 102 participants. This meta-analysis of TSHR for detecting the recurrence of TC generated a pooled sensitivity of 0.88 (95%CI: 0.76–0.96) and pooled specificity of 0.78 (95%CI: 0.65–0.89) (Fig. 4). The AUC of SROC curve is 0.91. The pooled DOR for the TSHR was 40.01 (95%CI: 10.49-152.63) (Fig. 5), thus showing the high discriminative power.
The Youden index, indicating the diagnostic test accuracy, calculated as (sensitivity + specificity − 1). A diagnostic test is supposed to be suitable if its Youden index is above 0.5. The Youden index of the TSHR is 0.66, which was higher than that of EpCAM (0.6). Based on the comparative outcomes of the AUC and Youden index for detecting the recurrence or metastases, EpCAM and TSHR were both excellent choices and TSHR was slightly superior to EpCAM.
Evaluation of threshold effect and heterogeneity
To evaluate the reason of high heterogeneity, the threshold effect was firstly considered. For biomarker of EpCAM, no evident threshold effect was presented in this meta-analysis, supported by the value of proportion of heterogeneity likely due to threshold effect (0.25).
Given study design, socio-demographic characteristics and geographical location might also represent sources of the heterogeneity, we conducted a meta-regression analysis with geographical location, quality of study, study design and detection methods include in Table 2. No significant heterogeneity was suggested with respect to region (coefficient=-1.118, p = 0.2360), quality of study (coefficient = 0.036, p = 0.7824), study design (coefficient = 0.201, p = 0.8996) and detection methods (coefficient = -1.862, p = 0.5867).Thus, other factors might contribute to the observed high heterogeneity.
Table 2
Results of meta-regression on EpCAM
Variable | Coefficient | Standard Error | P value | RDOR |
Region | -1.118 | 0.4346 | 0.2360 | 0.33 |
Detection method | -1.862 | 2.4535 | 0.5867 | 0.16 |
Quality of study | 0.036 | 0.7824 | 0.9708 | 1.04 |
Study design | 0.201 | 1.2607 | 0.8996 | 1.22 |
P < 0.05 was considered as to be statistically significant. |
Publication bias
Potential publication bias was assessed by Deeks’ funnel plots. Note that, p < 0.05 indicated the existence of publication bias. There was no evidence of publication bias for the pooled analysis of EpCAM (p = 0.66) and TSHR (p = 0.34). The funnel plots of publication bias on EpCAM and TSHR are shown in Fig. 7.