3.1. Study characteristics
We did a search through PubMed, EMBASE, the Cochrane Library and Web of Science and identified a total of 907 records. Among these records, 337 were duplicate studies and were, therefore, excluded. An additional 329 articles were excluded after reading the titles and another 203 publications after reviewing the abstracts. The remaining 38 full-text articles were assessed for relevance according to our pre-determined inclusion and exclusion criteria. Subsequently, 32 studies, including four meetings and 28 without clinical data were excluded, and a two-by-two contingency table was made to calculate HRs. We only extracted enough data to calculate the diagnostic value of miRNAs. Only six diagnostic-related researches were ultimately included in this study [4, 5, 8-11]. A flow chart of the selection process for this study is presented in Fig.1.
The included six articles (ranging from the year 2010 to 2018) reported 37 studies, involving a total of 170 TBI patients and 121 controls composed of healthy controls and other diseases (Table 1). Among the 37 studies, 23 reported a single miRNA, while the additional 14 discussed a panel miRNAs (Table 2). Out of the 37 articles, ten detected miRNA in plasma, five detected miRNA in serum, six detected miRNA in saliva, one identified miRNA in brain‐derived extracellular vesicles, and six evaluated the brain tissue. Of the 37 studies, the populations of 35 studies were Caucasian, whereas two studies were Asian. A total of 24 studies were conducted in severe TBI patients, eleven in mild TBI patients, and the remaining two studies focused on TBI patients.
3.2. Diagnosis
The diagnostic value of miRNAs for TBI is shown in Fig.2. Forest plots revealed a significant heterogeneity and we, therefore, used the mixed effect model in this meta-analysis. We also summarized sensitivity, specificity, and diagnostic accuracy of all miRNAs in TBI (Table 3). The sensitivity, specificity, PLR, NLR, and DOR of overall miRNA for diagnosis of TBI were 0.92 (95% CI: 0.87–0.95), 0.92(95% CI: 0.78–0.97), 11.1 (95% CI: 3.8–32.7), 0.09 (95% CI: 0.05–0.15) and 128 (95% CI: 29–575). Diagnostic accuracy was evaluated by plotting the summary receiver operating characteristic (SROC) curve (Fig.3a). The diagnostic accuracy of overall miRNAs was outstanding since the area under the Curve (AUC) was 0.96 (95% CI: 0.94–0.97). We performed subgroup analyses according to ethnicity, detected sample, and miRNA profiling in order to find the heterogeneity (Fig. 4b). The diagnostic value of single miRNAs was as follows: sensitivity, 0.91; specificity, 0.90; PLR, 8.9; NLR, 0.10; DOR, 93; and AUC, 0.95. However, miRNA panels have a higher overall diagnostic accuracy: sensitivity, 0.93; specificity, 0.94; PLR, 15.4; NLR, 0.07; DOR, 216; and AUC, 0.97 (Fig. 5 a, b, and c). The sensitivity, specificity, PLR, NLR, DOR and AUC of saliva, brain tissue, and blood were 0.73, 0.17, 0.90, 1.59, 0.6 and 0.40; 0.88, 0.87, 6.8, 0.13, 53 and 0.94; 0.99, 0.99, 162, 0.01, 12000 and 1.00, respectively (Fig. 3 b, c and d). This result suggested that miRNAs dected in blood have the highest overall diagnostic accuracy. In the severe TBI patients, the results were 0.92 for sensitivity, 0.92 for specificity, 12 for PLR, 0.09 for NLR, 129 for DOR, and 0.97 for AUC (Fig. 4a).
3.3. Sensitivity analysis and meta-regression analysis
The goodness of fit and bivariate normality analyses revealed that the random effects bivariate model was best suited for sensitivity analysis (Fig. 6aand 6b). Influence analysis showed that studies of Schober et al., Di Pietro et al., and Yang et al. were the leading researches in weight (Fig. 6c). Outlier detection identified that no research would significantly affect the heterogeneity of our meta-analysis (Fig. 6d). Considering the bias of miRNAs, ethnicity, and the detected sample, we conducted a meta-regression analysis and found that the detected sample may influence sensitivity and specificity. Results on subgroup analyses indicated that miRNA detected in blood exhibit the highest sensitivity and specificity in the diagnosis of TBI. We further conducted a subgroup analysis according to the type of TBI. In sTBI, there was no apparent heterogeneity because the I2 value was only 27.07% for sensitivity and 47.04% for specificity. After excluding non-severe TBI studies, the sensitivity and specificity of I2 value dramatically decreased 59.87% and 41.66% respectively (Fig. 5d). We thought that non-severe TBI studies could be the reason for heterogeneity. However, we did not do a subgroup analysis of mTBI studies due to the limitation of the number of mTBI studies.