Search strategy
The systematic literature search was performed based on PubMed, Embase and other databases for eligible original literatures until December 1st, 2018. The relevant keywords included the following items: “miR-155” or “microRNA-155” or “miRNA-155”, and “lung cancer” or “NSCLC” or “lung”, and “prognosis” or “diagnosis” or “detection” or “variants”. The MeSH terminology and relevant keywords were randomly combined in order to ensure acquiring the most comprehensive data. In addition, we also sifted through the reference lists of original articles and manually searched from relevant reviews for additional literatures.
Inclusion and exclusion criteria
In order to screen out eligible studies, the following criteria were adopted: (1) Research focus on patients with definite diagnosis of LCa; (2) Detection of miR-155 expression in plasma, serum or other human body fluids; (3) Sufficient data of assessing the coorelation between miR-155 over-expression and poor overall survival (OS), disease free survival (DFS) and progression-free survival (PFS) in LCa patients; (4) Available data of true positive (TP), false positive (FP), false negative (FN), true negative (TN) or clear sample size combined with sensitivity (SEN) and specificity (SPE) to calculate the area under the ROC curve (AUC) for diagnostic analysis. In addition, the studies would be excluded according to the following criteria: (1) Non case-control studies, letters, reviews, comments; (2) Non-English or Chinese studies; (3) No data available for LCa diagnosis and prognosis; (4) Duplicates or the same samples used in previous publications.
Data extraction
All Data were extracted from the included studies by two investigators (SCC and YFM), the uncertain results were reviewed by the third investigator (QZQ). The extracted data consist of following items: first author's name, country, year of publication, ethnicity of the population studied; number of patients and controls; assay type; and diagnostic results of SEN, SPE, TP, FP, FN, and TN; or prognostic outcomes including HRs of elevated miR-155 expression for OS/DFS/PFS. In addition, if not directly available from each article, data was extracted from the Kaplan-Meier curve using the previously described method to infer HR with 95% CI.
Quality assessment
Two researchers (SCC and YFM) in our institution assessed whether each included literature met the quality standards separately. Then, another researcher (QZQ) will reevaluate and make a unified conclusion if there is a discrepancy between first two researchers. For diagnostic meta-analysis, the quality assessment was conducted following the guidelines of the the Quality assessment of diagnostic accuracy studies 2 (QUADAS-2) [14]. This tool consists of 4 domains for assessing the risk and applicability of bias. The four areas (patient selection, index test, reference standard, and flow and timing) are refined into 14 specific questions. Each item has a rating of "Yes", "No" or "Unclear", corresponding to the scores of -1, 1 and 0, respectively (Figure 2). For prognostic meta-analysis, the quality of each included article was assessed by the Newcastle-Ottawa Scale (NOS), which is one of the most authoritative tools for assessing the quality of non-randomized studies [15, 16]. By scoring one by one, the total quality score ranges from 0 to 9. Studies with a final score > 6 were considered high-quality.
Statistical analysis
For diagnostic accuracy studies, the SEN, SPE, PLR, NLR and corresponding 95% CI from included studies were pooled to initially assess the diagnostic value of circulating miR-155 in LCa. The summary receiver operating characteristic (SROC) curve was then drawn based on the original data, and the area under the SROC curve (AUC) was calculated to comprehensively determine the diagnostic accuracy of miR-155, taking into account the trade-off between SEN and SPE. To assess the heterogeneity across studies, the X2-based Q-statistic and I2 statistic were utilized. The I2 square value typically fluctuates within a range of 0 (unobserved heterogeneity) to 100% (maximum heterogeneity). P value <0.05 or I2>50% was recognized statistically significant [17]. If the studies were proved to be homogenous, a fixed-effect model would be utilized for further analysis. If not, the random-effect model would be utilized [18]. Subsequently, subgroup and meta-regression analyses were carried out to find the potential sources of heterogeneity. Finally, the publication bias of all the included diagnostic accuracy studies was assessed by Deeks' funnel plots (significant at P < 0.05) [19].
For prognostic meta-analyses, a combination of the pooled HR and 95% CI was calculated to elucidate the link between high expression of miR-155 and cooresponding OS/DFS/PFS of LCa patients. Cochran’s Q test and I2 statistics were applied to evaluate the heterogeneity of the pooled results [20]. In addition, Begg’s and Egger’s tests were selected to evaluate the potential publication bias. All above statistical analysis was performed by the statistical software STATA (version 12.0) [21].