Studies included in the meta-analysis need to meet the following criteria: (1) the diagnosis of gastric cancer must be confirmed by pathological examination; (2) there was available data to be investigated which was provided by the original article, such as HR and 95%CI for the OS and (or) DFS, the number of patients with various clinicopathological features; (3) the blood samples was obtained before treatment, PLR was calculated as the ratio of the platelets to lymphocytes; Studies were excluded based on the following criteria: (1) letters, conference abstracts, posters or review articles; (2) insufficient data to estimate from the article; (3) not human research or not published in English.
Data extraction and quality assessmentAll studies were assessed independently by two authors according to the designed eligibility criteria.Any questions or disagreements were resolved by consulting another co-author. The extracted data from each study included: first author, publication year, country, study design (retrospective or prospective), study period, treatment regimens, follow-up time, cut-off value of PLR, the number of patients with various clinicopathological features, such as tumor location, differentiation, size, depths of tumor invasion, lymph node metastasis, TNM stage, HRs with 95% CIs of OS and DFS. The quality of each study was assessed according to the Newcastle-Ottawa Scale (NOS) by two authors [14]. Studies with an NOS score ≥ 6 were considered as high-quality researches.
Statistical analysisHRs and 95% CIs for OS and DFS were obtained directly from each study if available or were calculated from the necessary data according to the methods published by Tierney et al. [15].Cochran's Q test and Higgins I-squared statistic were used to evaluate the heterogeneity of pooled results. A p-value < 0.1 for the Q-test or Ι2 > 50% indicated significant heterogeneity among studies, and the random-effects model (DerSimonian-Laird method) was performed to calculate the pooled HRs [14]. Otherwise, the fixed-effects model (Mantel-Haenszel method) was applied [16]. Odds ratios (OR) and 95% CI were used to analysis the relationship between PLR and clinicopathological factors.Publication bias of the literature was evaluated by Begg's funnel plot and Egger's linear regression tests, and p > 0.05 indicated that there was no significant publication bias. Sensitivity analysis was