Cohort studies assessing the prognostic significance of serum LDH levels in EWS were searched in the PubMed, Embase and Web of Science databases. The search strategy was based on combinations of the following terms: ‘Lactate Dehydrogenase’ (OR ‘Lactate Dehydrogenases’ OR ‘LDH’) AND ‘Ewing's sarcoma’ (OR ‘Ewing sarcoma’ OR ‘Sarcoma, Ewing’). An upper date limit of October 10, 2015, was applied. Only studies published in English were included, and unpublished reports were considered. Studies eligible for inclusion in the present meta-analysis conformed with the following criteria: i) A prospective or retrospective cohort study, or randomized controlled trial; ii) tumors has been histologically confirmed as EWS; iii) the studies had examined the association between serum LDH levels and clinical outcome; and iv) the studies had provided sufficient information to estimate the hazard ratio (HR) and 95% confidence interval (CI) of overall survival (OS) or 5-year disease-free survival (DFS). When the identical authors had reported two or more publications on possibly the same patient populations, only the most recent or most complete study was included in this meta-analysis. Qing Yang and Zhuoying Wang performed the study selection, Suoyuan Li checking the other's work.

The present study was approved by the institutional review board of Shanghai General Hospital (Shanghai. China), and it conforms to the provisions of the Declaration of Helsinki.

Each paper was analyzed for extraction of the relevant data, including first author, publication year, country of the patients, age and gender of the patients, the number of cases, site of the primary tumor and follow-up time. The cutoff of a high level of LDH was also extracted. When the studies involved time-to-event data, the most appropriate statistics to use were log hazard ratio (logHR) and its variance; however, these were not always stated explicitly in each study. Thus, the logHR and standard error of each study were obtained through the following methods: i) Directly extracting the unadjusted HR and 95% CI from each article; ii) estimating HR using the log-rank test, P-values, total events, high level and control group figures (12); and iii) estimating HR using data from the Kaplan-Meier survival curves read by the Engauge Digitizer software (http://markummitchell.github.io/engauge-digitizer/), as well as the minimal and maximal follow-up times (13). Again, Qing Yang and Zhuoying Wang performed the study selection, Suoyuan Li checking the other's work.

The quality of each included study was assessed by two independent reviewers using the Newcastle Ottawa Quality Assessment Scale (NOQAS) (14). These scales were used to allocate a maximum of nine points for the quality of selection, exposure, comparability and outcome of study participants.

HR with 95% CI was used to assess the prognostic role of LDH expression. A heterogeneity test, with inconsistency index (I²) and Q statistical values, was performed. The random effects model was used when significant heterogeneity was observed among the included studies (I² >50%). The fixed-effects model was used for the analysis when there no significant heterogeneity was observed across the included studies (I² ≤50%) (15). According to the convention, an observed HR >1 implies worse survival rates for the group with high LDH levels. If the 95% CI did not overlap with 1, the impact of LDH on the survival rate was considered to be statistically significant. To validate the credibility of outcomes in the present meta-analysis, sensitivity analysis was performed by the sequential omission of individual studies. The possibility of publication bias was evaluated by visually assessing the symmetry of Begg's funnel plots and Egger's test, where P<0.05 was considered to indicate a statistically significant value (16). Statistical analyses were performed using the software STATA version 12 (StataCorp LP, College Station, TX, USA). A two-tailed value of P<0.05 was considered to indicate a statistically significant value.

By searching the PubMed, Embase and Web of Science databases, a total of 52 studies were initially identified. Fig. 1 shows a flow diagram of the selection process for defining the relative studies. Following the process of selection, nine studies met the inclusion criteria and were included in the meta-analysis (17–25). The major characteristics of the nine eligible publications are shown in Table I. These studies were performed in seven countries and published between 1980 and 2014. The total number of patients included the present meta-analysis was 1,412, ranging from 27 to 596 patients per study (median, 157). Six studies with a total of 644 patients used OS as the primary endpoint, whereas four studies with a total of 795 patients used 5-year DFS (one study contained data for OS and 5-year DFS). Quality assessments revealed average NOQAS scores from the two reviewers of 7.0 and 7.5, indicating that all nine eligible studies were of moderate quality.

No significant heterogeneity existed across the OS (I²=6.3%, P=0.376) or the 5-year DFS (I²=16.0%, P=0.311) studies; therefore, the fixed effect model was used. Overall, elevated levels of LDH expression were associated with poor prognosis in patients with EWS during the follow-up (OS studies: HR=2.90, 95% CI 2.09–4.04, P<0.001; 5-year DFS studies: HR=2.40, 95% CI 1.93–2.98, P<0.001) (Fig. 2). Sensitivity analysis suggested that the pooled HR was stable, and omitting a single study did not change the significance of the pooled HR (Fig. 3).

The assessment of publication bias in the selected literature was performed using Begg's test and Egger's test (19,20,22–25). The funnel plot for this meta-analysis revealed evidence of symmetry, and the P-value from the Egger's test was 0.707 in the OS studies (Fig. 4). Thus, there was no significant publication bias risk in the meta-analysis.