MATERIALS AND METHODS
Literature search strategy
This meta-analysis was reported according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.13 Systematic computerized searches of PubMed, Embase, and Cochrane data bases for reports dated up to March 26, 2017 were performed with the following search terms: “cyclooxygenase-2 inhibitor”, “COX-2 inhibitor”, “non-small-cell lung cancer”, “NSCLC”, “chemotherapy”. All reference lists from the trials selected by electronic searching were scanned to further identify relevant trials. Ethical approval was not required for this study.
Literature selection and exclusion
The following criteria were used for study selection: 1) the RCTs compared the efficacy and safety profile of adding COX-2 inhibitors to chemotherapy alone; 2) only including patients with cytologically or histologically confirmed NSCLC stage IIIB or IV; 3) full paper in English language was published; and 4) studies needed to have measured at least one of the following outcomes as their end points: OS, PFS, 1-year survival rate (SR), ORR, and toxicities.
If a study was a duplicate or the study’s data could not be extracted or obtained through contact with the author, it was excluded from our analysis.
The final articles included were independently assessed by two authors. In the case of disagreement, another author was consulted to resolve the dispute, and a final decision was made by majority vote. The relevant information included study design, patient characteristics, interventions, controls, and outcomes. For some missing survival indices such as OS and PFS, HR and 95% confidence interval (CI) were extracted from the survival curve.14 Regarding toxicity, we considered both hematological (leukopenia, thrombocytopenia, and anemia) and nonhematological (nausea/vomiting, diarrhea, asthenia, and cardiotoxicity) grade 3 and grade 4 effects of treatment.
Quality assessment of included studies
Two investigators independently evaluated the methodological quality of eligible trials using the Cochrane collaboration tool for assessing risk of bias15 (random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias).
Dichotomous data, including 1-year SR, ORR, and toxicities, were compared with a pooled risk ratio (RR) with a 95% CI. Survival indices of OS and PFS were expressed as HR with a 95% CI.16,17 This meta-analysis was performed using Stata 12.0. Heterogeneity between studies was also analyzed using chi-square tests, with the significance level set to P<0.1.18 No heterogeneity was observed when I2=0%. However, when I2>50%, studies were considered to have significant heterogeneity. If the data were homogeneous under a fixed-effects model, the type of COX-2 inhibitors and treatment line were identified as key sources of heterogeneity in the main outcomes (OS, PFS, 1-year SR, and ORR).18Heterogeneity was then dealt with using subgroups based on these modifiers. If the data were still heterogeneous, we introduced a random-effects model. Whereas when I2<50%, a fixed-effects model was used instead.18
A funnel plot was used to estimate potential publication bias, with an asymmetric plot suggesting possible bias.19 In the funnel plot, larger studies that provide a more precise estimate of an interventions effect form the spout of the funnel, whereas smaller studies with less precision form the cone end of the funnel. Finally, the Egger’s test was employed to address quantitative detection bias.20