*2.3. Statistical Analysis*

The DerSimonian–Laird random effects models for main and subgroup analyses was implemented, assessing heterogeneity of effect-size estimates from the individual studies by Cochran's Q test and the I2 statistic. Additionally, MA corresponding to analysis of binary data of proportions was also performed using a DerSimonian–Laird random effects model without transformation of the proportion. A high level of heterogeneity was considered if I2 was greater than 50%. Due to the relatively low number of trials involved in this MA, values and significance of heterogeneity must be considered as guidance only [26]. Statistical significance was reached for *p*-values less than 0.05. Analyses were not controlled for multiplicity; no alpha was assigned to the different analyses. The nature of this study is therefore exploratory, mainly in the subgroup analysis. Hazard ratios (HRs) and 95% confidence intervals (CIs) for OS and PFS from the overall population and subgroups from each individual trial of advanced NSCLC were calculated. For dichotomous data, odds ratios (ORs) were estimated. The MA was performed using Open Meta Analyst v. 10 (Center for Evidence Synthesis in Health, Brown University). Recommendations of the Cochrane Collaboration and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were followed for this MA [27].

For the ORR, different endpoints, including complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD), were alternately modeled. These sensitivity analyses along with those for OS and PFS did not quantitatively alter the results and conclusions of the main analyses.
