OBJECTIVES: In the presence of EM, the consistency assumption in network meta-analysis (NMA) may be violated, leading to biased estimation of treatment effect. In the setting of first-line aNSCLC, we sought to systematically evaluate clinical sources of EM.

METHODS: We conducted a systematic literature review of first-line aNSCLC RCTs assessing efficacy of systemic anticancer therapy. We extracted baseline characteristics; stratification factors used for randomization; two efficacy endpoints (progression-free survival (PFS) and overall survival (OS)) and all hazard ratios for all subgroups presented by study authors. The ten most frequently reported within-trial subgroups were systematically evaluated as potential effect modifiers, for each pairwise comparison, using: within-trial subgroup analysis; between-trial subgroup meta-analysis; and meta-regression using log-transformed treatment effect estimates against aggregate-level patient characteristics.

RESULTS: We identified 36 eligible RCTs involving 11 pairwise comparisons. 15 RCTs reporting at least one subgroup were included in the within-trial subgroup analysis; 27 RCTs were included in the between-trial subgroup meta-analysis; and 32 RCTs were included in the meta-regression and between-trial analysis. Within-trial analysis showed the potential of EM by histology in pemetrexed-containing RCTs, and of PD-L1 expression thresholds and tumor mutational burden for RCTs with immune checkpoint inhibitors. For between-trial meta-analysis, bevacizumab RCTs revealed potential EM by Asian race. Meta-regression did not reveal signs of EM; models generally produced implausibly steep slopes based on few trials and narrow ranges. Other sources of EM (e.g. sex, smoking status) were inconclusive and potentially correlated with histology.

CONCLUSIONS: These hypothesis-generating findings around EM should be considered in the design of NMA of PFS and OS in the first-line aNSCLC setting, to avoid violating the consistency assumption. A limitation of the research is the inability to evaluate unmeasured and unknown factors that may lead to EM.