OBJECTIVES: Individual Participant Data (IPD) meta-analysis is recognized as a gold standard in evidence synthesis due to its enhanced accuracy and flexibility. Unlike traditional aggregate data meta-analyses, IPD meta-analysis uses raw, individual-level data from multiple studies, enabling precise estimation of treatment effects and advanced subgroup analyses. Despite many comparisons, uncertainty remains about when to use the one-stage or two-stage IPD meta-analysis. We describe and compare one and two-stage IPD meta-analysis methods to assess the effectiveness of treatments on time-to-event (TTE) outcomes.

METHODS: This study utilized one-stage and two-stage meta-analysis techniques to process IPD from two phase 3 trials in metastatic breast cancer, evaluating overall survival (OS), progression-free survival (PFS), and time-to-deterioration (TTD) on EORTC QLQ-C30 and EQ-5D instruments. The treatment effects were estimated using a stratified Cox proportional hazards regression analysis. The one-stage IPD meta-analysis pooled raw data from two studies into a single model, adjusting for clinically and statistically relevant covariates. The two-stage IPD meta-analysis analyzed each study separately with relevant covariates and then combined the summarized results.

RESULTS: The trials included in the IPD meta-analysis were sufficiently similar in terms of trial and patient characteristics. Both one-stage and two-stage models consistently estimated treatment effects on survival, PFS, and TTD, favoring the intervention. The one-stage IPD meta-analysis was chosen as the base case due to the limited number of studies in the analysis. The alignment of both models, both directionally and statistically, underscores the robustness of the estimates.

CONCLUSIONS: The one-stage and two-stage models used in this IPD meta-analysis effectively captured and validated treatment effects. For a nuanced analysis, the one-stage approach is preferable for IPD meta-analyses of fewer studies, while the two-stage approach is better for large meta-analyses in the presence of heterogeneity. Where possible, both one-stage and two-stage approaches should be used to leverage their complementary strengths for TTE analysis.