Objectives/Goals: To develop a novel meta-analysis model that estimates pooled risk ratios (RR) at prespecified milestones, accounts for unequal follow-up across trials, and leverages between-time correlation, thereby providing clinically interpretable estimates under potential non-proportional hazards (PH). Methods/Study Population: Our model combines study-specific RRs – the ratios of cumulative event probabilities – from Kaplan-Meier curves at prespecified times (12, 24, 36 months, etc.). Using the delta method, we derive within-trial variances for log-RRs and cross-time covariances to reflect correlation among milestones. These study-level estimates and variance-covariance matrices enter a multivariate random effects model to obtain pooled RRs. We illustrate this method with a meta-analysis of progression-free survival in 7 randomized clinical trials of first-line therapy immunotherapy combinations in metastatic renal cell carcinoma. Results/Anticipated Results: Follow-up for PFS across the 7 trials was heterogeneous: all 7 trials contribute at 12 months, 6 at 24 months, 5 at 36 months, 4 at 48 months, and only 2 contributing at 54 and 60 months. This heterogeneity motivates a multivariate approach that borrows strength across times. PH diagnostics indicate deviations from PH for PFS in 3 trials, supporting the need for pooled effects at milestones rather than a single summary measure. We will report pooled RRs at 12, 24, 36, 48, 54, and 60 months. We expect the multivariate model to improve precision at shared milestones while honestly reflecting increasing uncertainty at sparse later times. Discussion/Significance of Impact: When follow-up differs across trials and PH is questionable, a multivariate RR meta-analysis preserves interpretability and efficiently uses all available milestones without forcing a single-number summary. This approach clarifies when benefits accrue and provide clinically interpretable results.
Reddy et al. (Wed,) studied this question.