Surrogate end points for overall survival: A meta-analysis exploring adequacy of progression-free survival and objective response in randomized NSCLC trials.

Abstract

e20702 Background: Surrogate endpoints such as progression-free survival (PFS) and objective response rate (ORR) are widely used in advanced non–small cell lung cancer (NSCLC) trials, yet their ability to reliably predict overall survival (OS) remains uncertain, particularly in the post-2015 era of immune checkpoint inhibitors and targeted therapies. We evaluated trial-level surrogacy of PFS and ORR for OS across randomized clinical trials using random-effects weighted meta-regression, with prespecified analyses by trial era. Methods: We conducted a systematic, trial-level meta-analysis of randomized studies in advanced or metastatic NSCLC reporting OS hazard ratios (HRs) and at least one surrogate endpoint (PFS HRs and/or ORR). The protocol was prospectively registered in PROSPERO (CRD420251266621). PubMed/MEDLINE, Embase, Scopus, CINAHL, and Web of Science were searched from inception through the final search date. Two reviewers independently screened records and full texts, resolving discrepancies by consensus or third-party adjudication. Treatment effects were analyzed on the log scale (log[HR] for OS and PFS; log[RR] for ORR), with standard errors derived from 95% confidence intervals when required. Trial-level surrogacy was assessed using random-effects weighted least-squares meta-regression, modeling log(HR_OS) as a function of log(HR_PFS) and separately log(RR_ORR). Era-stratified analyses (pre-2015, 2015, post-2015) and interaction testing were prespecified. Surrogacy strength was quantified using a weighted R² analogue. Small-study effects were evaluated using Egger’s regression. Analyses were performed using Python (version 3.14.2). Results: A total of 190 trials were included; 150 contributed to the PFS–OS analysis and 98 to the ORR–OS analysis. Overall, PFS showed a statistically significant but modest association with OS (β = 0.331; 95% CI, 0.230–0.433; p < 0.001), explaining limited between-trial variability (R² = 0.216). In post-2015 trials (k = 124), the association remained significant (β = 0.332; p < 0.001) but explained little OS variability (R² = 0.193), with no significant interaction by era. ORR demonstrated a weak and inconsistent association with OS (overall β = −0.256; p = 0.029), with non-significant associations in pre-2015 trials. Egger regression showed no evidence of small-study effects. Conclusions: Across randomized trials in advanced NSCLC, PFS and objective response demonstrate limited trial-level surrogacy for overall survival. Surrogacy strength did not meaningfully improve in the post-2015 era despite widespread adoption of immune checkpoint inhibitors and targeted therapies. These findings support cautious reliance on PFS or ORR as substitutes for OS in contemporary NSCLC trials and underscore the need for OS validation or context-specific endpoint selection.