Presentation (CTI)

OBJECTIVES: Recent advances in treatment of R/R CLL may lead to delayed collection of mature OS data, underscoring the need to establish reliable SEs. This study evaluated CR and PFS as SEs of OS in R/R CLL at the trial-level.
METHODS: Analyses were based on a previously published evidence base comprising 19 randomized controlled trials. SE-OS associations were assessed separately using bivariate random-effects meta-analyses (BRMA). Across 13 trials reporting PFS, OS, and evaluable treatment effects on CR, observed OS HRs were compared with predicted OS HRs from (1) BRMA using CR benefit, (2) BRMA using observed PFS benefit, (3) a two-step approach in which PFS benefit was predicted from CR benefit using a published model, which was then used to predict OS benefit using the PFS-OS BRMA, (4) a published Cox PH model using arm-level differentials in CR, partial response (PR), and prior LoT distribution.
RESULTS: CR-OS association was weak and negative (−0.16; 95% credible interval [CrI]: −0.95-0.92), whereas PFS-OS association was moderate and positive (0.59; 95% CrI: −0.86-1.00). Visual patterns in treatment effects on SEs and OS, and model slopes (−0.007 for CR, 0.05 for PFS), were consistent with the degree of corresponding associations. Mean absolute gaps between observed and predicted OS HRs were 0.13, 0.11, 0.12, and 0.17 for approaches (1)-(4), respectively, indicating comparable predictive accuracy for approaches (1)-(3) and slightly weaker predictive accuracy for approach (4).
CONCLUSIONS: PFS demonstrated a closer correlation with OS than CR. Incorporating PR, prior LoT distribution, and temporal OS trends alongside CR in a PH model did not improve OS benefit estimates beyond BRMAs, potentially reflecting the limitations of applying arm-level response proportions uniformly across patients. Further research is needed to strengthen the evidence supporting CR and PFS as surrogate endpoints for OS in R/R CLL.