OBJECTIVES : Surrogate predictors may allow earlier assessment of clinical efficacy when overall survival (OS) requires prolonged follow-up times. This study devised a frequentist approach to cope with the absence of individual-level patient data (IPD) in estimating correlation between progression-free survival (PFS) and OS, and presented a case-study from first-line glioblastoma multiforme (GBM).

METHODS : A literature search for recent randomized controlled trials for newly diagnosed GBM identified 13 studies (n = 5971) published between 2005-2019, 2 of which included only MGMT-methylated and 4 of which included only MGMT-unmethylated patients. For each study, pseudo-IPD was reconstructed using the WebPlotDigitizer version 4.2 after digitizing each endpoint’s Kaplan-Meier curve. Using the pseudo-IPD and the dependent modeling framework proposed by Fleischer et al. (2009), Pearson correlation coefficients between PFS and OS were calculated via rates of time-to-progression, and pre- and post-progression death which were derived by maximum-likelihood estimation. Predictive accuracy of the model was assessed by comparing estimated PFS and OS rates with 95% confidence intervals of the observed PFS and OS rates from the trials.

RESULTS : Of the 13 studies consisting of 37 arms in total, 2 were considered to have high association between the endpoints, 16 were considered to have low association, and the remaining 19 had medium association per the IQWiG criteria. The strength of association depended on treatment, subgroup, and sample size. The sample average of the correlation coefficients across all studies was 0.73 in the intent to treat population, 0.67 in MGMT-methylated population, and 0.68 in the MGMT-unmethylated population. The respective prediction accuracies for PFS and OS were 75% and 80% on average across the evidence base.

CONCLUSIONS : The analyses show a moderate relationship between PFS and OS in the first-line GBM setting. Due to simplifying assumptions in the study (e.g. time-homogeneous hazards), results should be approached with caution.