Patient-Level Meta-Analysis of End-of-Therapy PET-CR as a Surrogate Endpoint for PFS and OS in Patients with Previously Untreated DLBCL: Implications for Clinical Trial Design

Introduction:

The use of PET-CR as a surrogate endpoint would expedite the development of novel therapies and enable better estimates of sample size based on early outcomes of a trial. Previous studies have reported an association between end-of-therapy (EOT) PET results and long-term progression-free (PFS) and overall survival (OS) in diffuse large B-cell lymphoma (DLBCL) patients receiving standard first-line chemoimmunotherapy. We have also previously shown that the overall predictability of PET-CR for PFS and OS in these trials is similar to that in 18 literature-based studies. (15-ICML 2019, P195). To assess the potential for PET-CR as a surrogate endpoint in registration trials, we conducted a prospectively designed individual patient-level-data meta-analysis of available clinical trials.

Methods:

We synthesized patient-level data from three prospective phase II and III trials (GOYA [NCT01287741], GATHER [NCT01414855], MAYO [NCT00670358]) conducted in previously untreated DLBCL patients, using a Bayesian hierarchical model. We considered the two treatment arms in GOYA (GOYA-R-CHOP [rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone] and GOYA-G-CHOP [obinutuzumab, cyclophosphamide, doxorubicin, vincristine, and prednisone]) separate; hence a total of four arms, excluding patients without baseline PET scans. We investigated the relationship between PET-CR and long-term survival outcomes overall and within patient subgroups. We used Kaplan-Meier plots to compare survival endpoints by PET-CR status. We considered hypothetical RCTs that have PET-CR and PFS as endpoints to show how our model of the relationship between PET-CR and PFS can be used to predict the trial outcome.

Results:

We included 1496 patients (GOYA-R, GOYA-G, GATHER, MAYO, respectively: 665, 669, 100, 62). EOT PET-CR status was determined by the Lugano criteria in GOYA and by IHP in the other two studies. The overall rate of PET-CR was 72%; respectively 72%, 73%, 62% and 74% per arm. Panel A of the Figure shows Kaplan-Meier plots of PFS by PET-CR status. The Bayesian modeled hazard ratio (HR) comparing PET-CR versus nonCR was 0.13 (95% CI: 0.10, 0.15). The model can be used in planning a clinical trial as follows. Suppose a new therapy improves the PET-CR rate from 70% to 85%. Based on our model the expected HR for PFS treatment effect would be 0.71 (95% CI: 0.58, 0.84). The trial sample size required to demonstrate such an improvement with 90% power is about 650 patients, assuming an accrual rate of 50 patients per month and a minimum follow-up time of 12 months. Panel B of the Figure shows the model-based PFS HR and its CI (shaded) as it depends on the PET-CR rate of the new therapy. This plot also shows the estimated total trial sample size (in red), again assuming the relationship between PET-CR and PFS in our model. However, our model may not apply for the new therapy; and the trial could have an adaptive design with final sample size tailored to the accruing information about PET-CR and PFS and their relationship for the therapies in the trial.

Conclusions:

Achieving an EOT PET-CR in newly diagnosed DLBCL is highly predictive of favorable outcome in the populations we considered. The estimated HR is 0.13 for PFS, PET-CR versus nonCR, and it is 0.10 for OS. Based on our model, a treatment that improves PET-CR rate could be reasonably expected to have a benefit on PFS and OS, for the populations we considered. Whether a new therapy, with a different mechanism of action, that improves PET-CR rate will extend PFS in a clinical trial is less clear. An adaptive trial could be initiated based on our model as a hypothesis. This hypothesis can be verified or updated using accruing information in the trial itself. Our model can also help in planning clinical trials for re-estimating sample size during the trial and considering early stopping for futility. Further work should investigate the applicability of PET-CR in predicting PFS and/or OS for treatments that have different mechanisms of action and for other populations of patients.

Acknowledgements: This work was funded by Genentech/Roche.

View largeDownload slide

View largeDownload slide

Close modal