The importance of evaluating long term responder fraction using complementary statistical approaches in immune checkpoint inhibitors phase III trials: An in-silico study using keynote 045 study.

Abstract

e16512 Background: Immune-checkpoint-inhibitors (ICI) provide durable antitumor activity even for recurrent and/or metastatic disease. As illustrated in the Keynote-045 study (pembrolizumab (pembro.) in second line metastatic urothelial carcinoma; Fradet, 2019), improvement in long-term responder fraction does not always translate into progression-free survival (PFS) improvement. Despite a long-term benefit improvement (2 years PFS: 12.4% vs 3%), there was no significant difference in PFS (HR 0.98; 95% CI, 0.81-1.19; p=0.42). Benefits in terms of OS were not deemed sufficient by the French health authorities in charge of health technology assessment and reimbursement of medicinal products to consider that pembro. brought a significant improvement upon existing drugs in urothelial carcinoma. Thus, although recommended by scientific societies, this treatment has only been available in France since January 2020. The main objective is to illustrate the importance of using complementary statistical approaches when evaluating ICI phase III trials. Methods: PFS curves of the Keynote-045 study available in publications (Bellmunt, 2017; Fradet, 2019) were digitized to reconstruct pseudo individual patient data as per established methods. Survival rates were estimated using Kaplan-Meier method and comparison between groups were performed using Cox model. Flexible parametric survival models with cure (FPCM) were then used to estimate the treatment benefit in terms of long-term responder fraction and evaluate the treatment effect on PFS in the non–long-term responder population. Results: In both first and updated data, there were no significant between-group difference in the duration of PFS in the total population (2017: HR=0.95, 95%CI 0.79–1.15; 2019: HR=0.93, 95%CI 0.77–1.11). Using FPCM, we demonstrated that the long-term responder fraction (LRF) was higher with pembro. compared to chemotherapy (2017: Pembro 14.0% 95%CI 10.2-18.3; Chemo.: 7.7% 95%CI 4.9-11.2; 2019: Pembro 11.0% 95%CI 7.8-14.9; Chemo.: 4.6% 95%CI 2.7-7.3). Differences in LRF were estimated to 6.3% (95%CI 1.5–11.13) and 6.4% (95%CI 2.4-10.4]) in the first and updated analysis, respectively. In the non–long-term responder population, ICI was first associated with a higher risk of progression or death and then a lower risk. Conclusions: While ICIs show a substantial shift towards long-term benefit in terms of PFS, greater emphasis should be placed on rigorously evaluating this quantity instead of only reporting Kaplan-Meier estimation or comparing PFS curves using classical Cox model. Despite the long-term responder fraction improvement was constant between analysis, relative variations in LRFs were observed. This suggests that the follow-up of this study was not long enough to properly characterize long term responders.