Presentation (CTI)

OBJECTIVES: Treatment sequencing models in oncology allow for evaluation of alternative subsequent treatment options to those observed in the key clinical trial being evaluated, with different treatment options with each progression. Using a hypothetical example, we explore the challenges and uncertainty when combining trial data and external studies to model treatment sequences.
METHODS: We built a four-state semi-Markov model (progression-free [PF], progressed disease [PD]1, PD2, death) in R. Five time-to-event endpoints informed disease transitions, all using a 'clock-reset' assumption modelled via tunnel states. A treatment effect hazard ratio (HR) of 0.78 (95%CI 0.65-0.93) was applied to derive intervention arm PF. For the comparator (switch) arm during PD1, HRs of 0.75 (0.55-0.95) and 0.82 (0.68-0.93) were applied to the progression and death transitions respectively, derived from hypothetical trial and external sources. Outputs were probabilistic restricted mean survival times (RMST) for overall survival. Incremental RMST (intervention vs. switch) was estimated across key assumptions.
RESULTS: RMST was 31.6 months (95%CrI 27.0-36.9) for intervention and 30.0 (25.8-34.7) for switch, with 1.6 months incremental RMST (-0.7-4.4) in the base case. Incremental RMST was sensitive to the switch treatment effect during PD1 (-0.1-2.6 months) but less sensitive to baseline hazards for transitions from PD1 (0.8-2.0 months). Results were sensitive to whether transitions were independent of PF treatment (6.1 vs 1.6 months).
CONCLUSIONS: The relative treatment effect of a subsequent switch substantially impacts incremental sequencing outcomes. The baseline hazard for later transitions had a modest impact when equivalent across comparators but had a considerable impact when these differ by PF treatment received. The evidence source for post-progression hazards (trial vs. external), and whether these differ by PF treatment received, warrant sensitivity analyses in treatment sequence models. Overall, higher post-progression hazards relative to pre-progression will diminish the benefit associated with switching.