OBJECTIVES: The ECHELON-1 trial (NCT01712490) demonstrated superior progression-free survival for brentuximab vedotin plus doxorubicin, vinblastine, and dacarbazine (A+AVD) versus doxorubicin, bleomycin, vinblastine, and dacarbazine (ABVD) in patients with frontline stage III/IV Hodgkin lymphoma with a median follow-up of 2 years and superior overall survival (OS) with a median follow-up of 6 years. The objective of this study was to compare the long-term survival extrapolations using trial data from early vs late datacuts.

METHODS: A Markov cohort model with three health states (pre-, post-progression, and death) was developed to extrapolate survival of A+AVD vs ABVD over a 70-year time horizon. Transition probabilities were informed by the 2-year and 6-year datacuts from ECHELON-1. Before a specified cure timepoint, observed treatment-specific time-to-progression (TTP) and time-to-death (TTD) data were applied. Afterwards, extrapolation was based on the last observed follow-up point for TTP and general population mortality (UK) accelerated by an excess mortality rate for TTD. A constant transition probability was applied using a joint exponential model with or without a post-progression (PPS) treatment effect. Life-years (LYs) were discounted at 3.5%.

RESULTS: Extrapolation based on the 2-year data underestimated 7-year OS for both treatments when compared with the observed 7-year data and underestimated the difference in landmark OS between treatments (3.55% vs 4.6%). When the 6-year data were used, incremental OS at 7 years more closely matched the observed data (4.36% with PPS treatment effect). Over the lifetime time horizon, longer LYs were generated with extrapolations based on 6-year versus 2-year data: 18.60/18.49 (with/without PPS treatment effect) vs 17.81 for A+AVD; 17.61 vs 16.97 for ABVD. Incremental LYs of A+AVD vs ABVD also improved: 0.99/0.88 vs 0.84.

CONCLUSIONS: When based on the latest datacut from ECHELON-1, absolute and incremental survival estimates improved vs the early datacut, indicating that earlier estimates were underestimated.