OBJECTIVES: R/R-LBCL is associated with historically poor prognosis (median overall survival [OS]=6.3 months). In the two-year update of the ZUMA-1 trial of axi-cel, median OS was not reached after a median follow-up of 27.1 months. Prior analysis at one-year suggests that traditional parametric survival models cannot accurately estimate long-term OS with axi-cel when a large proportion of patients achieve durable remissions, thereby supporting the use of mixture cure model (MCM) for survival extrapolations (Bansel 2018). We analyzed the updated two-year ZUMA-1 data to further explore the robustness of MCM.

METHODS: The ZUMA-1 data were analyzed using traditional parametric model, MCM, general mixture model (GMM), and integrated Markov cure model (IMCM). The MCM assumed patients with long-term OS would have similar mortality as the age-adjusted general population. The GMM relaxed this assumption by separately fitting and then combining parametric survival functions for poor- and good-prognosis patients in a single survival function. The IMCM simultaneously accounted for OS and disease progression in a single framework. All parameters were estimated via maximum likelihood estimations and model fit was assessed via Akaike information criterion (AIC).

RESULTS: The best-fitting traditional model estimated a mean OS of 10.6 years (AIC=444.3). The best-fitting MCM estimated a mean OS of 13.5 years with a 51.0% long-term OS fraction (AIC=439.3). The best-fitting GMM produced a mean OS of 13.5 years with a 52.7% long-term OS fraction (AIC=442.1). The best-fitting IMCM estimated a mean OS of 13.4 years with a 50.5% long-term OS fraction (AIC=916.6). The higher AIC for GMM and IMCM is due, partly, to specifying more parameters than MCM.

CONCLUSIONS: Analyzing the updated ZUMA-1 two-year data, both GMM and IMCM showed good model fit and consistent results to MCM. These results support the long-term survival associated with axi-cel in R/R-LBCL, driven by ≥50% long-term OS rates.