Expert elicitation of long-term survival (LTS) for patients (pts) with relapsed/refractory multiple myeloma (RRMM) treated with idecabtagene vicleucel (ide-cel, bb2121) in the KarMMa phase 2 trial

Background Cost-effectiveness analyses (CEA) of new treatments often need LTS extrapolations, but models can lack clinical expert evaluation. A systematic process can integrate expert opinion (EO) in an unbiased way. In the global multicenter, single-arm phase 2 KarMMa trial (NCT03361748), ide-cel, a B-cell maturation antigen (BCMA)-targeted chimeric antigen receptor (CAR) T cell therapy, showed deep, durable responses in triple-class exposed (TCE) pts with RRMM, with survival data reported up to 12 months (Munshi NC, et al. NEJM 2021;384:705-716). Without EO, predicted LTS varies with the extrapolation model. This study estimated expected survival rates at 3, 5, and 10 years for pts treated with ide-cel, based on integrating observed trial data with EO. Methods This prospective qualitative study incorporated semi-structured interviews, adapted from the Sheffield Elicitation Framework. Oncologists and hematologists with experience in treating pts with RRMM with BCMA-targeted therapies including CAR T cell therapies participated in the elicitation exercise (n=6). Evidence on pts in the KarMMa clinical trial from the Jan 2020 data cut (median follow-up 13.3 months) and conventional care (CC) for TCE pts with RRMM were summarized to give a common baseline data set for EO. In a facilitator-guided web-based application, experts gave upper and lower plausible limits and likely survival values at 3, 5, and 10 years. Experts were given the blinded, individual estimates and collectively gave consensus estimates from a ‘rationale impartial observer’ perspective. To assess the impact of incorporating EO, separate models were based on observed data with or without EO. Exponential, Weibull, Gompertz, generalized gamma (GG), lognormal, and log-logistic survival distributions were used to assess model fit. Analyses used a Frequentist framework and the Akaike information criterion compared the goodness-of-fit in survival models. This process estimated LTS for CC from the observational studies KarMMa-RW (Jagannath S, et al. JCO 2020;38;8525) and MAMMOTH (Gandhi UH, et al. Leukemia 2019;33:2266-2275). Results The best-fitting distributions for models using observed ide-cel (KarMMa) data only vs observed data and EO were Gompertz and GG, respectively. Survival estimates were 11% (without EO) to 35% (with EO) at 3 years; 0% (without) to 15% (with) at 5 years; and 0% (without) to 2% (with) at 10 years. In comparison, for the real-world CC cohort in KarMMa-RW survival estimates (with EO) were 16%, 6%, and 1% at 3, 5, and 10 years, respectively, for the best-fitting (GG) model. For the MAMMOTH CC cohort, survival estimates (with EO) were 5%, 0%, and 0% at 3, 5, and 10 years for the best-fitting (GG) model. Conclusion These findings suggest when observed data and EO are integrated, ide-cel treatment provides extended estimated survival rates vs CC. This study indicates that including EO is informative in assisting decision-making processes.