OBJECTIVES: Overall survival (OS) data from randomized clinical trials might be immature at the time of submission. External data can be used to inform the long-term OS extrapolations. We compared four methodologies where historical trial data is used to guide survival extrapolations in MM.

METHODS: An immature trial in previously untreated MM patients comparing active treatment bortezomib (V) combined with melphalan and prednisone (MP) and placebo MP with maximum follow up of 30 months was informed by mature historical data on MP with max follow up of 100 months. Four Bayesian methodologies were compared: (1) the shape of historical data is used as prior for the shape of the pivotal trial (informed shape), (2) the historical data is added as a treatment arm in the parametric model fit (informed fit), (3) expert elicitation based on historical data was used as prior (informed survival -> the expected survival was estimated to be between 10% and 20% after 70 months for placebo), and (4) a combination of (2) and (3). All methods were compared with standard parametric methods (non-informed) using Weibull. The predictions were validated based on incremental restricted mean survival (RMS) comparing mature (maximum follow up of 70 months) of the VMP and MP trial with the predicted survival.

RESULTS: RMS of the validation data set was 43 months for MP and 50 months for VMP. The ΔRMS with non-informative priors is 1.94 and 3.06, informed shape 0.06 and 1.52, informed fit -0.04 and 1.47, informed survival -5.31 and -0.54, informed survival and fit -5.39 and -0.47 respectively.

CONCLUSIONS: Clinically meaningful extrapolations can be improved when informed by historical trial data. The impact of informed survival on placebo was largest as expert opinion expectation was set at point estimate of historical data whereas the other methodologies apply a proportional hazard assumption compared to historical data.