Online Program

Use of Hybrid Control Arms with an Individually-Weighted Adaptive Power Prior for Time-to-Event Data (302401)

*Evan Kwiatkowski, University of North Carolina at Chapel Hill 
Jiawen Zhu, Genentech, Inc. 
Devin Incerti, Genentech, Inc. 
Xiao Li, Genentech, Inc. 
Herbert (Herb) Pang, Genentech 
Gracie Lieberman, Genentech, Inc. 
Matthew Psioda, University of North Carolina at Chapel Hill 

Keywords: Historical controls, Hybrid designs, Power prior, Real world data

We develop a hybrid design that uses external controls from real world data (RWD) to augment internal control arms in randomized controlled trials (RCTs) where the degree of borrowing is evaluated based on similarity between RCT and RWD patients to account for systematic differences (e.g. unmeasured confounders). We develop a novel extension of the power prior where the discounting weight is computed separately for each external control subject based on compatibility with the randomized control data. The discounting weights are assigned using the predictive distribution for the external controls derived via the Bayesian posterior distribution for time-to-event parameters estimated from the RCT. This method is applied in an example based on a completed trial in non-small cell lung cancer using a proportional hazards model with piecewise constant baseline hazard, and is compared to propensity score methods, frailty models, and the commensurate prior approach. It is shown that the individually-weighted adaptive power prior provides robust inference under various forms of heterogeneity in the external control population.