Abstract:
|
Treatment switching, also called crossover, is a common issue seen in clinical trials for oncology drugs and can lead to confounded estimation of long-term endpoints, e.g., overall survival. For example, after disease progression occurs, patients randomized to the control arm may switch to off-label use of the investigational agent being studied or switch to agent in the same drug class already approved for the disease. Another example occurs when a study reads out positive on an intermediate endpoint like progression free survival, and control arm patients are offered the opportunity to switch therapies. This later occurrence is often called administrative crossover. There exist several methods to account treatment switching including marginal structural models with inverse probability censoring weighting, two-stage estimation, rank preserving structural failure time models, iterative parameter estimation, and Bayesian competing risk models. We borrow from previous work in the Bayesian competing risk space and expand a model to account for administrative censoring.
|