Keywords: Bayesian method, Probability of study success
Developing informative and efficient statistical methods for go/no-go decisions is crucial for clinical trials, as planning late phase trials is costly and time consuming. Bayesian approach has become quite common in clinical trials practice to formulate the go/no-go decisions on the basis of the data of early phase trial to try and quantify the benefits of moving to the late phase setting, especially in prediction of study success in late phase trial.
To quantitatively assess the study success, the predictive probability statement is generally adopted. The probability of study success has been treated as an unconditional power in Bayesian methods based on the distribution of the parameters, and it can be calculated through a weighted success probability by using historical data or interim data. It is also implemented by step process using simulations thus it could be applied in the situations when the explicitly distribution function do not exist. The limitation of the method is the absence of covariates and their contributions in determining the parameters and eventually the probability of study success. Also, most of these aforementioned approaches promote a hybrid approach, that is, a combination of Bayesian and frequentist methods. By introducing the fitting prior and validation prior, the probability of study success could be calculated after the inclusion of covariates through a covariate distribution as well as historical data. We will present few data examples to illustrate those methods and discuss the pros and cons.