Abstract

There is currently strong interest in the possibility of utilising historical placebo (or other control) data to supplement the data obtained in a new clinical trial, thereby increasing its power and/or reducing its size. It is generally considered that historical data should be discounted relative to data from the new trial, but determination of the appropriate degree of discounting is a major difficulty. This presentation provides methods for assessing the consequences of different choices or decisions. The discount can be expressed in terms of the difference between the historical and future means, represented by an offset variance component tau-squared. In a Bayesian analysis, there must be a strong prior belief that is small if much value is to be obtained from the historical data, but a range of such priors can be explored in a sensitivity analysis. Alternatively, the same numerical results can be viewed in terms of minimisation of the mean square error when biased historical data are included in the estimate, and a range of values of a bias parameter, |delta|, can be explored. The choice of value for tau-squared or |delta| can have a substantial effect on the inferences made, yet a wide range of values may be consistent with the data. Fortunately, external information can sometimes provide additional guidance on the appropriate value. On this basis, it is likely that historical data can provide a valuable contribution to exploratory studies in drug development, where bias can be accepted as one of the research project’s risks.