The combination of randomized and historical controls in clinical trials

Abstract In many clinical trials the objective is to compare a new treatment with a standard control treatment, the design being to randomize equal numbers of patients onto the two treatments. However, there often exist acceptable historical data on the control treatment and this paper describes procedures for incorporating such historical controls into both the design and analysis of a randomized trial. The statistical model supposes that treatment evaluation consists of a single quantitative measure for each patient and the objective of the trial is to estimate the true difference in treatment means for this measure. In general, historical controls cannot be considered as reliable as randomized controls and this leads one to expect some bias in the historical data. This bias cannot be determined, even as regards its direction, and in the statistical model it is defined as a random variable with zero mean and variance to be specified. In practice, one might choose several values for this variance to represent varying degrees of mistrust, i.e. potential bias, in the historical data. As regards analysis, the best estimate of the control treatment mean is a weighted average of the means for the randomized and historical controls. This leads to a more accurate comparison with the new treatment than the use of randomized controls alone. In the design of a randomized trial the presence of historical data enables one to enter a reduced proportion of patients into a randomized control group, the precise amount of this reduction depending on the size of the historical data and also its potential bias. Examples from actual sequences of clinical trials run by the Eastern Co-operative Oncology Group illustrate the practical use of the methods. In conclusion, it is current practice in clinical trials to rely exclusively on either randomized controls or historical controls, but not both. The methods described in this paper provide an objective, quantitative approach for the combination of these two sources of control data and this should lead to a more efficient use of patients in the execution of clinical trials.