Tutorial in biostatistics Bayesian data monitoring in clinical trials

Many clinical trials organizations use regular interim analyses to monitor the accruing results in large clinical trials. In disease areas such as cancer, where survival is usually a major outcome variable, ethical considerations may lead to a stipulated requirement for data monitoring of mortality. This monitoring has frequently taken the form of limiting interim analyses to be few in number, and specifying an extreme p-value of, for example, p < 0.001 or p < 0.01 as grounds for early termination of the trial. Group-sequential methods are also used. However, none of these approaches formally assesses the impact that the results of a clinical trial may have upon clinical practice. Thus a trial might be terminated early because of apparent treatment benefits, but might fail to influence sceptical clinicians to modify their future treatment policy. We discuss the application of Bayesian methods, including the use of uninformative, sceptical and enthusiastic priors, and demonstrate that the necessary calculations are both straightforward to perform and easy to interpret statistically and clinically. Methods are illustrated with interim analyses of a clinical trial in oesophageal cancer.