Early stopping rules and long-term follow-up in phase III trials.

The main advantage of interim analyses is that a randomized trial can be concluded in advance. As a consequence, the cost is reduced and the protection of the individual is increased from an ethical point of view. Formal statistical methods are, however, necessary to control the overall significance level and the power of the trial. Group sequential analysis and triangular test properties are compared in the setting of randomized trials with a long-term follow-up. Potential biases secondary to the conduct of the first analyses on a small number of events are discussed. Conservative methods which employ stringent criteria for the first analyses should be used to avoid loss of power in subsequent comparisons of long-term survival when early stopping occurs. Early stopping leads to less precision in the estimation of the treatment effect as the size of the sample is reduced. Furthermore, the observed effect could be biased upwards when trials are stopped early because the effect is larger than expected and biased downwards because the effect is lower than expected. Finally, as the ultimate goal of a trial is to alter therapeutic strategies, data should always be consistent so that the medical community can be convinced that treatment management should be modified.