Interim analysis for randomized clinical trials: simulating the predictive distribution of the log-rank test statistic.

This paper presents an analog to stochastic curtailment, where the simulation of the remaining trial data from interim analysis is constructed using the existing data, rather than the design planning parameters. The nonparametric approach for the predictive distribution as proposed by Berliner and Hill (1988, Journal of the American Statistical Association 83, 772-779) forms the basis of the procedure. The methods are applied retrospectively to a clinical trial in childhood cancer.