Sequential treatment allocation procedures in clinical trials--with particular attention to the analysis of results for the biased coin design.

We discuss the advantages and problems of using sequential allocation procedures for assigning treatments in a clinical trial. We use Monte Carlo methods to study the exact distribution of the fourfold chi-squared statistic for the case of no stratifying variables except sequence of entry. We display situations in which the distribution is well-approximated by the chi-squared distribution or by Efron's adjusted chi-squared and situations where these approximations fail. We provide suggestions for recognition and handling of situations in which both approaches fail.