2006 年 27 巻 1 号 p. 69-79
We showed a dynamic allocation procedure to achieve a specified proportion of sample sizes among treatment groups not only within all patients, but also within patients with strata composed of centers and/or prognostic factors in randomized controlled clinical trials. This procedure allocates a patient to one of treatment groups that are within a pre-specified range from the best balance, as well as the treatment groups with the best balance. When target sample sizes of two treatment groups are the same, Zelen (1974) showed a restriction for a range of obtained sample sizes. Proposed method is an extension of Zelen's restriction to the case of different target sample sizes of more than two treatment groups. This method can introduce randomness into a dynamic allocation procedure keeping comparability among treatment groups.