Comparison Of Five Minimization Methods On Balance, Predictability, Randomicity And Balance Stability

Objective To compare the balance ability, predictability, randomicity and balance stability for five minimization methods. Additionally, to establish linear models for predictability. Methods (1)36 kinds of simulations with varying sample size (N=48,96,192), probability to group leading to the best balance(P=0.7,0.8,0.9,1.0) and the number of groups(K=2,3,4) were conducted with 1000 samples for each kind randomly sampled from a clinical dataset consisting of 2000 patients to compare the win probability of comprehensive balance concerning on the balance of participants and four variables expected to keep balance across groups, predictability and randomicity for five minimization methods, namely goodness method(created by the designer of the study), range method, variance method, signal method and dynamic upper limit method, contrasted by complete randomization. Specially, center balance hasn't been taken count into consideration and weight for each factor was set to one in our simulations. (2)one fixed sample with 96 participants sampled from the same clinical dataset was obtained to observe the balance ability focusing on factors desired to keep balance and factor levels with smaller number of participants, two simulations(P=0.7,1.0,K=3) were conducted with randomly order for participants. (3) Stability on balance was compared with four fixed samples(N=48,P=0.7,0.85;N=72,P=0.7;N=96,P=0.9), fixed order and three groups. (4)Additionally, linear regression was used to establish predictability model for goodness and variance method. Results (1)Dynamic upper limit method is the worst on balance among the five minimization methods, range method, variance method and signal method on the difference of the number of participants across groups are close to each other. Goodness method outperforms other four minimization methods on win...