| BACKGROUND&OBJECTIVERandomization is one of the important principles of test (experimental) study design, in particular, randomized controled trial (RCT) for randomization have higher requirements. Methods RCT randomized studies are generally divided into dynamic randomization method and two types of non-dynamic randomization methods, including former urn, biased coin method, minimization method and dynamic balanced randomization, and the latter include simple randomization, block randomization, and stratified block randomization.Currently, the application of dynamic randomization has become a mainstream method used by the Institute of RCT. Consider the balance of a randomized trial, there are two aspects to consider. A balance between the two groups is the number of cases, and the other is the important controlling factors in the pre-test to determine the balance between the groups. The advantage of non-dynamic randomization is that the operation is easy, the management is also very convenient, the "random" feature in the group process keeps on a high level, and Researchers could not guess the result of a new patient’s group information. But the disadvantage of these categories randomization method is also obvious. Since the patient who has been enrolled is not considered, so the balance of the important factors is difficult to ensure a balanced. Some non-dynamic randomization method such as simple randomization method applicable to small sample size, the banlance of the number of patients between the two groups is also not as good as dynamic randomization method. The main advantage of minimization the law can consider the number of cases between the two groups and keep each important control factor in the balance between the two groups simultaneously. But the disadvantage is also obvious, specifically in the "forward", the Improper use under the conventional statistical test and the applicability of certain conditions in the complexity. Advantages of dynamic balanced randomization is simple calculation, no need to calculate complex parameters. In the small sample size of the case, the balance of the number of patients treated slightly better than minimization, the number of cases treated group gained more balanced. However, in the situation of many factors, it is difficult to ensure the balance of all factors. In all the algorithms described above, we did not find a specific algorithm can take the situation that the balance of important control factors in stratification factors into the final algorithm. Especially in multi-center clinical trial, it did not consider the balance of the center of each hierarchical internal control factors. In a multi-center clinical trial, if the resercher do not consider the balance of internal center, then after the end of the study, that balance of control factors in portion of the centers is likely to be very poor. Leading to the strong center effects and interaction effects and reduce the effectiveness of the final test. Similarly, in another study included other stratification factors, if the researcher do not consider hierarchical internal balance, the unbalanced will lead to heavy bias in the final result.In addition, the study also found that if we only consider the overall balance and neglect the balance of internal factors of each center can easily lead to imbalance of internal factors between two groups. If considering only within the center of balance is likely to come with the overall balance algorithm inconsistencies, thus affecting the overall balance of the control factors. While only consider the balance of the center will lead investigator of the research centers easier to deduce information about the next grouping patients according to the patient prior to the inside of the center of the group.To this end. the present study is to propose a new and improved minimization algorithm to ensure the overall balance of the premise,at the same time achieve internal balance of stratification factors, in order to meet the actual demand and provide a new dynamic randomized clinical trials effective tool.RESEARCH METHODSWe want to propose a new algorithm formula, and compared with the before minimization algorithm. It does not change the fundamental algorithms, but retaining the core of before algorithm, and add the internal balance of the center to the final algorithm models. It should be noted that the determination of the results of the last packet is composed of internal imbalance of the center where the new patient joined and the overall imbalance in two parts. Avoid potential problems consider only one aspect brought. Under the premise to ensure the overall balance of the centers while maintaining internal balance is maintained, while control the balance of the center to prevent the leakage of algorithms and blindind information.Using Monte Carlo techniques, compare the effect of the new minimization algorithm with other commonly used randomized algorithms. Simulation tools is SAS9.4, random number generation using ranuni uniformly distributed random numbers to generate the control factors with two classifications. Center’s number is 10, the block length is 6, sample size is set to 12 cases each center,30 cases,60 cases, 90 cases of four cases, the patients ratio of 1:1. The number of control factors are 2, and were considered to have three control factors and six control factors in the case. Because the simulation takes a long time. Therefore, the number of simulations in the analysis set for 1000.Center is the stratified factor in stratified randomization method. Minimization method to measure the two unbalanced degree algorithm is Range Method, the right to equal weight among the factors. Assigning probabilities to 0.8 and 0.2. Dynamic balanced randomization assumes that all factors are set threshold size 4. Evaluation of the final between the group’s overall difference between the groups and the control group factors minimum chi-square test P value of PI, P25. P50, P75 and P99 and sub-centers inside the corresponding index. Because the most extreme cases appear to be compared in a random process, so the difference between the center of the interior of the group selected within 10 center of the largest center difference, minimum internal control factors also select 10 centers three or six control factor P value of the minimum value. If it is found in the inspection in the case of the grid is less than 5. then replaced with exact test P-value alternative to chi-square test P value.RESEARCH RESULTComparison of results of different controlling factorsControlling factors for the three and six have similar results.6 Ways to the merits of the degree of change does not occur in the control of a number of different factors.Group balance between the number of cases:Similar balance between the number of cases within the center group and the number of cases among the overall balance of the results of the group of simple randomization. The number of cases among the overall group dynamic random four kinds of methods of control when the three factors are slightly worse than the controlling factors for the six case. Internal group in the number of cases of balance control factors of the center when the three were also slightly worse than the control factor is 6,while the results of the sub-center randomization are almost unchanged.Group balance between control factors:the overall and center internal of six kinds of randomized control methods in controlling the balance of factors when three factors are slightly better than the control factor is 6.Overall balance:The results of simple randomization method show the mean difference between groups increase obviously with increasing sample sizes. The rest of each percentile also showed the same trend. The overall P value of the result of the control factors is also less than ideal, and with increasing sample size, the value of P is worth almost unchanged. Stratified-oriented results reflect the algorithm itself with the characteristics. Because the role of random block, so the difference between the two groups can ensure balance within each block, thus ensuring the balance between overall group. Almost no difference with the control factors result in the patient’s overall results of simple randomization. and with increasing sample size, the value of the P has hardly changed. The results of minimization about overall balance between the number of groups of cases is very much better than simple randomization, and with increasing sample size, the results are hardly changed. Compared with the stratified block randomization, although it can not guarantee that all the results can be balanced between the groups, but the P95 and P99 are very small, the number of cases indicates the bias are very small and are acceptable. In the overall control factors with P values of all samples showed that the amount of P values were significantly greater than 0.05, and with the increasing sample size, mean P value also showed a growing trend. And in each case the sample size, P1 are far greater than 0.05. in 6 kinds of methods the original minimization method is optimal. Dynamic balanced randomization’s results show good balance of the number of cases on the whole group, which is much better than a simple randomization, compared to the original minimization method, the dynamic balanced randomization is a little bit worse. In the overall control factors in patients with P values showed that the method has obvious advantages compared to simple randomization and stratified randomization, and with the sample mean and other percentile increase, the amount of P values also showed growth trend. Compared with the original minimization method Dynamic balanced randomization’s results were lower in each sample. Sub-center randomization in each sub-centers are carried out when the respective party after randomization, overall balance still received more or less affected. Compared the mean difference between the integral assembly, we find it.s only better than simple randomization. In the overall balance control factors, sub-center randomized results are better than just simple randomization and stratified randomization. Of all four kinds of dynamic randomization method it,s the worst one. The mean difference between groups of improved method of minimization show with increasing sample size the result was almost unchanged, the rest of each percentile also hardly changed. Compared to a minimization and dynamic balancing algorithm, the results are relatively similar. The overall results of control factors show P value is also ideal, it is second only to the minimization of six methods, and with increasing sample size, P value also show growing trend.Internal balance in stratification factors:The results of simple randomization method show the mean difference between groups in stratification factors increase obviously with increasing sample sizes. The rest of each percentile also showed the same trend. The overall P value of the result of the control factors in stratification factors is also less than ideal. Stratified-oriented ensure the balance just like the overall result. Almost no difference with the control factors result in the patient’s overall results of simple randomization. The results is the best to balance the mean difference between groups in stratification factors, The results of minimization show no advantage compared to the simple randomization about either the mean or various percentiles, even no difference. With the increasing sample size, the difference between the two groups also showed relatively clear upward trend in the value of the P results of control factor. Dynamic balanced randomization’s results show good balance in sub-centers, the method is much better than the original simple randomization and minimization methods. In the balance of control factors, there is no different between dynamic balanced randomization and simple randomization method, and the trend is exactly the same. It’s evidence of the purely control the various control factors between whole group and the difference between the number of cases of sub-center group can not change the difference between the respective control factors component of the center of balance, which overturned the ues of a series of similar dynamic randomization Related algorithm in the control center of balance. In sub-center randomization,the result of the central mean difference between the groups is the best, and with the increasing sample size, the result are almost unaffected, indicating the control ability is very obvious. In the result of P value, the result is optimal. The mean difference between groups in stratification factors of improved method of minimization show with increasing sample size the result has an sight increasing trend, but the trend is not obvious, the rest of each percentile also hardly changed. In the six kinds of methods it is second only to sub-center randomization, and show significantly better than the original minimization algorithm and dynamic balanced randomization, it has good control of the balance between the number of cases of group sub-centers. The result of control factors in patients with P-value sub-centers also show ideal, and also the second beller one after sub-centers randomization, and with increasing sample size, P value is also worth growing trend, the results of P50 are also significantly higher than 0.05.CONCLUSIONSImproved minimization could control the overall balance and the internal balance in stratification factors at the same time, the balance was better than the other five methods. Meanwhile, the method in principle retained the core theory of minimization algorithm, so it’s conducive to the promotion of the current most central randomization system.. |
PDF Full Text Request