Statistical evaluation in non-randomized clinical trial
Randomized controlled trials on evaluating the intervention effect can provide higher quality data and reduce the treatment biases, compared with other study design. Its the international recognized gold standard to evaluate whether the intervention measure is effective. In the RCT study, Subjects get randomized to different treatments. On average, no systematic differences (bias) in observed and unobserved covariates between subjects assigned to different treatments. However, RCT study is not applied to all the trials, especially for medical device clinical trials, because of ethnic or clinical feasibility problem. At this time, we tend to apply the non-randomized controlled clinical trial. According to the number of control groups, there are usually two kinds of non-RCT research, one is Parallel Group Trial, and the other is Single Arm trial. In the Parallel Group Trial, data in each arm will be prospectively collected, and through the trial we get clinical endpoints. Due to nonrandomization of the group assignment, baseline information in two groups may not be balanced. So we should use some statistical method to adjust the imbalance of concomitant variable between two groups. The traditional method is to use multiple regression to adjust the confounding, its said that we introduce some confounding variables as covariant to regression equation, so as to achieve the purpose of the balance between groups. Another method is to calculate the Propensity Score.
Jin Guo
National Center for Cardiovascular Diseases Fu Wai Hospital & Cardiovascular Institute Peking Union Medical College
国际会议
Second Joint Biostatistics Symposium(第二届生物统计国际研讨会2012)
北京
英文
47-49
2012-07-08（万方平台首次上网日期，不代表论文的发表时间）