Comparative Research Of Non-inferiority Evaluation Approachs In Chinese Medicine Clinical Trials

BackgroundAlong with the development of new drugs and new dosage-forms and new preparation technology, the clinical trials of chinese medicine which using non-inferiority design and analysis are increasing gradually. So it is significant to make a correct non-inferiority appraisal through a scientific and reasonable design and a correct statistical analysis method. At present, there are some methods to appraise the clinical non-inferiority. except frequency statistical approach (hypothesis testing and confidential interval), Bayesian approach receives some researcher’s favor in certain domains. There are some uncertaintyes in the non-inferiority clinical studies of Chinese medicine, but Bayesian statistics approach can make statistical inference using prior information, can deal with multiple hypotheses and complex models. So, If bayesian statistical approach be used in non-inferiority clinical trials of Chinese medicine, its non-inferiority testing result is in conformity with the result of frequency approach or not? How much application value of Bayesian statistical approach in non-inferiority clinical trial of Chinese medicine? These questions need to be discussed.Another aspect, there are many ordered categorical data in clinical trials. in normal conditions, these ordered categorical data were combined and transferred into binomial data, and then make non-inferiority or equivalence testing. While in the process of data transfer, it will lose some information. So, some scholars discussed non-inferiority or equivalence testing methods of ordered categorical data. In Chinese medicine clinical trials, the outcomes of TCM symptom can be resonded with the effect index or effect rank or overall response rate. Naturally, people will ask whether the non-inferiority testing result has significant different between these three outcomes of TCM symptom? which outcome is more advantageous in TCM clinical trials? These questions need researchers to seek for the answer from the practice.Objective1. In Chinese medicine clinical trial, whether it has significant difference between the result of non-inferiority testing of frequency statistical approach and Bayesian statistical approach. how to select these two methods;2. whether it has significant difference between the result of non-inferiority testing of ordered categorical data and binary data; If we choose to apprise the non-inferiority of ordered categorical data, which questions need to pay attention.Methodsthis paper took a research base on a non-inferiority clinical trial of chinese medicine, which is a multicenter, randomized, double-blind, double-simulation, positive-controlled, parallel group, non-inferiority trial.1. We adopt frequency statistics approach (hypothesis test and confidence interval) make non-inferiority test to the definite outcomes, and adopt Bayesian statistical approach to make non-inferiority test to response rate. then realize the advantages and disadvantages and discuss how to make a choice between above two approachs by some literatures of Bayesian statistical approach and by compare the results of non-inferiority testing.2. Refers to the basic theory of "the difference of row mean scores method " and "the ratio of row mean ranks method" put forward by Xin Weiquan in equivalence testing of ordered categorical data, We established the non-inferiority testing model of ordered categorical data, and make a non-inferiority test to the ordered categorical data outcome of a TCM clinical trial using these two methods, and then compared the results of non-inferiority test of ordered categorical data and binomial data and measurement data.(1) Firstly, To calculate the mean (μR) of effect ordered scores of control group using "the difference of row mean scores method", then obtain The lower limit of 95% confidence interval of the difference of row mean scores by the Bootstrap resampling 1000 times. Finally, refer to the non-inferiority margin (δ=0.20μR) to accomplish the non-inferiority statistical inference. (2) For "the ratio of row mean ranks method ", To obtain The lower limit of 95% confidence interval of the the ratio of row mean ranks by the bootstrap resampling 1000 times. And then, refer to the non-inferiority margin (δ=0.80) to accomplish the non-inferiority statistical inference.Result1. To the TCM syndrome Comprehensive Effectiveness (The effective rates), the normal rate and the improvement rate of the function of gastric motility in the full analysis set and per-protocol population set, the non-inferiority test results of frequency statistical approach suggest that experimental treatment is non-inferiority to active control treatment. but to the change value of the function of gastric motility, we cannot make a deduction that experimental treatment is non-inferiority to active control treatment.2. To the TCM syndrome Comprehensive Effectiveness (The effective rates), the non-inferiority test results of bayesian approach suggest that experimental treatment is non-inferiority to active control treatment. The result is consistent with the frequency statistical approach.3. To the ordered categorical data of TCM syndrome, the levels of gastric motility function, and the change degree of gastric motility function, the results of non-inferiority test is inconsistent using "the difference of row mean scores method" compared with "the ratio of row mean ranks method". "the difference of row mean scores method" obtained non-inferiority conclusions, but "the ratio of row mean ranks method" cannon make a deduction that experimental treatment is non-inferiority to active control treatment.4. Compared with the result of non-inferiority test to effective rate and the improvement rate of the function of gastric motility, "the difference of row mean scores method", which test to the ordered categorical data of TCM syndrome and gastric motility function, obtained consistent conclusions, but "the ratio of row mean ranks method" obtained inconsistent conclusions.Conclusion1. Frequency statistical approach and bayesian statistical approach can be used to make non-inferiority test in chinese medicine clinical trial, But, each of them has its own advantagesand disadvantages.(1) if only stems from the goal of non-inferiority appraisal, for some relatively simple trial and data, the frequency statisticcal approach is an overwhelming choice; But for some complex model and data, Bayesian statistical approach can be used as a supplementary to the frequency statistical approach.(2) If taking the design of non-inferiority trial into consideration, bayesian statistical approach may be primarily option. Moreover, Bayesian statistical approach need to be discussed in the design aspect of non-inferiority clinical trial.2. In non-inferiority clinical trials of TCM, in view of many kinds of manifestation of clinical syndrome outcome, which outcome shoud be choosen to make a non-inferiority test, besides consider from clinical specialized angle, we also need consider the degree distributions and sample size of two groups, the non-inferiority margin of ordered categorical data, and so on.(1) when the degree distributions are same, to the ordered categorical data of TCM syndrome and the effective rates, the results of non-inferiority test are solid using "the difference of row mean scores method". And this time, classical statistics approach is simple and convenient to make non-inferioritytesting.(2) When the degree distributions are different, we should consider to make non-inferiority testing to ordered categorical data directly. But when chooce "the difference of row mean scores method", we need to consider the non-inferiority margin is reasonable or not; when chooce "the ratio of row mean ranks method", we need to consider the sample size is enough or not.