On Bayesian Response Adaptive Randomisation And Its Web Application Implementation

Background:Randomisation plays an essential role in clinical research.A thoughtful randomisation scheme can not only control bias and reduce confounders but also act as the theoretical foundation of effective hypothesis testing.The randomisation scheme which features fixed allocation probabilities,such as equal randomisation,rules the field because of its simplicity and efficiency.However,in light of the improvement of clinical trial methodologies,the ethics of equal randomisation has been questioned.The traditional concept is that clinical trials should emphasise on testing the differences between therapies as quickly as possible so that potential patients outside the trial can benefit,while the ethics of subjects in the trial are neglected.Nevertheless,is there any patient or clinician who does not want to try best to cure the disease?Furthermore,because of lacking information on interventions at the beginning of the trial,equal randomisation adheres to the principle of equality,but as the trial data accumulates,it is not wise to continue to ignore what they tell.The adaptive design allows reasonable adjustment of the trial protocol based on currently collected trial data and improves the efficiency and the ethics without destroying the authenticity and integrity of the trial.As one of them,response adaptive randomisation can assign the following subjects to the best treatment group with the highest allocation probability based on accumulated response data.At present,in-depth systematic research on response adaptive randomisation using frequentist methods has been conducted.On the other hand,the Bayesian methods fit well the process of adjusting the following allocation probabilities based on the latest knowledge due to Bayesian peculiarities of continually learning.Although there are many pieces of research on Bayesian response adaptive randomisation so far,problems such as the lack of a comprehensive scheme establishment framework,limited methods on the construction of allocation probabilities,and the undiscussed effects of design parameters still exist.Given that a Bayesian response adaptive randomisation scheme can only be evaluated through statistical simulation,nonetheless,only MD Anderson Cancer Center has released a computer program for simulation,which implements just one allocation probability construction method and does not support for more sophisticated schemes with block settings and time trend issues.Besides,unlike trials with fixed randomisation schemes,those using response-adaptive randomisation schemes do not support prespecified subject masking.A central randomisation system should be utilised to fulfil the dynamic subject unbalanced allocation procedures.This thesis will explore and try to solve the above methodological problems as well as practical difficulties.Method:①.A literature review was carried out to build a comprehensive scheme establishment framework.②.A statistical simulation script for evaluation of Bayesian response adaptive randomisation schemes was written in R and compared with the MD Anderson one for verification purpose.③.Statistical simulation approaches were applied to study the effects of each design parameter on evaluation indexes.④.A comparison of Bayesian response adaptive randomisation schemes,frequentist response adaptive randomisation schemes and an equal randomisation scheme was conducted based on an actual clinical trial with adjustments.⑤.JavaScript(Vue.js for the client and Express.js for the server)was used to develop an interactive web application which supports for subjects enrollment according to Bayesian response adaptive randomisation schemes.Result:①.A framework for the Bayesian response adaptive randomisation scheme establishment was raised,which includes the core model definition and the design parameters specification.The former employs the Bayesian analysis procedures as the basis,and then constructs allocation probabilities according to the posteriors and finally controls the extreme allocation probabilities.The latter uses statistical simulation techniques to obtain frequentist operating characteristics,ethics and bias indexes to adjust design parameters until satisfied evaluation indexes are attained.②.A new allocation probabilities construction method was proposed.③.An efficient statistical simulation program using R parallel programming techniques that can simultaneously call multiple cores was written.④.Indepth discussion of the influence of each design parameter on the evaluation indexes,which provides references and suggestions for practical scheme designs were provided.⑤.Three Bayesian response adaptive randomisation schemes,a parametric frequentist response adaptive randomisation scheme,a non-parametric frequentist response adaptive randomisation scheme and an equal randomisation scheme were compared based on the adjusted Fluoxetine Hydrochloride Trial.⑥.An interactive web application that supports subjects enrollment by means of Bayesian response adaptive randomisation schemes was developed.Conclusion:①.The Bayesian response adaptive randomisation scheme establishment framework proposed in this thesis can summarise existing works of literature and help to build schemes.②.Although in any adaptive randomisation scheme,pursuing high trial ethics will reduce power is inevitable,for the different needs of actual clinical trials,a satisfactory scheme can be designed through extensive simulation.③.The Bayesian response adaptive randomisation schemes constructed based on the three kinds of point estimates of each arm’s response rate posterior pay more attention to the power and are less ethical,which provide new choices for the actual trials.④.The statistical simulation script we developed and the MD Anderson Adaptive Randomization V5.2.2 can be mutually verified,and the former supports for more sophisticated settings.⑤.The interactive Web application provides core functions such as scheme design,subject number and drug number acquisition,response data update and storage.The client provides a user-friendly Chinese-English bilingual interface,with online help and responsive designs,which can be accessed on any terminal with a browser.The server implements the persistent storage of data and can call the R engine for statistical calculation.The thesis provides preliminary research on this domain and can be used as a reference for future professional central randomisation system development.