Research On Application Of Bayesian Statistics In Clinical Trials Of Medical Devices

Research background and purpose:With the increasing demand for innovative design in the field of medical device evaluation,Bayesian statistics is gradually being applied to medical device clinical trial evaluation because of its unique advantages.The popularity of Bayesian statistics in the evaluation of medical device clinical trials was further established after the U.S.Food and Drug Administration issued Guidelines for the Application of Bayesian Statistics in Medical Device Clinical Trials.However,the application of Bayesian statistics in medical device clinical trials in China is still in its infancy,and no medical device approval project using Bayesian statistics has been identified.Therefore,this paper will conduct an exploratory study on the application of Bayesian statistical methods in medical device clinical trials,aiming to introduce Bayesian statistical methods commonly used in the evaluation of medical device clinical trials into China through simulation studies and case studies,and lay the foundation and reference for the introduction of new ideas and methods in the evaluation of medical device clinical trials in China.Methods:(1)Based on simulation studies,a comparative study of several prior utilization methods in Bayesian statistics is conducted to compare the degree of influence of several methods on the utilization of a prior information and on the results of a posterior estimation,also to explore the ability of several methods to deal with heterogeneity and non-exchangeable situations.(2)Based on Simulation Research and case analysis,this paper makes a comparative study of three Bayesian methods and frequency methods in non-inferiority test design and discusses the influencing factors of Bayesian method posterior estimation and frequency operation characteristics.(3)Based on Simulation Research and case analysis,this paper makes a comparative study of Bayesian interim analysis and frequency interim analysis design,explores the differences and influencing factors on the estimation of results between the several approaches.The results of the study were compared between two methods based on Bayesian a posterior probability and predicted probability in a non-inferior scenario,comparing the differences and associations in the estimation of the results.A comparative study of three Bayesian filling methods based on predicted probability in the case of delayed outcomes,and the differences and influencing factors in result estimation between them are compared.Results:(1)The results of comparisons between the prior information utilization methods found that,under the same setting,the difference between the posterior estimates based on current data and the prior full-utilization method was greatest,and the other prior discount utilization method estimates falling between the two.When?Z2=0.0,0.01,the posterior estimates of power prior method,the hierarchical prior method and meta analysis prediction prior method are close.When ?Z2=0.10.0.25,the posterior estimate of hierarchical prior method is the largest,the posterior estimates of power prior method are greater than meta analysis prediction prior method.When N=50,the estimated values of various methods are quite different from the real effect of the test;when N=200,the estimated values of various methods are close to the real effect of the test.Under the condition of exchangeability,the posterior estimates of power priori method,hierarchical prior method and meta analysis prediction prior method are close.Under the condition of unexchangeability,the posterior estimate of hierarchical prior method is the largest,the posterior estimates of power prior method are larger than meta analysis prediction prior method.(2)The comparison results of non-inferiority evaluation methods show that,except for the posterior probability threshold,the Bayesian statistical method and the frequency method are approximated at the mean and confidence interval estimates of the difference in rates and the non?inferiority judgement conclusions under the same conditions.A prior has a great influence on the posterior estimation results of Bayesian statistical method.the posterior probability estimation value of Bayesian method is the largest at the optimistic prior and smallest at the pessimistic prior setting.Under the same settings,the power and type I error of Bayesian statistical method are greater than those of frequency method.The larger the posterior probability threshold,the lower the power and type I error of Bayesian statistical method.(3)The results of Bayesian and frequency interim analysis show that:the posterior result and confidence interval of Bayesian estimation were close to the results of the three methods of frequency estimation under the same setting,and the average sample size of Bayesian method was smaller than frequency methods.The analysis based on both the posterior and predictive probability methods found that the effective termination probability of the predictive probability-based method was greater than the posterior probability-based method for the same setting,the predictive probability-based method had a greater type I error rate and power than the posterior probability-based method,and the average sample size of the predictive probability-based method was slightly lower than the posterior probability-based method.The comparison of the results of the interim analysis in the case of delayed outcome found that,under the same conditions,the three effective termination probabilities based on predicted probability estimation were approximated,and the type I error rate,power and average sample size were also approximated.Conclusion:(1)The use of prior information has a great influence on Bayesian posterior results.When the heterogeneity between studies is small and the exchangeable hypothesis conditions are met,each method has a good effect on a prior information utilization.When the heterogeneity between studies is large or the exchangeable hypothesis conditions are not met,each method has a poor effect on a prior information utilization.(2)In non-inferiority trial designs,Bayesian statistical method and frequency method are similar in estimation and conclusion without prior information.The three Bayesian methods are interchangeable,Bayesian sampling method is more universal and worth popularizing.(3)The adaptive design of Bayesian interim analysis can be more practical value not only for effective early termination but also for ineffective early termination,and the average sample size is lower than frequency methods.The posterior estimated value and power of the predictive probability-based method are greater than the posterior probability-based.Both the two methods have their advantages and disadvantages,which should be used in conjunction with the actual application and purpose.In the case of delayed outcome,the methods based on prediction probability and auxiliary variable prediction can make up for the loss of information and improve the research efficiency.