Comparison of Bayesian and Frequentist Methods in Cardiovascular Clinical Trials

描述（由申请人提供）：贝叶斯临床试验（BCT）据称在成本和人类受试者方面比现有的频率论方法更有效。目前，BCT已用于癌症试验、器械试验和一些HIV试验，但在其他健康研究领域是罕见的例外。虽然传统的频率论方法在实践中被广泛理解和接受，但贝叶斯方法尚未得到充分审查，并且由于它们依赖于繁重的计算，直到最近才在典型的临床试验环境中容易获得。本比较有效性研究旨在考虑贝叶斯方法在临床试验中的应用，并与使用真实数据的其他分析方法进行比较。本申请提出了一系列并排比较现有的频率组序贯方法，和新的贝叶斯方法的临床试验分析数据完成的心血管试验。这些试验因规模、结果、对其领域的影响以及原始分析日期而异。研究者将使用两种典型的频率论组序贯计划（Obrien-Fleming和Lan- DeMets α消耗函数）和具有信息性、非信息性和怀疑性先验分布的贝叶斯分析的几种变体重新分析每项试验的数据，以确定结局和试验持续时间对标准组序贯方法和BCT分析假设的敏感性。在每个分析时间点，将进行计算以确定存在临床相关效应的概率和无效概率。将使用具有模拟的随机分布的Bootstrap抽样来估计每次分析的实验误差率。此外，在每个分析时间点，将计算检测效应的条件把握度。这样就可以直接比较不同分析方法的相对有效性。最后，将综合所有试验的结果，以确定BCT方法或多或少有利的心血管临床试验的特征。本研究的结果将提供关键信息，以进一步开发临床试验的分析技术，并使用真实的生活数据的创新方法而不是模拟试验数据的标准使用来评估贝叶斯方法的相对效率。 公共卫生相关性：传统的频率论方法在临床试验分析中被普遍接受为研究实践中的金标准。然而，较新的贝叶斯方法对于更有效的分析方法有很大的希望，从而降低临床研究的成本，并通过减少暴露于劣质治疗组的时间长度来增强临床试验的伦理建设。本研究的结果将提供有价值的信息，这两种方法的真实比较性能，并帮助确定贝叶斯方法可能提供优于传统分析技术的情况。

DESCRIPTION (provided by applicant): Bayesian clinical trials (BCT) are purported to be more efficient in terms of cost and human subjects than existing frequentist methods. At this time, BCTs have been implemented for cancer trials, device trials, and some HIV trials, but are the rare exception in other fields of health research. While traditional frequentist methods are widely understood and accepted in practice, Bayesian methods are not yet fully vetted, and, due to their reliance on heavy computations, have not until recently, been readily accessible in the typical clinical trial setting. This comparative effectiveness study aims to consider applications of Bayesian methodology for the clinical trial in comparison to other analytic alternatives using real-life data. This application proposes a series of side-by-side comparisons between existing frequentist group sequential methods, and newer Bayesian methods of analysis of clinical trials using data from completed cardiovascular trials. The trials vary by size, outcome, impact on their field, and date of original analysis. Investigators will re- analyze the data from each trial using two typical frequentist group sequential plans (Obrien-Fleming and Lan- DeMets alpha spending functions) and several variations of Bayesian analysis with informative, non- informative, and skeptical prior distributions in order to determine the sensitivity of outcome and trial duration to analytic assumptions for both standard group sequential methods and BCTs. At each analysis time point, calculations will be made to determine the probability that there is a clinically relevant effect and probability of futility. Bootstrap sampling with a simulated Null distribution will be used to estimate experiment-wise error rates for each analysis. In addition, at each analysis point over time, conditional power to detect an effect will be calculated. In this way direct comparisons can be made as to the relative effectiveness of the different methods of analysis. Finally, results from all trials will be synthesized to identify the characteristics of a cardiovascular clinical trial for which BCT methods would be more or less advantageous. The results from this study will provide critical information to further the development of analytic techniques for clinical trials and to assess the relative efficiency of Bayesian methods using the innovative approach of real- life data rather than the standard use of simulated trial data. PUBLIC HEALTH RELEVANCE: Traditional frequentist methods for analysis of clinical trials are generally accepted as the gold standard in research practice. Newer Bayesian methods, however, hold great promise for a more efficient analytic approach, thereby lowering the costs of clinical research and enhancing the ethical construct of clinical trials by reducing the length of exposure time to inferior treatment arms. The results of this study will provide valuable information about the true comparative performance of the two methods and help identify the circumstances for which Bayesian approaches may provide advantages over conventional analytic techniques.