GROUP SEQUENTIAL CLINICAL-TRIALS - A CLASSICAL EVALUATION OF BAYESIAN DECISION-THEORETIC DESIGNS
GROUP SEQUENTIAL CLINICAL-TRIALS - A CLASSICAL EVALUATION OF BAYESIAN DECISION-THEORETIC DESIGNS
复制标题
DOI：
10.2307/2291016
复制
发表时间：
1994-12-01
影响因子：
3.7
通讯作者：
BERRY, DA
中科院分区：
文献类型：
--
作者：
LEWIS, RJ;BERRY, DA
贝叶斯决策理论设计的临床试验，比较两种治疗方法，具有双重结果的发展和评估。治疗i的成功结果概率由p(I)表示，i=1，2，并且假设关于每个p(I)的先验知识服从贝塔分布。假定p(I)是独立的。为了便于与常见的临床试验设计进行比较，我们采用了假设检验的方法。零假设是Delta>；Delta(0)，替代假设是Delta>；0，其中Delta(0)是试验寻求的最小治疗效果，Delta=p(2)，-p(1)是真正的治疗差异。我们使用一个简单的终端损失函数来反映试验的假设检验目标，试验的总成本是最终样本量加上终端损失函数。使总费用期望值最小的停止规则和决策规则由反向归纳法确定。用蒙特卡罗模拟比较了这些贝叶斯设计与Pocock和O‘Brien-Fleming的单尾经典分组序贯设计的贝叶斯错误率、频率错误率和平均样本量。正如预期的那样，贝叶斯决策理论设计比经典设计具有更小的平均成本。更令人惊讶的是，当选择终端损耗函数的大小来产生与经典设计相似的频率错误率时，贝叶斯设计的平均样本量通常较小。
Bayesian decision-theoretic designs for a clinical trial comparing two treatments for a disease with binary outcomes are developed and evaluated. The probability of successful outcome with treatment i is denoted by p(i), i = 1, 2, and prior knowledge regarding each p(i) is assumed to follow a beta distribution. The p(i) are assumed to be independent. To facilitate comparison with frequentist clinical trial designs, we take a hypothesis-testing approach. The null hypothesis is delta > delta(0), and the alternative hypothesis is delta > 0, where delta(0) is the minimum treatment effect sought by the trial and delta = p(2), - p(1) is the true treatment difference. We use a simple terminal loss function reflecting the hypothesis-testing goal of the trial, and the total cost of the trial is the final sample size plus the terminal loss function. The stopping and decision rules that minimize the expectation of the total cost are determined by backward induction. Monte Carlo simulation is used to compare Bayesian and frequentist error rates and mean sample sizes of these Bayesian designs with one-tailed classical group-sequential designs of Pocock and O'Brien-Fleming. As expected, the Bayesian decision-theoretic designs have smaller mean costs than the classical designs. More surprising, when the magnitude of the terminal loss function is chosen to yield frequentist error rates similar to those for classical designs, the mean sample sizes of the Bayesian designs are usually smaller.