Unit of Biostatistics, Epidemiology and Public Health, Department of Cardiac Thoracic Vascular Sciences and Public Health, University of Padova, 35128 Padova, Italy.
Department of Translational Medicine, University of Eastern Piedmont, 28100 Novara, Italy.
Int J Environ Res Public Health. 2021 Feb 21;18(4):2095. doi: 10.3390/ijerph18042095.
In the conduction of trials, a common situation is related to potential difficulties in recruiting the planned sample size as provided by the study design. A Bayesian analysis of such trials might provide a framework to combine prior evidence with current evidence, and it is an accepted approach by regulatory agencies. However, especially for small trials, the Bayesian inference may be severely conditioned by the prior choices. The Renal Scarring Urinary Infection (RESCUE) trial, a pediatric trial that was a candidate for early termination due to underrecruitment, served as a motivating example to investigate the effects of the prior choices on small trial inference. The trial outcomes were simulated by assuming 50 scenarios combining different sample sizes and true absolute risk reduction (ARR). The simulated data were analyzed via the Bayesian approach using 0%, 50%, and 100% discounting factors on the beta power prior. An informative inference (0% discounting) on small samples could generate data-insensitive results. Instead, the 50% discounting factor ensured that the probability of confirming the trial outcome was higher than 80%, but only for an ARR higher than 0.17. A suitable option to maintain data relevant to the trial inference is to define a discounting factor based on the prior parameters. Nevertheless, a sensitivity analysis of the prior choices is highly recommended.
在临床试验中，常见的情况是潜在的困难，与潜在的困难招募计划的样本量由研究设计。这样的试验贝叶斯分析可能提供了一个框架，结合先前的证据与当前的证据，它是一个被监管机构接受的方法。然而，特别是对于小的试验，贝叶斯推断可能会受到先验选择的严重限制。肾脏瘢痕尿路感染（RESCUE）试验，一个儿科试验，由于招募不足而成为早期终止的候选者，作为一个激励的例子，调查先验选择对小试验推断的影响。通过假设不同样本量和真实绝对风险降低（ARR）的 50 种组合来模拟试验结果。使用贝叶斯方法对模拟数据进行分析，使用贝塔幂先验的 0%、50%和 100%贴现因子。小样本的信息推断（0%贴现）可能会产生对数据不敏感的结果。相反，50%的贴现因子可以确保确认试验结果的概率高于 80%，但只有在 ARR 高于 0.17 时才会如此。保持与试验推断相关的数据的一个合适的选择是基于先验参数定义贴现因子。然而，强烈建议对先验选择进行敏感性分析。
