Daniells Libby, Mozgunov Pavel, Barnett Helen, Bedding Alun, Jaki Thomas
STOR-i Centre for Doctoral Training, Lancaster University, Fylde College, Lancaster, Lancashire, LA1 4YF, United Kingdom.
MRC Biostatistics Unit, University of Cambridge, East Forvie Building, Robinson Way, Cambridge, Cambridgeshire, CB2 0SR, United Kingdom.
Biostatistics. 2024 Dec 31;26(1). doi: 10.1093/biostatistics/kxaf016.
In basket trials a single therapeutic treatment is tested on several patient populations simultaneously, each of which forming a basket, where patients across all baskets on the trial share a common genetic aberration. These trials allow testing of treatments on small groups of patients, however, limited basket sample sizes can result in inadequate precision and power of estimates. It is well known that Bayesian information borrowing models such as the exchangeability-nonexchangeability (EXNEX) model can be implemented to tackle such a problem, drawing on information from one basket when making inference in another. An alternative approach to improve power of estimates, is to incorporate any historical or external information available. This paper considers models that amalgamate both forms of information borrowing, allowing borrowing between baskets in the ongoing trial whilst also drawing on response data from historical sources, with the aim to further improve treatment effect estimates. We propose several Bayesian information borrowing approaches that incorporate historical information into the model. These methods are data-driven, updating the degree of borrowing based on the level of homogeneity between information sources. A thorough simulation study is presented to draw comparisons between the proposed approaches, whilst also comparing to the standard EXNEX model in which no historical information is utilized. The models are also applied to a real-life trial example to demonstrate their performance in practice. We show that the incorporation of historic data under the novel approaches can lead to a substantial improvement in precision and power of treatment effect estimates when such data is homogeneous to the responses in the ongoing trial. Under some approaches, this came alongside an inflation in type I error rate in cases of heterogeneity. However, the use of a power prior in the EXNEX model is shown to increase power and precision, whilst maintaining similar error rates to the standard EXNEX model.
在篮子试验中，单一治疗方法会同时在多个患者群体上进行测试，每个群体构成一个篮子，试验中所有篮子里的患者都有共同的基因畸变。这些试验允许对小群体患者进行治疗测试，然而，篮子样本量有限可能导致估计的精度和效能不足。众所周知，可以实施贝叶斯信息借用模型，如可交换性 - 不可交换性（EXNEX）模型来解决此类问题，即在对另一个篮子进行推断时借鉴一个篮子的信息。提高估计效能的另一种方法是纳入任何可用的历史或外部信息。本文考虑将两种信息借用形式结合的模型，既允许在正在进行的试验中的篮子之间借用信息，同时也借鉴历史来源的反应数据，目的是进一步改善治疗效果估计。我们提出了几种将历史信息纳入模型的贝叶斯信息借用方法。这些方法是数据驱动的，根据信息源之间的同质程度更新借用程度。本文进行了全面的模拟研究，以比较所提出的方法之间的差异，同时也与未使用历史信息的标准EXNEX模型进行比较。这些模型还应用于一个实际试验示例，以展示它们在实际中的表现。我们表明，当此类历史数据与正在进行的试验中的反应同质时，在新方法中纳入历史数据可导致治疗效果估计的精度和效能大幅提高。在某些方法下，在异质性情况下会伴随着I型错误率的上升。然而，在EXNEX模型中使用功效先验被证明可以提高效能和精度，同时保持与标准EXNEX模型相似的错误率。
