Chen Wei-Chen, Lu Nelson, Wang Chenguang, Xu Yunling
Center for Devices and Radiological Health, Food and Drug Administration, 10903 New Hampshire Avenue, Silver Spring, MD, 20993, USA.
Biostatistics & Data Management, Regeneron Pharmaceuticals Inc, 777 Old Saw Mill River Rd, Tarrytown, NY, 10591, USA.
Ther Innov Regul Sci. 2025 Jul 22. doi: 10.1007/s43441-025-00839-2.
Non-randomized comparative studies are often used to compare treatment effects between an investigational product and a control when randomization is not feasible or difficult in practice. A typical situation is that the product is investigated in a single-arm study, and the control data are collected in an external data source. For such a situation, we propose an alternative approach to draw inference on the treatment effect difference. First, a potential outcome model (POM) for the outcome under control treatment is built based on the external control data source. Next, the POM is utilized to impute outcomes of subjects in the single-arm study as if they were treated with the control treatment. Then the inference on the treatment effect difference can be made by comparing imputed outcomes (for the control) and observed outcomes (for the investigational product). The main purpose of this paper is to provide a proof of concept regarding how to perform inference on the treatment effect between the investigational product and the control under this scenario. We illustrate our approach by assuming the endpoint to follow a normal distribution and the POM to be a linear regression model.
当随机化在实际中不可行或困难时，非随机对照研究常被用于比较试验产品与对照之间的治疗效果。一种典型的情况是，产品在单臂研究中进行研究，而对照数据则从外部数据源收集。针对这种情况，我们提出了一种推断治疗效果差异的替代方法。首先，基于外部对照数据源构建对照治疗下结局的潜在结果模型（POM）。接下来，利用POM对单臂研究中受试者的结局进行插补，就好像他们接受了对照治疗一样。然后，通过比较插补结局（对照的）和观察到的结局（试验产品的）来推断治疗效果差异。本文的主要目的是提供一个概念验证，说明在这种情况下如何对试验产品与对照之间的治疗效果进行推断。我们通过假设终点服从正态分布且POM为线性回归模型来说明我们的方法。
