From the Centre for Clinical Epidemiology, Lady Davis Institute, Jewish General Hospital, and Departments of Epidemiology, Biostatistics, and Medicine, McGill University, Montreal, Canada.
Epidemiology. 2021 Jan;32(1):94-100. doi: 10.1097/EDE.0000000000001267.
Regulatory agencies now recognize single-arm trials with external historical controls, particularly common in oncology, to assess promising treatments for rare or specific indications. When a new treatment indication depends on events over time, such as treatment failures, this design can introduce time-related biases in comparisons with external controls.
We describe two potential biases resulting from calendar time and choice of time zero. We illustrate these biases using simulated data, emulating those from a single-arm trial of the effectiveness of blinatumomab in treating relapsed or refractory acute lymphoblastic leukemia on the outcome of mortality.
The trial compared 189 patients treated with blinatumomab with 1112 external historical control patients. First, calendar time was not concurrent, with the blinatumomab arm diagnosed during 2010-2014 and the control cohort during 1990-2013. The median survival under blinatumomab was 6.1 months compared with 3.3 months in the control arm, though for the latter it increased from 2.4 to 4.2 months over the 24-year period. Second, using the latest line of salvage treatment as cohort, entry for the control cohort introduces selection bias. The corresponding hazard ratio of death with blinatumomab compared with control was 0.56 (95% CI = 0.47, 0.67) but became 0.98 (95% CI = 0.83, 1.15) after redefining cohort entry by the matched line of salvage treatment rather than the latest line.
While single-arm trials with external historical controls are gaining recognition, a proper understanding of time-related sources of bias is essential if such trials will be used to provide valid evidence for drug approval from regulatory agencies.
监管机构现在认可采用外部历史对照的单臂试验，特别是在肿瘤学中，以评估针对罕见或特定适应症的有前途的治疗方法。当新的治疗适应症取决于随时间推移的事件时，例如治疗失败，这种设计可能会在与外部对照的比较中引入与时间相关的偏倚。
我们描述了两种潜在的偏倚，分别源于日历时间和时间零的选择。我们使用模拟数据来说明这些偏倚，这些数据模拟了治疗复发或难治性急性淋巴细胞白血病的blinatumomab 的单臂试验，结果是死亡率。
该试验比较了 189 名接受blinatumomab治疗的患者和 1112 名外部历史对照患者。首先，日历时间并不一致，blinatumomab 组在 2010-2014 年诊断，对照组在 1990-2013 年诊断。在blinatumomab 组的中位生存期为 6.1 个月，而对照组为 3.3 个月，尽管后者在 24 年期间从 2.4 个月增加到 4.2 个月。其次，使用最新的挽救治疗线作为队列，对照组的入组会引入选择偏倚。与对照组相比，blinatumomab 的死亡风险比为 0.56（95%CI=0.47,0.67），但在重新定义队列入组为匹配的挽救治疗线而不是最新的治疗线后，该比值变为 0.98（95%CI=0.83,1.15）。
虽然采用外部历史对照的单臂试验越来越受到认可，但如果此类试验将被用于为监管机构提供药物批准的有效证据，那么正确理解与时间相关的偏倚来源至关重要。
