Di Maio Danilo, Mitchell S A, Batson S, Keeney E, Thom Howard H Z
F. Hoffmann-La Roche Ltd, Basel, Switzerland.
Mtech Access, Bicester, Oxfordshire, UK.
BMC Med Res Methodol. 2025 Feb 1;25(1):30. doi: 10.1186/s12874-025-02456-x.
The National Institute for Health and Care Excellence (England's health technology assessment body) recommend the use of the average treatment effect (ATE) as an estimand for economic evaluations. However there is limited literature on methods to estimate the ATE, particularly in the case of survival outcomes. Single-arm trials and real-world data are playing an increasing role in health technology assessments, particularly in oncology/rare diseases, generating a need for new ATE estimation methods. This study aimed to present the adaptation and utility of this methodology for survival outcomes.
The approach is based on a "doubly robust" method combining matching with regression adjustment (Austin 2020) using a Weibull model (lowest Akaike information criteria [AIC] specification) to estimate counterfactual event times. As a case study, we compared mosunetuzumab versus rituximab/bendamustine, as a proxy for rituximab/chemotherapy, in 3L+ relapsed/refractory follicular lymphoma. Individual patient data for mosunetuzumab (NCT02500407) and a combination of two rituximab/bendamustine 3L+ follicular lymphoma cohorts (NCT02187861/NCT02257567) were used. Endpoints included overall survival (OS) and progression-free survival (PFS). Sensitivity analyses were performed to test robustness to different distributional assumptions (log-normal, log-logistic and exponential) or model specifications (second, third and fourth lowest AIC) for event times.
The case study found improved PFS (hazard ratio [HR] 0.43 [95% confidence interval (CI): 0.13, 0.91]) and OS (HR 0.30 [95% CI: 0.05, 5.28]) for mosunetuzumab. Consistent findings (HR range 0.25-0.47 and 0.21-0.50 with all CIs excluding/including 1 for PFS/OS, respectively) were observed in sensitivity analyses.
DISCUSSION/CONCLUSIONS: The proposed adaptation expands the range of available approaches for the estimation of the (local) ATE for survival outcomes in health technology assessments using "doubly robust" methods. This approach appeared relatively robust to modelling decisions in our case study.
英国国家卫生与临床优化研究所（英格兰的卫生技术评估机构）建议将平均治疗效果（ATE）用作经济评估的估计量。然而，关于估计ATE的方法的文献有限，尤其是在生存结局方面。单臂试验和真实世界数据在卫生技术评估中发挥着越来越重要的作用，特别是在肿瘤学/罕见病领域，这就需要新的ATE估计方法。本研究旨在介绍该方法在生存结局方面的适应性和实用性。
该方法基于一种“双重稳健”方法，将匹配与回归调整相结合（奥斯汀，2020年），使用威布尔模型（最低赤池信息准则[AIC]规格）来估计反事实事件时间。作为一个案例研究，我们在3L+复发/难治性滤泡性淋巴瘤中比较了莫苏奈妥单抗与利妥昔单抗/苯达莫司汀（作为利妥昔单抗/化疗的替代）。使用了莫苏奈妥单抗的个体患者数据（NCT02500407）以及两个利妥昔单抗/苯达莫司汀3L+滤泡性淋巴瘤队列的组合数据（NCT02187861/NCT02257567）。终点包括总生存期（OS）和无进展生存期（PFS）。进行了敏感性分析，以测试对事件时间的不同分布假设（对数正态、对数逻辑和指数）或模型规格（第二、第三和第四最低AIC）的稳健性。
案例研究发现，莫苏奈妥单抗的PFS有所改善（风险比[HR]为0.43[95%置信区间（CI）：0.13，0.91]），OS也有所改善（HR为0.30[95%CI：0.05，5.28]）。在敏感性分析中观察到了一致的结果（HR范围为0.25 - 0.47和0.21 - 0.50，所有CI分别排除/包括1用于PFS/OS）。
讨论/结论：所提出的适应性扩展了使用“双重稳健”方法在卫生技术评估中估计生存结局的（局部）ATE的可用方法范围。在我们的案例研究中，这种方法对建模决策似乎相对稳健。
