Tremblay Gabriel, Haines Patrick, Briggs Andrew
Eisai Co. Ltd., Woodcliff Lake, NJ, USA.
Curo Consulting, Marlow, Buckinghamshire, UK.
J Health Econ Outcomes Res. 2015 Feb 14;2(2):147-160. doi: 10.36469/9896. eCollection 2015.
Trial data often does not cover a sufficiently long period of time to truly capture time-toevent endpoints, however, Health Technology Assessment (HTA) bodies often require overall survival (OS) and progression-free survival (PFS) estimates. Often, significant survival effects are found beyond the time period observed in clinical trials, thus, extrapolation of trial results is required for health economic and HTA evaluations. This paper looks at different techniques that can be used to extrapolate trial data, as well as criteria that should be used to select the most appropriate technique. Using these insights a formal decisionmaking criteria will be established, allowing users to follow a systematic approach to extrapolating survival estimates. The techniques are then applied to a metastatic breast cancer (MBC) example. A criterion-based guide was devised to allow the accurate extrapolation and justification of survival estimates in a MBC study comparing eribulin (Halaven) monotherapy with treatment of their (patient's) physician's choice (TPC). Parametric and piecewise models are used to extrapolate survival estimates, and statistical as well as visual tests are used to decide the most appropriate modelling technique. In the case study presented, the optimal model was identified as the Accelerated Failure Time (AFT) Parametric model using a Gamma distribution with a treatment covariate for OS, and the Kaplan-Meier survival estimates for PFS. Survival estimates must be extrapolated to a time point such that the benefits of a therapy can be clearly demonstrated. A systematic approach combined with a formal decision-making structure should be used to minimize the potential for bias as well as making the process transparent.
试验数据通常没有涵盖足够长的时间来真正获取事件发生时间终点，然而，卫生技术评估（HTA）机构通常需要总生存期（OS）和无进展生存期（PFS）估计值。通常，在临床试验观察期之外会发现显著的生存效应，因此，卫生经济和HTA评估需要对试验结果进行外推。本文探讨了可用于外推试验数据的不同技术，以及应使用的选择最合适技术的标准。利用这些见解将建立一个正式的决策标准，允许用户采用系统方法来外推生存估计值。然后将这些技术应用于转移性乳腺癌（MBC）的例子。设计了一个基于标准的指南，以便在一项比较艾日布林（海乐卫）单药治疗与其（患者）医生选择的治疗（TPC）的MBC研究中，准确外推和证明生存估计值。使用参数模型和分段模型来外推生存估计值，并使用统计和视觉检验来确定最合适的建模技术。在所呈现的案例研究中，最佳模型被确定为加速失效时间（AFT）参数模型，对于OS使用具有治疗协变量的伽马分布，对于PFS使用Kaplan-Meier生存估计值。生存估计值必须外推到一个时间点，以便能够清楚地证明一种疗法的益处。应采用系统方法与正式决策结构相结合，以尽量减少偏差的可能性，并使过程透明。
