Bayesian multivariate evidence synthesis methods to incorporate surrogate endpoints in health care evaluation

在卫生技术评估（HTA）中，决策是由NICE等机构就新卫生技术在卫生保健系统（如英国的NHS）中的可用性做出的。最近的趋势是缩短临床试验的持续时间和加快决策速度。在评估新的卫生技术（如新的药物、设备或外科手术）时，有效性结果是在随机对照试验（RCTs）中衡量的。然而，测量最终临床结果可能需要延长随访时间。因此，越来越多地寻求短期终点作为长期终点的替代品，以加快药物开发过程。候选替代终点需要验证，以确保由这些终点测量的治疗效果能够很好地预测由真实结果测量的治疗效果。例如，在癌症中，通过比较新治疗组患者的总生存期（OS）与对照组患者的生存期来衡量治疗效果，可以通过测量无进展生存期（PFS），即无复发或无进展的生存期来预测。与OS相比，PFS是早期测量的，因此PFS是临床试验中理想的替代终点。替代结果可用于预测临床疗效或映射到与健康相关的生活质量测量，两者都用于经济评估。因此，使用代理端点可以加快决策过程。拟议的项目旨在评估当前的方法，并进一步开发合适的方法来评估替代终点并将其纳入HTA。已经提出了一些旨在评估替代结果的元分析方法。不同的方法基于不同的建模假设，并考虑到这些模型参数的不同程度的不确定性。这可能会影响验证和预测的准确性。多变量荟萃分析方法最适合评估替代结果，因为它们考虑了结果之间的关系。贝叶斯方法的优点是允许分析人员对预测做出直接的概率陈述，例如，当用真实结果衡量新治疗优于标准治疗的概率时，假设用替代结果衡量新治疗优于标准治疗。贝叶斯方法在包含外部信息的能力上也是独一无二的，基于外部临床数据或专家意见，以所谓的先验分布的形式在模型中。这种方法使我们能够考虑到HTA中所有可用的证据，目标是使概率陈述更准确，从而实现更有效的决策。将开发新的证据合成方法，以纳入HTA的替代终点。该项目将扩展申请人最近在多元荟萃分析方面的工作，通过调查可选择的建模假设（关于模型参数之间的关系以及代理和临床结果之间的关系），以确保它们适用于不同的医疗保健环境。这项工作将扩展到多个替代结果和间接比较（IC）和网络荟萃分析（NMA）方法。当比较两种治疗的试验数据有限时，IC和MTC允许我们通过使用使用不同治疗比较物的研究的间接证据来估计这些治疗之间的相对效果。评估更广泛干预措施的研究可能报告更广泛的结果。因此，有必要发展多元集成电路和NMA方法。还将开发方法，通过使用多变量方法将特定疾病的有效性和与健康相关的生活质量措施映射到经济评估中使用的标准生活质量终点（通常是EQ-5D），将替代结果纳入卫生技术的经济评估。

In health technology assessment (HTA) decisions are made, by agencies such as NICE, about the availability of new health technologies in health care systems such as the NHS in the UK. There has been a recent trend towards clinical trials of shorter duration and towards faster decision making. When evaluating new health technologies, such as new medications, devices or surgical procedures, effectiveness outcomes are measured in randomised controlled trials (RCTs). However, measuring a final clinical outcome may require extended follow-up time. Hence shorter term endpoints are increasingly sought as surrogates for long term endpoints to expedite the drug development process. Candidate surrogate endpoints need to be validated to ensure that the treatment effect measured by those endpoints predicts well the treatment effect that would be measured by the true outcome. For example, in cancer the effect of treatment measured by comparing overall survival (OS) of patients in the new treatment group with survival of those in the control group can be predicted by measuring progression free survival (PFS), the survival without relapse or progression. PFS is measured early compared to OS, hence PFS is a desirable surrogate endpoint in clinical trials. Surrogate outcomes can be used to predict clinical effectiveness or be mapped onto health-related quality of life measures, both used in economic evaluation. Hence the use of surrogate endpoints can lead to faster decision-making process.The proposed project aims to assess current methodology and develop further suitable methods to evaluate surrogate endpoints and incorporate them in HTA. A number of meta-analytical methods have been proposed that aim to evaluate surrogate outcomes. Different approaches are based on different modelling assumptions and take into account different levels of uncertainty about parameters of such models. This may impact on accuracy of the validation and predictions. Multivariate meta-analysis methods are most suitable to evaluate surrogate outcomes as they take into account the relationship between outcomes. Bayesian methods have the advantage of allowing analysts to make direct probability statements about predictions, for example about the probability that the new treatment is superior to the standard care when measured by the true outcome given that it is superior when measured by the surrogate outcome. Bayesian methods are also unique in their ability to include external information, based on external clinical data or expert opinions in a model in the form of so called prior distributions. This approach allows us to take into account all available evidence in HTA with the goal of making the probability statements more accurate, leading to more efficient decision making.Novel evidence synthesis methods will be developed to incorporate surrogate endpoints in HTA. The project will extend the applicant's recent work on multivariate meta-analysis by investigating alternative modelling assumptions (about the relationship between parameters of the model and between the surrogate and clinical outcomes) to ensure they are suitable in different health care settings. The work will be extended to multiple surrogate outcomes and indirect comparisons (IC) and network meta-analysis (NMA) methods. When data from trials comparing two treatments is limited, IC and MTC allow us to estimate the relative effect between those treatments by use of indirect evidence from studies using different treatment comparators. Studies evaluating a wider range of interventions may report a wider range of outcomes. Hence there is a need to develop multivariate IC and NMA methods. Methods will also be developed to incorporate surrogate outcomes in economic evaluation of heath technologies, by using multivariate methods to map disease specific effectiveness and health related quality of life measures onto standard quality of life endpoints (usually EQ-5D) used in economic evaluation.