MRC Biostatistics Unit, Cambridge, UK (CHJ, SGT, LDS)
Centre for Health Economics, University of York, UK (LB, KC)
Med Decis Making. 2011 Jul-Aug;31(4):662-74. doi: 10.1177/0272989X11406986. Epub 2011 May 20.
Decision analytic models used for health technology assessment are subject to uncertainties. These uncertainties can be quantified probabilistically, by placing distributions on model parameters and simulating from these to generate estimates of cost-effectiveness. However, many uncertain model choices, often termed structural assumptions, are usually only explored informally by presenting estimates of cost-effectiveness under alternative scenarios. The authors show how 2 recent research proposals represent parts of a framework to formally account for all common structural uncertainties. First, the model is expanded to include parameters that encompass all possible structural choices. Uncertainty can then arise because these parameters are estimated imprecisely from data, for example, a treatment effect of doubtful significance. Uncertainty can also arise if there are no relevant data. If there are relevant data, uncertainty can be addressed by averaging expected costs and effects generated from probabilistic analysis of the models with and without the parameter. The weights used for averaging are related to the predictive ability of each model, assessed against the data. If there are no data, additional parameters can often be informed by eliciting expert beliefs as probability distributions. These ideas are illustrated in decision models for antiplatelet therapies for vascular disease and new biologic drugs for the treatment of active psoriatic arthritis.
用于健康技术评估的决策分析模型会存在不确定性。这些不确定性可以通过对模型参数进行概率分布，并从这些分布中进行模拟，从而生成成本效益估计值来进行量化。然而，许多不确定的模型选择，通常被称为结构假设，通常只是通过呈现替代方案下的成本效益估计值来进行非正式的探讨。作者展示了最近的两项研究提案如何代表了一个正式考虑所有常见结构不确定性的框架的一部分。首先，模型被扩展到包括涵盖所有可能结构选择的参数。不确定性可能会因为这些参数是从数据中不准确地估计出来的，例如，治疗效果存在疑问。如果没有相关数据，也会产生不确定性。如果有相关数据，则可以通过对有和没有该参数的模型进行概率分析来生成成本和效果的预期平均值来解决不确定性。用于平均值的权重与针对数据评估的每个模型的预测能力有关。如果没有数据，则可以通过启发式地将专家的信念作为概率分布来为其他参数提供信息。这些想法在血管疾病的抗血小板治疗和治疗活动性银屑病关节炎的新型生物药物的决策模型中得到了说明。
