Fairley Michael, Cipriano Lauren E, Goldhaber-Fiebert Jeremy D
Department of Management Science and Engineering, Stanford University, Stanford, CA, USA.
Ivey Business School and the Department of Epidemiology and Biostatistics at Schulich School of Medicine and Dentistry, Western University, London, ON, Canada.
Med Decis Making. 2020 Aug;40(6):797-814. doi: 10.1177/0272989X20944875.
Health economic evaluations that include the expected value of sample information support implementation decisions as well as decisions about further research. However, just as decision makers must consider portfolios of implementation spending, they must also identify the optimal portfolio of research investments. Under a fixed research budget, a decision maker determines which studies to fund; additional budget allocated to one study to increase the study sample size implies less budget available to collect information to reduce decision uncertainty in other implementation decisions. We employ a budget-constrained portfolio optimization framework in which the decisions are whether to invest in a study and at what sample size. The objective is to maximize the sum of the studies' population expected net benefit of sampling (ENBS). We show how to determine the optimal research portfolio and study-specific levels of investment. We demonstrate our framework with a stylized example to illustrate solution features and a real-world application using 6 published cost-effectiveness analyses. Among the studies selected for nonzero investment, the optimal sample size occurs at the point at which the marginal population ENBS divided by the marginal cost of additional sampling is the same for all studies. Compared with standard ENBS optimization without a research budget constraint, optimal budget-constrained sample sizes are typically smaller but allow more studies to be funded. The budget constraint for research studies directly implies that the optimal sample size for additional research is not the point at which the ENBS is maximized for individual studies. A portfolio optimization approach can yield higher total ENBS. Ultimately, there is a maximum willingness to pay for incremental information that determines optimal sample sizes.
包含样本信息期望值的卫生经济评估为实施决策以及进一步研究的决策提供支持。然而，正如决策者必须考虑实施支出的组合一样，他们还必须确定研究投资的最优组合。在固定的研究预算下，决策者决定资助哪些研究；分配给一项研究以增加研究样本量的额外预算意味着可用于收集信息以减少其他实施决策中决策不确定性的预算减少。我们采用一种预算受限的投资组合优化框架，其中的决策是是否投资一项研究以及确定样本量大小。目标是使各项研究的总体抽样预期净效益（ENBS）之和最大化。我们展示了如何确定最优研究投资组合以及特定研究的投资水平。我们用一个程式化的例子来说明我们的框架，以展示其求解特征，并通过对6篇已发表的成本效益分析进行实际应用演示。在被选中进行非零投资的研究中，最优样本量出现在所有研究的边际总体ENBS除以额外抽样的边际成本都相同的点上。与没有研究预算限制的标准ENBS优化相比，最优预算受限样本量通常较小，但能资助更多的研究。研究预算的限制直接意味着额外研究的最优样本量不是单个研究的ENBS最大化的点。投资组合优化方法可以产生更高的总ENBS。最终，存在一个对增量信息的最大支付意愿，它决定了最优样本量。
