Coyle Doug, Oakley Jeremy
Department of Epidemiology and Community Medicine, University of Ottawa, 451 Smyth Road, Ottawa, ON, Canada, K1H 8M5.
Eur J Health Econ. 2008 Aug;9(3):251-9. doi: 10.1007/s10198-007-0069-y. Epub 2007 Jul 19.
Value of information analysis provides a framework for the analysis of uncertainty within economic analysis by focussing on the value of obtaining further information to reduce uncertainty. The mathematical definition of the expected value of perfect information (EVPI) is fixed, though there are different methods in the literature for its estimation. In this paper these methods are explored and compared.
Analysis was conducted using a disease model for Parkinson's disease. Five methods for estimating partial EVPIs (EVPPIs) were used: a single Monte Carlo simulation (MCS) method, the unit normal loss integral (UNLI) method, a two-stage method using MCS, a two-stage method using MCS and quadrature and a difference method requiring two MCS. EVPPI was estimated for each individual parameter in the model as well as for three groups of parameters (transition probabilities, costs and utilities).
Using 5,000 replications, four methods returned similar results for EVPPIs. With 5 million replications, results were near identical. However, the difference method repeatedly gave estimates substantially different to the other methods.
The difference method is not rooted in the mathematical definition of EVPI and is clearly an inappropriate method for estimating EVPPI. The single MCS and UNLI methods were the least complex methods to use, but are restricted in their appropriateness. The two-stage MCS and quadrature-based methods are complex and time consuming. Thus, where appropriate, EVPPI should be estimated using either the single MCS or UNLI method. However, where neither of these methods is appropriate, either of the two-stage MCS and quadrature methods should be used.
信息价值分析通过关注获取更多信息以减少不确定性的价值，为经济分析中的不确定性分析提供了一个框架。完美信息期望值（EVPI）的数学定义是固定的，尽管文献中有不同的估计方法。本文对这些方法进行了探索和比较。
使用帕金森病疾病模型进行分析。采用了五种估计部分EVPI（EVPPIs）的方法：单一蒙特卡罗模拟（MCS）方法、单位正态损失积分（UNLI）方法、使用MCS的两阶段方法、使用MCS和求积法的两阶段方法以及需要两次MCS的差分法。对模型中的每个个体参数以及三组参数（转移概率、成本和效用）估计了EVPI。
使用5000次重复时，四种方法得出的EVPPIs结果相似。使用500万次重复时，结果几乎相同。然而，差分法反复给出与其他方法有显著差异的估计值。
差分法并非基于EVPI的数学定义，显然是一种不适合估计EVPPI的方法。单一MCS和UNLI方法是使用起来最不复杂的方法，但适用性有限。基于两阶段MCS和求积法的方法复杂且耗时。因此，在合适的情况下，应使用单一MCS或UNLI方法估计EVPPI。然而，当这两种方法都不适用时，应使用两阶段MCS和求积法中的任何一种。
