THE SEARCH FOR INFORMATION - A PATIENT PERSPECTIVE ON MULTIPLE OPINIONS

作     者：MEHREZ, A YUAN, Y GAFNI, A 

作者机构：MCMASTER UNIVCTR HLTH ECON & POLICY ANALDEPT CLIN EPIDEMIOL & BIOSTATHAMILTONON L8N 3Z5CANADA MCMASTER UNIVMICHAEL G DEGROOTE SCH BUSINESSHAMILTONON L8S 4M4CANADA KENT STATE UNIVGRAD SCH MANAGEMENTKENTOH 44242 MCMASTER UNIVCTR HLTH ECON & POLICY ANALHAMILTONON L8N 3Z5CANADA 

出 版 物：《EUROPEAN JOURNAL OF OPERATIONAL RESEARCH》 (Eur J Oper Res)

年 卷 期：1995年第85卷第2期

页      面：244-262页

核心收录：

学科分类：1201[管理学-管理科学与工程(可授管理学、工学学位)] 07[理学] 070104[理学-应用数学] 0701[理学-数学] 

主　　题：STOCHASTIC DYNAMIC PROGRAMMING EXPECTED VALUE OF SAMPLE INFORMATION (EVSI) 2ND OPINION 

摘      要：This paper describes the first dynamic formulation of the optimal strategy, from a patient perspective, for the search for additional information in the form of multiple opinions. The Mehrez and Gafni framework for choosing an optimal treatment from a patient perspective is employed to study the optimal search strategy for a representative individual who is an expected utility maximizer. Two different processes for gathering information regarding the probability of treatment success (P) are studied: In the first, physicians provide information in a yes or no (treatment) type of recommendation;in the second, physicians provide probabilistic information about the likelihood of success in the treatment. For the first system the optimal strategy is derived using an algorithm of an order O(n) (n is the number of opinions to be purchased from expert doctors) for a prior beta probability density function of P and a Bernoulli process of information gathering. EVSI (expected value of sample information) measures are incorporated into the Bayesian analysis to study this case and to obtain an upper bound on n* (the optimal number of opinions to be purchased). The approach developed here is different than the well known one provided by Raiffa and Schlaifer in 1961 and it enables us to present additional results regarding the optimal search strategy and how to identify it. For example, if (i) the process of information gathering is generated by a Bernoulli process, (ii) the natural conjurate prior distribution is beta, and (iii) the utility functions are linear functions of the probability of treatment success, then for each n, there exists at most one value of x (the number of experts recommending treatment) for which a search should continue. At this value the incremental increase of the individual s lifetime expected utility stemming from the decision to seek additional expert s opinion is positive. This observation might explain a reality where the likelihood of seeking a la