Two-level resampling as a novel method for the calculation of the expected value of sample information in economic trials

Mohsen Sadatsafavi, Carlo Marra, Stirling Bryan

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

It has already been pointed out that the bootstrap can be used to calculate the expected value of perfect information (EVPI) when individual-level data from a randomized controlled trial (RCT) is at hand. However, as mentioned by others, it is not clear if and how such a method can be extended to calculate the expected value of sample information (EVSI). In this article, we provide a nonparametric definition for EVPI and EVSI, which is based on considering the entire population distribution as the uncertain entity for which the current RCT provides partial information. This enables a theoretical justification for using the bootstrap for EVPI calculation and allows us to propose a two-level resampling method for EVSI calculation. What is considered as the sampling unit in this algorithm can range from the individual level net benefits to the full panel of the RCT data for an individual, enabling the analyst to decide on a trade-off between computational efficiency and comprehensiveness in value of information analysis. As such, we argue that this method, in addition to being consistent with the popular bootstrap method of RCT-based economic evaluations, is a flexible approach for EVPI and EVSI calculations.
Original languageEnglish
Pages (from-to)877-882
Number of pages6
JournalHealth Economics
Volume22
Issue number7
Early online date30 Aug 2012
DOIs
Publication statusPublished - Jul 2013

Fingerprint

Randomized Controlled Trials
Economics
Cost-Benefit Analysis
Hand
Demography

Keywords

  • Bayesian statistics
  • value of information
  • bootstrap
  • clinical trials

Cite this

Two-level resampling as a novel method for the calculation of the expected value of sample information in economic trials. / Sadatsafavi, Mohsen; Marra, Carlo; Bryan, Stirling.

In: Health Economics, Vol. 22, No. 7, 07.2013, p. 877-882.

Research output: Contribution to journalArticle

@article{7027a8aded1341f9b191aaaa829d12bb,
title = "Two-level resampling as a novel method for the calculation of the expected value of sample information in economic trials",
abstract = "It has already been pointed out that the bootstrap can be used to calculate the expected value of perfect information (EVPI) when individual-level data from a randomized controlled trial (RCT) is at hand. However, as mentioned by others, it is not clear if and how such a method can be extended to calculate the expected value of sample information (EVSI). In this article, we provide a nonparametric definition for EVPI and EVSI, which is based on considering the entire population distribution as the uncertain entity for which the current RCT provides partial information. This enables a theoretical justification for using the bootstrap for EVPI calculation and allows us to propose a two-level resampling method for EVSI calculation. What is considered as the sampling unit in this algorithm can range from the individual level net benefits to the full panel of the RCT data for an individual, enabling the analyst to decide on a trade-off between computational efficiency and comprehensiveness in value of information analysis. As such, we argue that this method, in addition to being consistent with the popular bootstrap method of RCT-based economic evaluations, is a flexible approach for EVPI and EVSI calculations.",
keywords = "Bayesian statistics, value of information, bootstrap, clinical trials",
author = "Mohsen Sadatsafavi and Carlo Marra and Stirling Bryan",
note = "Copyright 2012 Elsevier B.V., All rights reserved.",
year = "2013",
month = "7",
doi = "10.1002/hec.2869",
language = "English",
volume = "22",
pages = "877--882",
journal = "Health Economics",
issn = "1057-9230",
publisher = "John Wiley and Sons Ltd",
number = "7",

}

TY - JOUR

T1 - Two-level resampling as a novel method for the calculation of the expected value of sample information in economic trials

AU - Sadatsafavi, Mohsen

AU - Marra, Carlo

AU - Bryan, Stirling

N1 - Copyright 2012 Elsevier B.V., All rights reserved.

PY - 2013/7

Y1 - 2013/7

N2 - It has already been pointed out that the bootstrap can be used to calculate the expected value of perfect information (EVPI) when individual-level data from a randomized controlled trial (RCT) is at hand. However, as mentioned by others, it is not clear if and how such a method can be extended to calculate the expected value of sample information (EVSI). In this article, we provide a nonparametric definition for EVPI and EVSI, which is based on considering the entire population distribution as the uncertain entity for which the current RCT provides partial information. This enables a theoretical justification for using the bootstrap for EVPI calculation and allows us to propose a two-level resampling method for EVSI calculation. What is considered as the sampling unit in this algorithm can range from the individual level net benefits to the full panel of the RCT data for an individual, enabling the analyst to decide on a trade-off between computational efficiency and comprehensiveness in value of information analysis. As such, we argue that this method, in addition to being consistent with the popular bootstrap method of RCT-based economic evaluations, is a flexible approach for EVPI and EVSI calculations.

AB - It has already been pointed out that the bootstrap can be used to calculate the expected value of perfect information (EVPI) when individual-level data from a randomized controlled trial (RCT) is at hand. However, as mentioned by others, it is not clear if and how such a method can be extended to calculate the expected value of sample information (EVSI). In this article, we provide a nonparametric definition for EVPI and EVSI, which is based on considering the entire population distribution as the uncertain entity for which the current RCT provides partial information. This enables a theoretical justification for using the bootstrap for EVPI calculation and allows us to propose a two-level resampling method for EVSI calculation. What is considered as the sampling unit in this algorithm can range from the individual level net benefits to the full panel of the RCT data for an individual, enabling the analyst to decide on a trade-off between computational efficiency and comprehensiveness in value of information analysis. As such, we argue that this method, in addition to being consistent with the popular bootstrap method of RCT-based economic evaluations, is a flexible approach for EVPI and EVSI calculations.

KW - Bayesian statistics

KW - value of information

KW - bootstrap

KW - clinical trials

UR - http://www.scopus.com/inward/record.url?scp=84865542212&partnerID=8YFLogxK

U2 - 10.1002/hec.2869

DO - 10.1002/hec.2869

M3 - Article

VL - 22

SP - 877

EP - 882

JO - Health Economics

JF - Health Economics

SN - 1057-9230

IS - 7

ER -