Premium
Research prioritization based on expected value of partial perfect information: a case‐study on interventions to increase uptake of breast cancer screening
Author(s) -
Welton N. J.,
Ades A. E.,
Caldwell D. M.,
Peters T. J.
Publication year - 2008
Publication title -
journal of the royal statistical society: series a (statistics in society)
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.103
H-Index - 84
eISSN - 1467-985X
pISSN - 0964-1998
DOI - 10.1111/j.1467-985x.2008.00558.x
Subject(s) - bayesian probability , prioritization , computer science , value of information , psychological intervention , risk analysis (engineering) , data mining , econometrics , medicine , management science , mathematics , artificial intelligence , psychiatry , economics
Summary. We investigate whether Bayesian decision theory, in the form of expected value of partial perfect information (EVPPI) analysis, is a realistic and practical approach to research prioritization. We develop a simple cost‐effectiveness analysis of breast cancer screening as a typical case‐study, motivated by data from a cluster randomized 2 × 2 factorial trial of interventions to increase uptake. An EVPPI analysis is developed which shows that, on the basis of the evidence that was available beforehand, the trial was cost effective, but that after incorporating the results of the trial it would still be cost effective to carry out research that further reduced decision uncertainty. We identify key conceptual and technical issues: the relationship between the target interventions and the previous evidence, the distinction between variation and uncertainty and methods for correlated parameters. EVPPI methods have clear advantages over current methods of research prioritization, but we suggest that some specific sensitivity analyses are required before they can be used confidently in practice. These have limitations, and there is a need to develop robust methods to optimize research portfolios.