Adjusting Estimates of the Expected Value of Information for Implementation: Theoretical Framework and Practical Application

Background: Value of information (VoI) calculations give the expected benefits of decision making under perfect information (EVPI) or sample information (EVSI), typically on the premise that any treatment recommendations made in light of this information will be implemented instantly and fully. This assumption is unlikely to hold in health care; evidence shows that obtaining further information typically leads to improved rather than perfect implementation. Objectives: To present a method of calculating the expected value of further research that accounts for the reality of improved implementation. Methods: This work extends an existing conceptual framework by introducing additional states of the world regarding information (sample information, in addition to current and perfect information) and implementation (improved implementation, in addition to current and optimal implementation). The extension allows calculating the implementation-adjusted EVSI (IA-EVSI), a measure that accounts for different degrees of implementation. Calculations of implementation-adjusted estimates are illustrated under different scenarios through a stylized case study in non-small cell lung cancer. Results: In the particular case study, the population values for EVSI and IA-EVSI were 25 pound million and 8 pound million, respectively; thus, a decision assuming perfect implementation would have overestimated the expected value of research by about 17 pound million. IA-EVSI was driven by the assumed time horizon and, importantly, the specified rate of change in implementation: the higher the rate, the greater the IA-EVSI and the lower the difference between IA-EVSI and EVSI. Conclusions: Traditionally calculated measures of population VoI rely on unrealistic assumptions about implementation. This article provides a simple framework that accounts for improved, rather than perfect, implementation and offers more realistic estimates of the expected value of research.