Meeting Brochure and registration form      SMDM Homepage

Monday, October 22, 2007
P2-6

TOWARDS A BETTER UNDERSTANDING OF THE EVPI MEASURE

H. Koffijberg and M. P. Janssen. University Medical Center Utrecht, Utrecht, Netherlands

Purpose: The expected value of perfect information (EVPI) is recently introduced into the field of HTA. It represents the economical value of additional research on parameter values in cost-effectiveness models, and is already widely used. Its interpretation is straightforward, but this comes at a cost: differences in costs and in effects are combined into one single (cost) dimension, using a cost-effectiveness threshold, and relevant information in these separate two dimensions is lost.

Methods: In a simulation study we defined several scenarios in which we compared two treatment options and calculated the EVPI as function of the cost-effectiveness threshold. In each scenario we determined the EVPI contribution from cost differences and from effect differences separately. In addition, we derived the analytical form of the EVPI as function of the differences in costs and effects, given some simplifying assumptions.

Results: Similar EVPIs were found in different scenarios (thus for dissimilar differences between the treatment options). Given a fixed cost-effectiveness threshold, a single EVPI could represent an improvement in effects combined with an improvement (decrease) in costs but also a more substantial improvement in costs combined with a deterioration (increase) in costs. This ambiguity is also apparent in the analytical form of the EVPI.

Conclusions: The EVPI is a non-injective function and thus may deprive policy makers of relevant information and lead to suboptimal decisions. Since it is not possible, using the EVPI, to discern additional research expected to lead to both cost and effect benefits from addition research expected to lead to effect benefits at increased costs, it is useful to calculate the cost and effect contributions separately. An expected return on investment calculation could then be used to prioritize different additional research projects when similar EVPIs are reported in models on different diseases.