Quick & clean: computationally efficient methods for Value of Information measures

Abstract: Recently, there has been much research devoted to developing computationally efficient methods for various measures of the Value of Information in health economics, including the Expected Value of Partial Information (EVPPI) and the Expected Value of Sample Information (EVSI). I will present two sets of methods, one based on computationally efficient Gaussian Process regression based on Integrated Nested Laplace Approximation to compute the EVPPI and the other based on "moment-matching" to compute the EVSI. Both methods draw on existing methodologies but expand them by providing general-purpose algorithms that can be applied on a wide range of real-life modelling structures. I will present the methods using toy and real-life examples and discuss their advantages and limitations with respect to applicability in practice.