A New Approach to Calculating Expected Value of Sample Information for a Clinical Trial

• Activity Number: 186
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Robert A. Parker*+ and Pamela Pen-Erh Pei and Milton Weinstein
• Companies: Massachusetts General Hospital and Massachusetts General Hospital and Harvard School of Public Health
• Keywords: Cost-effectiveness ; Expected net benefit ; Monte-Carlo simulation ; Probabilistic Sensitivity Analysis ; Value of information

Abstract

In the cost-effectiveness framework, the value of a clinical trial - the expected value of sample information (EVSI) - is the difference between the expected net benefit of the decision made after and in the absence of a trial, due to the reduction in parameter uncertainty from the trial. The calculation of EVSI is performed for a range of willingness to pay values, and is built upon a probabilistic sensitivity analysis (PSA), which generates the distribution of outcomes conditional upon the pre- or post-trial joint distribution of parameters. Although these calculations are straightforward, the standard approach (Ades et al, MDM 2004;24:207) is computationally intensive involving nested simulations for the PSA. When each analysis itself is a Monte-Carlo simulation this can become computationally overwhelming, especially as calculations need to be assessed across a range of potential trials and trial sizes. We propose an alternative approach to EVSI calculations, which allows results from a single set of simulations to be applied across a range of study sizes, reducing the computational burden dramatically. A simple model comparing TB diagnostic strategies illustrates the approach.