Value of Information

Expected value of sample information (EVSI) 1 measures the average net-benefit gain from conducting new research and can be used to inform decisions on which new studies to fund and how best to design those studies. This helps avoid wasting resources researching treatments that were never likely to be cost-effective or conversely by adopting treatments that, if more evidence were collected, may be shown not to be cost-effective. However, the calculations in the general case rely on nested simulations, which can be very computationally demanding and even infeasible to compute in some cases. Since EVSI needs to be repeatedly computed over the potential study design space, this represents a clear barrier to the uptake of EVSI methods in practice. In some special situations, algebraic solutions are available that avoid the inner simulation step. 2–4 More generally, meta-modeling, which attempts to build a model to approximate the relationship between the model inputs (on which a new study can provide information) and model outputs (net benefit), is a promising approach that can lead to substantial computational savings. 5,6 In this issue, 2 novel meta-modeling methods are proposed for the calculation of EVSI, 7,8 both of which require only