Incorporating structural and other sources of uncertainty in decision models

Abstract

Decision models such as those used in health economic evaluations need to provide an assessment of the uncertainty surrounding conclusions in order to be useful for policy making. For instance, NICE guidelines state that "the uncertainty surrounding the estimates of clinical and cost effectiveness needs to be fully expressed". Unfortunately, this is difficult to implement in practice. Standard techniques such as probabilistic sensitivity analysis are typically only used to account for sample uncertainty in the data used to inform model parameters.

Other sources of uncertainty that can affect results include study biases, uncertainty around structural, model and methodological choices, as well as properties of the modellers themselves. These sources of uncertainty are particularly important in infectious disease models, both because of the complexity of model and methodological choices for which there are few guidelines to follow, and because changes in such choices can affect results in a non-linear fashion. However, they are also relevant to models of non-infectious diseases. This talk will propose a taxonomy of sources of uncertainty in models, suggest some ways they can be incorporated effectively, and give examples of recent economic models of infectious disease interventions where they have been important.