![[ Visit Client Website ]](images/banner.gif)
Tuesday, October 21, 2008: 12:00 PM
Grand Ballroom D (Hyatt Regency Penns Landing)
Background: Probabilistic sensitivity analysis, including value of information analysis, is well-developed as a means of expressing uncertainty when there is a limited number of clearly distinct decision options. However, realistic medical decision making often involves choice between a very wide range of options. The probability that the preferred intervention is cost-effective, and the Expected Value of Perfect Information (EVPI), are both highly sensitive to the choice of allowed options in the model. Analysis based on an essentially arbitrary selection of options cannot reflect the full decision uncertainty.
Aim: To explore what happens to the EVPI as an increasingly large number of decision options is considered.
Methods: Two stylised models are used as the basis for this work. One involves the choice of an optimal age for a once in a lifetime screening program, while the other involves the disease severity at which more aggressive treatment should be initiated. In each case, there is a single continuous-valued decision variable. Initially, this is chosen from a restricted range of possibilities, but the range of allowable values is increased arbitrarily.
Results: Effects of more refined choice sets on the Cost-Effectiveness Acceptability Frontier (CEAF) and EVPI are shown. The CEAF collapses to zero for threshold Incremental Cost-Effectiveness Ratios at which screening at some age is preferred to no screening, while the EVPI stabilises at a non-zero figure.
Conclusions: The Cost-Effectiveness Acceptability Frontier is of limited use when the options modelled are an essentially arbitrary selection from a very large set of possible options. In contrast, the Expected Value of Perfect Information can be reasonably approximated by a model with a limited choice set. It should, however, be noted that a model with a very restricted choice set may overestimate the total uncertainty measured through the EVPI. This can happen when the limited model does not include an option close to the true optimal choice under current uncertainty that would be found from the “perfect” model including all choices.
Aim: To explore what happens to the EVPI as an increasingly large number of decision options is considered.
Methods: Two stylised models are used as the basis for this work. One involves the choice of an optimal age for a once in a lifetime screening program, while the other involves the disease severity at which more aggressive treatment should be initiated. In each case, there is a single continuous-valued decision variable. Initially, this is chosen from a restricted range of possibilities, but the range of allowable values is increased arbitrarily.
Results: Effects of more refined choice sets on the Cost-Effectiveness Acceptability Frontier (CEAF) and EVPI are shown. The CEAF collapses to zero for threshold Incremental Cost-Effectiveness Ratios at which screening at some age is preferred to no screening, while the EVPI stabilises at a non-zero figure.
Conclusions: The Cost-Effectiveness Acceptability Frontier is of limited use when the options modelled are an essentially arbitrary selection from a very large set of possible options. In contrast, the Expected Value of Perfect Information can be reasonably approximated by a model with a limited choice set. It should, however, be noted that a model with a very restricted choice set may overestimate the total uncertainty measured through the EVPI. This can happen when the limited model does not include an option close to the true optimal choice under current uncertainty that would be found from the “perfect” model including all choices.