ESTIMATING THE EXPECTED VALUE OF PERFECT PARAMETER INFORMATION (EVPPI): APPLICATIONS OF SHORT-CUT ALGORITHMS

Purpose:    To demonstrate the use of short-cut algorithms for estimating the expected value of perfect parameter information (EVPPI) for health policy models.

Method:    EVPPI analysis is an increasingly important metric used for prioritizing research funding and health policy decision making.  However, the traditional approach for calculating EVPPI is extremely burdensome and often infeasible.  We demonstrate two applications of short-cut algorithms to calculate EVPPI for one or a subset of health policy model parameters.     In the first application, we apply a 1-level short-cut algorithm to several individual and combined parameters in a Markov-based model of hepatitis C.  We set all parameters others than those considered in the EVPPI analysis to their baseline mean values, and estimated model results based on an array of 1,000 EVPPI parameter values calculated based on Latin Hypercube sampling.     In the second application, we performed a 2-level EVPPI calculation for a single parameter using a microsimulation model of eye diseases.  We calculated fully probabilistic results based on 40 values of the EVPPI parameter arrayed across the parameter space.  We derived a function of outcomes based on the EVPPI parameter, and used this function to estimate the EVPPI for the parameter of interest.

Result:   We found that employing the 1-level shortcut algorithm significantly reduced the computational load of estimating EVPPI. Because of the presence of other nonlinear parameters and skewed distributions, the incremental cost-effectiveness ratio (ICER) calculated with parameters set to the baseline mean values differed slightly from the ICER calculated with a fully probabilistic model.    Calculating the 2-level EVPPI in a fully probabilistic model was extremely computationally demanding.  To be computationally feasible, we limited the number of EVPPI parameter values and limited the inner loop sample size.  Due to these limitations and the small impact of the EVPPI parameter, the simulation results were subject to wide variability.  This variability prevented calculation of EVPPI based on actual model output, necessitating the estimation of a function of outcomes based on the EVPPI parameter value. 

Conclusion:    EVPPI is a powerful tool for measuring the impact of uncertainty on outcomes, but the traditional calculation approach is often infeasible.  Short-cut algorithms can vastly decrease computational requirements of calculating EVPPI and may allow EVPPI to become an increasingly common metric for assessing policy implementation and research funding allocation.