Menzies importance-sampling EVSI estimator
This method efficiently estimates how much value a future study might add in terms of reducing uncertainty in health economic models, without requiring many complex calculations.
At a glance
Use when
Estimating EVSI in complex health economic models with limited computational budget; when study designs vary and rapid iteration is needed
Avoid when
Study evidence is very strong and only Algorithm 1 is used; when prior samples are insufficiently diverse to support reweighting
Inputs
Prior parameter samples from probabilistic sensitivity analysis, hypothetical study data or design, model outcomes
Outputs
Expected value of sample information (EVSI), reweighted posterior parameter distributions
How it works
A method to estimate the expected value of sample information (EVSI) using importance sampling by reweighting prior parameter draws to approximate posterior distributions given hypothetical study data. Algorithm 1 provides a fast EVSI estimate with minimal model evaluations; Algorithm 2 improves accuracy using smoothing. Both outperform conventional 2-level Monte Carlo and Brennan-Kharroubi methods in RMSE and computational efficiency.
- HTA domains
- Costs & Economic Evaluation
- Categories
- Decision Curve Analysis
- Assumptions
- Prior parameter samples are representative; hypothetical study data can be summarized in a likelihood function; model linearity or smoothness supports smoothing in Algorithm 2
- Strengths
- Dramatically reduces computational burden (3-4 orders of magnitude fewer model evaluations); applicable to complex models; integrates with standard probabilistic sensitivity analysis; improved accuracy over alternatives in most scenarios
- Limitations
- Performance degrades with strong study evidence (Algorithm 1 underestimates EVSI); accuracy depends on sufficient outer-loop sample size; smoothing (Algorithm 2) may introduce bias in non-smooth models
- Also known as
- Menzies EVSI estimator, importance-sampling EVSI
Questions this answers
- › What is the expected value of collecting additional data in a health economic model?
- › How can EVSI be computed efficiently without nested Monte Carlo simulations?
- › How does study design impact the value of information?
- › Can EVSI be estimated accurately with limited model evaluations?
- › How does this method compare to existing EVSI estimators in accuracy and speed?
References & sources
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.