Strong-Oakley-Brennan GAM regression method for EVPPI
This method estimates how much value there would be in reducing uncertainty about specific parameters in a health economic model. Instead of using slow, complex simulations, it uses regression on existing simulation results to quickly estimate this value, saving time and computational resources.
At a glance
Use when
Estimating partial EVPPI in health economic models where computational efficiency is important, especially when nested sampling is impractical or when using existing PSA outputs.
Avoid when
The number of parameters of interest is very high (e.g., >5–6) without dimension reduction, or when PSA samples are too small to support stable regression fitting.
Inputs
Probabilistic sensitivity analysis sample: joint draws of input parameters and corresponding net benefit values from a health economic model.
Outputs
Partial expected value of perfect information (EVPPI) for specified parameters or parameter groups.
How it works
The Strong-Oakley-Brennan method estimates partial expected value of perfect information (EVPPI) using nonparametric regression models—specifically Generalized Additive Models (GAM) or Gaussian processes—applied to probabilistic sensitivity analysis (PSA) samples. It avoids the computational burden of traditional 2-level Monte Carlo methods by regressing net benefit outcomes on input parameters of interest, enabling efficient EVPPI estimation even with complex models and correlated parameters.
- HTA domains
- Clinical Effectiveness, Costs & Economic Evaluation, Aspects Beyond HTA
- Assumptions
- The relationship between the input parameters of interest and the net benefit can be adequately modeled using nonparametric regression (e.g., GAM or Gaussian process); the PSA sample is sufficiently large and representative of the joint parameter distribution.
- Strengths
- Computationally efficient compared to 2-level Monte Carlo methods,Uses only standard PSA samples, avoiding complex nested sampling,Handles models of high complexity and flexible parameter distributions,Can accommodate correlation among input parameters
- Limitations
- Performance depends on sample size and dimensionality (curse of dimensionality in regression),Nonparametric regression may struggle with very high-dimensional parameter sets,Accuracy depends on appropriate model specification and validation
- Also known as
- Regression-based EVPPI, GAM regression method for EVPPI, Strong et al. method for partial EVPI
Questions this answers
- › What is the value of eliminating uncertainty in a specific subset of parameters in a decision model?
- › Which parameters contribute most to decision uncertainty in terms of economic value?
- › How can EVPPI be estimated efficiently without nested Monte Carlo simulations?
- › Can partial EVPI be computed when input parameters are correlated?
- › Is it possible to estimate EVPPI using only a standard PSA sample?
- › What nonparametric regression approaches are suitable for EVPPI estimation?
References & sources
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.