Strong-Oakley nonparametric regression EVSI method
A fast method to estimate how much value collecting new data would bring in a health technology assessment, without needing complex simulations.
At a glance
Use when
Estimating EVSI quickly from existing PSA results, when 2-level Monte Carlo is too slow or infeasible, or when posterior sampling is difficult
Avoid when
The data-generating process is intractable, or when the regression relationship between parameters and net benefits is highly nonlinear or unstable
Inputs
Probabilistic sensitivity analysis sample (parameter draws and corresponding net benefits), ability to simulate data sets
Outputs
Per-patient expected value of sample information (EVSI)
How it works
A nonparametric regression-based method for estimating per-patient expected value of sample information (EVSI) using only the probabilistic sensitivity analysis sample. It avoids the computationally intensive 2-level Monte Carlo procedure by regressing net benefits on sampled parameter values, enabling efficient EVSI estimation without re-sampling posteriors or re-running the model.
- HTA domains
- Clinical Effectiveness, Costs & Economic Evaluation, Aspects Beyond HTA
- Assumptions
- The probabilistic sensitivity analysis sample is representative of the joint parameter distribution; the relationship between parameters and net benefits can be captured via regression; data can be simulated under the proposed study design
- Strengths
- Computationally efficient, avoids inner Monte Carlo loop, works with any model complexity and parameter distribution, uses existing PSA outputs
- Limitations
- Relies on quality of regression fit; requires ability to simulate data; accuracy depends on sample size and regression method choice
- Also known as
- Regression-based EVSI method, Strong and Oakley method
Questions this answers
- › How valuable is it to collect more data for a specific health technology?
- › What is the expected net benefit of a proposed research study?
- › How much can decision uncertainty be reduced by future data?
- › Can we estimate EVSI without running complex simulations?
- › Is there a faster alternative to the 2-level Monte Carlo method for EVSI?
- › How can EVSI be computed efficiently from existing PSA samples?
References & sources
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.