Gaussian process emulation for individual-patient simulation models (Stevenson et al.)
This method uses Gaussian process emulation to speed up individual-patient simulation models in health economic evaluations. Instead of running slow, complex simulations repeatedly, a faster statistical model (the emulator) learns from a few simulation runs and predicts outcomes almost instantly, while preserving accuracy.
At a glance
Use when
Working with computationally intensive individual patient simulation models requiring probabilistic sensitivity analysis; when traditional methods are too slow or infeasible.
Avoid when
When the simulation model is already fast; when input-output relationships are highly discontinuous or chaotic; when insufficient resources exist to build and validate the emulator.
Inputs
Inputs to the original individual patient simulation model (e.g., patient characteristics, treatment parameters, cost and utility values, transition probabilities).
Outputs
Emulated outputs of the simulation model (e.g., mean cost, mean effectiveness, cost-effectiveness ratios, uncertainty distributions).
How it works
Gaussian process emulation is applied to an individual patient-level simulation model to create a fast, statistical surrogate that approximates the relationship between model inputs and outputs. The emulator enables rapid probabilistic sensitivity analyses by drastically reducing computational time—from 150 minutes to near-instantaneous—while maintaining high accuracy, thus facilitating the handling of both first- and second-order uncertainty in cost-effectiveness modeling.
- HTA domains
- Clinical Effectiveness, Costs & Economic Evaluation
- Assumptions
- The relationship between inputs and outputs is sufficiently smooth and can be accurately captured by a Gaussian process; a limited set of simulation runs is representative of the full input space.
- Strengths
- Dramatically reduces computational time; enables extensive sensitivity analyses; maintains high accuracy; handles complex, individual-level patient histories better than simpler models.
- Limitations
- Requires initial runs of the full simulation to train the emulator; performance depends on appropriate selection of training points and covariance structure; may not capture abrupt nonlinearities outside training range.
- Also known as
- Gaussian process emulation, Stevenson et al. method, emulation for individual patient models
Questions this answers
- › How can we reduce computational time in individual patient simulation models?
- › How can probabilistic sensitivity analysis be made feasible for complex models?
- › What is the relationship between model inputs and outputs in a patient-level simulation?
- › How can we handle first- and second-order uncertainty efficiently?
- › Can we maintain accuracy while accelerating health economic models?
References & sources
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.