ASME V&V 40: Credibility of Computational Models for Medical Devices
Concept: credibility matched to risk
Overview
We'd like to replace part of the strength testing of an orthopaedic implant with analysis. How much verification will be accepted?
ASME V&V 40 is the framework for exactly that. The key is not demanding equal rigour from every analysis. First define what decision the analysis supports (context of use), then rate how much that decision relies on the model (model influence) and how severe a wrong decision would be for patients (decision consequence). Together these set model risk, and higher risk demands stronger verification and validation evidence. Replacing tests raises model influence, so plan for substantial evidence.
Assessment flow
- State the question of interest.
- Define the context of use — how the model addresses the question.
- Set model risk from model influence and decision consequence.
- Set target rigour for each credibility factor (code verification, solution verification, input uncertainty, comparator, etc.) according to risk.
- Perform V&V, judge against the targets and document.
Model-risk matrix (example)
| Model influence \ Consequence | Low | Medium | High |
|---|---|---|---|
| Low | 1 | 2 | 3 |
| Medium | 2 | 3 | 4 |
| High | 3 | 4 | 5 |
Higher numbers mean higher risk. The standard doesn't mandate a specific table, but setting levels from the combination and scaling the evidence required at each level is the heart of the approach.
The road to “simulation instead of testing”
Medical devices long relied on physical, animal and clinical testing to demonstrate safety. From the 2000s computational models grew in importance — stent and joint fatigue, heating in MRI — and the US FDA encouraged their use in review. V&V 40 (2018) laid the foundation, linking the ideas of V&V 10 (structures) and V&V 20 (fluids and heat) to the risk of medical-device decisions. Its significance is in establishing rules for treating computational results as evidence.
Example: fatigue assessment of a hypothetical stent
Question and context of use
| Item | Example | |
|---|---|---|
| Question | Will a new stent size resist fatigue under pulsatile vessel deformation? | |
| Context of use | FEA finds the location and size of peak strain for comparison with the fatigue limit; physical tests are done only for existing sizes | |
| Model influence | High (new size judged by analysis alone) | |
| Decision consequence | High (fracture can cause serious harm) | |
| Model risk | 5 | 5 |
Number of validation specimens
When comparing analysis and test strains on the existing size, the number needed to estimate the mean difference with 95% confidence follows from specimen scatter (coefficient of variation) and tolerance via the $t$ distribution:
| Scatter (CV) | Tolerance ±10% | ±5% | ±3% |
|---|---|---|---|
| 5% | 4 | 7 | 14 |
| 8% | 5 | 13 | 30 |
| 12% | 9 | 25 | 64 |
Tightening the tolerance makes the specimen count explode.
Yes — halving the tolerance roughly quadruples the count. That's why V&V 40's approach helps: a risk-5 decision needs tight tolerance and many specimens, whereas a low-risk use might be fine at ±10% with a handful. The point of the standard is not 'strict everything' but choosing evidence matched to the decision's weight, with a documented rationale. These counts cover only the mean comparison; material scatter and test-rig uncertainty must be estimated separately.
Practical workflow
Evidence per credibility factor
- Code verification: software quality records and comparison with known solutions.
- Solution verification: mesh convergence, time-step effects, numerical uncertainty.
- Inputs and comparator: basis and uncertainty of material properties and loads, and how representative the test conditions are of real use.
- Applicability: how close the validation conditions are to the conditions of the actual decision (e.g. a new size).
“Validated — for a different use”
A model matched static tests within 2%, but the review asked about fatigue under cyclic loading, where material nonlinearity and surface condition matter; the static agreement was not sufficient evidence. That's why V&V 40 insists on defining the context of use first. Validation counts for the quantities and conditions used in the decision; good agreement on a different quantity is not evidence of the same credibility.
Common mistakes
Mistakes and fixes
| Mistake | Effect | Fix |
|---|---|---|
| V&V without a context of use | Evidence irrelevant to the decision | Document question and context first |
| Uniform rigour for everything | Excessive or insufficient effort | Scale to risk |
| Substituting agreement on another quantity | Not valid credibility evidence | Validate the decision quantity |
| Arbitrary specimen counts | Can't resolve differences | Compute from scatter and tolerance |
| Using outside validated range | Unknown credibility | Assess applicability |
I'd like to know more about V&V standards.
Structures are covered in ASME V&V 10, fluids and heat in ASME V&V 20, documentation in V&V plans and reports, grading evidence quality in pedigree matrix, and stepwise validation from components to systems in hierarchical validation. Try scatter calculations with the Monte Carlo statistics tool.
Related Topics
Experience the theory firsthand with the interactive simulator for this field
All Simulators