
1식품의약품안전평가원 의료제품연구부 의료기기연구과
1Medical Device Research Division, Department of Medical Product Research, National Institute of Food and Drug Safety Evaluation
Copyright © The Korean Society for Precision Engineering
This is an Open-Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
| No. [Ref.] | Assessment level | QOIs | COU | Model risk | Applied medical device |
|---|---|---|---|---|---|
| 1 [15] | Partial application | Are the flow-induced hemolysis levels of the centrifugal pump acceptable for the intended use? | COU1: CPB (Class II); COU2: Short-term VAD (Class III) | L2 (COU1)/L5 (COU2)a | Centrifugal blood pump |
| 2 [9] | ASME V&V 40 referenced | Prediction of metal–polyethylene contact areab | Design evaluation for TAA contact mechanicsb | - | Total ankle arthroplasty |
| 3 [10] | ASME V&V 40 referenced | Mechanical response of spinal rods under 3-point bendingb | UQ and validation for mechanical responseb | - | Spinal rod |
| 4 [11] | Full application | For an apically implanted LVAD, does the selected pump speed produce: (a) complete aortic valve opening >0.3[L/min]; and (b) a Cardiac output compatible with life>(4.2[L/min]) for a range of HR and EF covering a HF patient population? | See note c | L3a | LVAD |
| 5 [16] | Partial application | Which is the optimal effective dose for a new anti-osteoporosis drug in adults and older adults (from 55 years) according to multi-dose Phase II studies? | See note d | L3a | - |
| 6 [17] | Partial application | QOI for COU1: “How do decisions regarding the material and design influence the functional parameters of the custom-made 3D printed WHO?”; QOI for COU2: “How do decisions regarding the material and design influence the occurring strains and stresses on the custom-made 3D printed WHO?” | COU1: Performance evaluation of the functional properties of the custom-made 3D printed WHO; COU2: Superiority evaluation of the strain distribution of the custom-made 3D printed WHO | L2~3 (COU1)/L2 (COU2)a | Custom 3D-printed WHO (wrist hand orthosis) |
| 7 [18] | Partial application | “what is the most immunogenic dose of the new therapeutic vaccine to be used in patients affected by tuberculosis?” | See note e | L2a | - |
| 8 [12] | Full application |
1. What are the crimp strains and fatigue strains (strain amplitude and mean strain) and peak locations in the TAV under simulated in vivo conditions? 2. What test conditions are required to replicate in vivo strain amplitudes (and mean strains) for structural component fatigue testing 3. Will the TAV survive 600M cycles under in vivo loading, without fracture? |
To predict the fatigue strains in multiple device sizes under in vivo loading conditions. Results are used to identify the worst-case device size, location of peak fatigue strains and test conditions required to reproduce in vivo strain level in a benchtop structural component fatigue test. | Medium | TAV (transcatheter aortic valve) frame |
| 9 [13] | Full application | Does the hypothetical new total knee arthroplasty design provide sufficient resistance to wear of the polyethylene (PE) inlay under ISO 14243-1 and activities of daily living test conditions in displacement control? | WearPy is used to determine the amount of PE volumetric wear of the new design and identify the worst case condition (test and size). Bench testing will be performed on the identified worst-case. | Low-medium | Knee implant |
| 10 [14] | Full application | “does adding a 1.6 mm diameter cannulation to an existing 7.5 mm diameter pedicle screw design compromise mechanical performance of the rod-screw construct in static compression-bending?” | “model predictions of the original non-cannulated screw construct will be validated with benchtop testing per ASTM F1717 static compression bending conditions. The validated model framework will be used to evaluate the cannulated screw design undergoing identical static compression-bending conditions. No benchtop testing on the cannulated screw design will be performed to answer the ?OI.” | High-medium (L4) | Pedicle screw system |
aModel risk was categorized into five levels (L1–L5) in the selected literature.
bQOI and COU were not explicitly stated in the original publication and were inferred by the authors based on the study objectives and methodology.
cThe heart-LVAD computational model may be used by design engineers to assist in the preclinical development of LVAD, by characterising aortic root, LVAD and intra-LV flows for a given pump speed. The goal of the heart-LVAD computational model is to provide a computational replica of a benchtop experiment for a quantitative analyses in parametric explorations. The heart-LVAD computational model by no means is replacing animal experiments or clinical trials, but augmenting the totality of evidence.
dBBCT-hip is a methodology where a stochastic biophysics model provides an estimate, for a given subject, of the Absolute Risk of proximal femur Fracture upon falling at time zero (ARF0), from their height, weight, and a Quantitative Computed Tomography (QCT) scan of the hip region. This ARF0 is to be used as a response variable in multi-dose Phase II studies in place of the measured DXA-based aBMD. The average change in ARF0 over the period of treatment for all subjects treated with a given dose (AveΔARF0) can be used as response variable, by assuming the optimal dose amongst those tested is the one for which AveΔARF0 is most positive (or least negative).
eThe UISSTB-DR model will be used to support the decision about the most immunogenic dose of the new therapeutic vaccine against TB and inform phase II dose selection studies by predicting the human immune system response.
| No. | Activity | Credibility factor | Implementation strategy | Ref. | ||
|---|---|---|---|---|---|---|
| 1 | Verification | Code | SQA | ① Verification of regular quality maintenance | [10,11,18] | |
| ② Conducting simple internal quality assurance activities, such as identifying COU-related bugs | [14,15] | |||||
| ③ Verification of compliance with international quality management standards (third-party certification) | [12,13,16,17] | |||||
| 2 | NCV |
① Selection of benchmark solutions representing COU-related physical phenomena – Well-known solutions in relevant fields where analytical solutions exist – Benchmarks provided by the manufacturer – Numerical comparisons based on existing literature |
[10–18] | |||
| ② Alongside ①, conduct mesh or time step convergence studies and verify convergence | [11] | |||||
| 3 | Calculation | Discretization error | ① Perform mesh or time sensitivity analysis and evaluate convergence | [9–17] | ||
| ② Define permissible discretisation error criteria (e.g., ≤5%) and evaluate results | [10,14] | |||||
| ③ Apply standard convergence analyses such as GCI based on Richardson Extrapolation | [15] | |||||
| 4 | Numerical solver error | ① Parameter setting via literature, expert consultation, etc. | [11] | |||
| ② Solver parameter sensitivity analysis affecting numerical stability | [12,14,16,18] | |||||
| 5 | Use error | ① Practitioner directly reviews validity of key inputs (boundary conditions, material properties, etc.) and outputs | [12,13,16] | |||
| ② Internal peer independently reviews input files and modelling settings | [11,12,14,18] | |||||
| ③ Independent execution by two or more different organisations, with results compared to assess user and execution environment effects | [10] | |||||
| 6 | Validation | Computational model | Model form | ① Clearly present all model assumptions and simplifications (e.g., geometric simplifications, symmetry assumptions, boundary conditions, loading conditions, material models, physiological characteristics, etc.) | [9,10,13,15] | |
| ② Alongside ①, qualitatively/quantitatively analyse the impact of these assumptions on results | [11,12,14,16,18] | |||||
| 7 | Model input | Quantification of sensitivities | ① Conduct sensitivity analysis for key or comprehensive inputs to identify those with significant impact on results | [9–14,16] | ||
| 8 | Quantification of uncertainties | ① Quantify the uncertainty of key inputs identified through sensitivity analysis, etc. | [10,12,18] | |||
| ② Apply sampling-based uncertainty propagation techniques (LHS, Monte Carlo, etc.) to quantitatively assess the impact of input variable uncertainty on final outputs (QOIs) | [11,14–16] | |||||
| 9 | Comparator | Test samples | Quantity | ① Select two or more arbitrary sample quantities | [9,12] | |
| ② Select sample quantities according to relevant standards or guidelines | [10,13,14] | |||||
| ③ Select statistically significant sample quantities | [16] | |||||
| 10 | Range characteristics | ① Select samples representing typical usage conditions, including those corresponding to the Nominal Value | [14,16] | |||
| ② Select samples to include extreme values, such as those from high-risk groups or under severe conditions | [16] | |||||
| ③ Select diverse samples to represent the variety of the product/patient population | [12] | |||||
| 11 | Measurements | ① Measure some or all key characteristics of the test samples required for the comparator study | [9–14,16] | |||
| 12 | Uncertainty of test sample measurements | ① Manage measurement uncertainty by performing the study using an already calibrated measurement system | [10,13] | |||
| ② Measure test condition values repeatedly to reflect measurement uncertainty | [14,16] | |||||
| ③ Estimate the combined measurement uncertainty for the test samples | [12] | |||||
| 13 | Test conditions | Quantity | ① Establish two or more diverse test conditions representing the clinical usage environment or mechanical operating conditions the model aims to evaluate | [11–13,15–17] | ||
| ② Select condition quantities according to relevant standards or guidelines | [9,10,13,14] | |||||
| 14 | Range characteristics | ① Select a range including the most common or average usage conditions, or those specified by relevant standards or guidelines | [10,13,14] | |||
| ② Include boundary conditions or extreme conditions that could most significantly impact safety | [12,15] | |||||
| ③ Select conditions to cover the entire range, including various points between normal and extreme conditions | [9,11] | |||||
| 15 | Measurements | ① Measure some key or all conditions required for comparator study | [10–15,17] | |||
| 16 | Uncertainty of test condition measurements | ① Measure test condition values using an already calibrated measurement system | [13] | |||
| ② Measure test condition values repeatedly to reflect measurement uncertainty | [10,15,17] | |||||
| ③ Estimate the combined measurement uncertainty for the test conditions | [12,14] | |||||
| 17 | Assessment | Equivalency of input parameters | ① If identical values or units (types) cannot be used, employ similar types | [12,18] | ||
| ② Use the inputs from the comparator as inputs for the CM&S model | [9,11,13–16] | |||||
| 18 | Output comparison | Quantity | ① Select QOIs for multiple parameters when choosing them | [9–17] | ||
| 19 | Equivalency of output parameters | ① If identical variable values cannot be measured in the computational model, select a similar physical quantity and document the correlation | [12,16] | |||
| ② Match the output parameters of the comparator to the outputs of the computational model | [9–11,13–15] | |||||
| 20 | Rigor of output comparison | ① Compare visual similarities such as the shape and trend of output curves | [13,17] | |||
| ② Compare arithmetic differences in key metrics such as maximum values, mean values, and mean root mean square error (NRMSE) | [9,10,12–16] | |||||
| ③ Compare the uncertainty of computational model results or comparator results together (e.g., checking whether confidence intervals overlap) | [11,12,14,15,17] | |||||
| 21 | Agreement of output comparison | ① Where quantitative comparison is impossible, conduct a qualitative comparison of output consistency | [13] | |||
| ② Conduct a quantitative, statistical comparison of output consistency | [9–17] | |||||
| ③ Confirm whether validation acceptance criteria exist in the same or similar field before analysis | [9,16] | |||||
| 22 | Applicability | Relevance of the QOIs | ① Even if QOIs are not directly linked to the COU, explicitly state their mathematical/logical association with the COU and demonstrate this relationship | [12,16,17] | ||
| ② Select parameters directly related to the core issues of the COU as QOIs for validation activities | [9–11,13–18] | |||||
| 23 | Relevance of the Validation Activities to the COU | ① If validation activities cover only part of the COU, clearly describe these limitations | [11,14,16,18] | |||
| ② Design validation activities to encompass as much of the COU as possible | [10,12,13,15,17] | |||||
| No. [Ref.] | Assessment level | QOIs | COU | Model risk | Applied medical device |
|---|---|---|---|---|---|
| 1 [ |
Partial application | Are the flow-induced hemolysis levels of the centrifugal pump acceptable for the intended use? | COU1: CPB (Class II); COU2: Short-term VAD (Class III) | L2 (COU1)/L5 (COU2)a | Centrifugal blood pump |
| 2 [ |
ASME V&V 40 referenced | Prediction of metal–polyethylene contact areab | Design evaluation for TAA contact mechanicsb | - | Total ankle arthroplasty |
| 3 [ |
ASME V&V 40 referenced | Mechanical response of spinal rods under 3-point bendingb | UQ and validation for mechanical responseb | - | Spinal rod |
| 4 [ |
Full application | For an apically implanted LVAD, does the selected pump speed produce: (a) complete aortic valve opening >0.3[L/min]; and (b) a Cardiac output compatible with life>(4.2[L/min]) for a range of HR and EF covering a HF patient population? | See note c | L3a | LVAD |
| 5 [ |
Partial application | Which is the optimal effective dose for a new anti-osteoporosis drug in adults and older adults (from 55 years) according to multi-dose Phase II studies? | See note d | L3a | - |
| 6 [ |
Partial application | QOI for COU1: “How do decisions regarding the material and design influence the functional parameters of the custom-made 3D printed WHO?”; QOI for COU2: “How do decisions regarding the material and design influence the occurring strains and stresses on the custom-made 3D printed WHO?” | COU1: Performance evaluation of the functional properties of the custom-made 3D printed WHO; COU2: Superiority evaluation of the strain distribution of the custom-made 3D printed WHO | L2~3 (COU1)/L2 (COU2)a | Custom 3D-printed WHO (wrist hand orthosis) |
| 7 [ |
Partial application | “what is the most immunogenic dose of the new therapeutic vaccine to be used in patients affected by tuberculosis?” | See note e | L2a | - |
| 8 [ |
Full application | 1. What are the crimp strains and fatigue strains (strain amplitude and mean strain) and peak locations in the TAV under simulated in vivo conditions? 2. What test conditions are required to replicate in vivo strain amplitudes (and mean strains) for structural component fatigue testing 3. Will the TAV survive 600M cycles under in vivo loading, without fracture? |
To predict the fatigue strains in multiple device sizes under in vivo loading conditions. Results are used to identify the worst-case device size, location of peak fatigue strains and test conditions required to reproduce in vivo strain level in a benchtop structural component fatigue test. | Medium | TAV (transcatheter aortic valve) frame |
| 9 [ |
Full application | Does the hypothetical new total knee arthroplasty design provide sufficient resistance to wear of the polyethylene (PE) inlay under ISO 14243-1 and activities of daily living test conditions in displacement control? | WearPy is used to determine the amount of PE volumetric wear of the new design and identify the worst case condition (test and size). Bench testing will be performed on the identified worst-case. | Low-medium | Knee implant |
| 10 [ |
Full application | “does adding a 1.6 mm diameter cannulation to an existing 7.5 mm diameter pedicle screw design compromise mechanical performance of the rod-screw construct in static compression-bending?” | “model predictions of the original non-cannulated screw construct will be validated with benchtop testing per ASTM F1717 static compression bending conditions. The validated model framework will be used to evaluate the cannulated screw design undergoing identical static compression-bending conditions. No benchtop testing on the cannulated screw design will be performed to answer the ?OI.” | High-medium (L4) | Pedicle screw system |
| No. | Activity | Credibility factor | Implementation strategy | Ref. | ||
|---|---|---|---|---|---|---|
| 1 | Verification | Code | SQA | ① Verification of regular quality maintenance | [ | |
| ② Conducting simple internal quality assurance activities, such as identifying COU-related bugs | [ | |||||
| ③ Verification of compliance with international quality management standards (third-party certification) | [ | |||||
| 2 | NCV | ① Selection of benchmark solutions representing COU-related physical phenomena – Well-known solutions in relevant fields where analytical solutions exist – Benchmarks provided by the manufacturer – Numerical comparisons based on existing literature |
[ | |||
| ② Alongside ①, conduct mesh or time step convergence studies and verify convergence | [ | |||||
| 3 | Calculation | Discretization error | ① Perform mesh or time sensitivity analysis and evaluate convergence | [ | ||
| ② Define permissible discretisation error criteria (e.g., ≤5%) and evaluate results | [ | |||||
| ③ Apply standard convergence analyses such as GCI based on Richardson Extrapolation | [ | |||||
| 4 | Numerical solver error | ① Parameter setting via literature, expert consultation, etc. | [ | |||
| ② Solver parameter sensitivity analysis affecting numerical stability | [ | |||||
| 5 | Use error | ① Practitioner directly reviews validity of key inputs (boundary conditions, material properties, etc.) and outputs | [ | |||
| ② Internal peer independently reviews input files and modelling settings | [ | |||||
| ③ Independent execution by two or more different organisations, with results compared to assess user and execution environment effects | [ | |||||
| 6 | Validation | Computational model | Model form | ① Clearly present all model assumptions and simplifications (e.g., geometric simplifications, symmetry assumptions, boundary conditions, loading conditions, material models, physiological characteristics, etc.) | [ | |
| ② Alongside ①, qualitatively/quantitatively analyse the impact of these assumptions on results | [ | |||||
| 7 | Model input | Quantification of sensitivities | ① Conduct sensitivity analysis for key or comprehensive inputs to identify those with significant impact on results | [ | ||
| 8 | Quantification of uncertainties | ① Quantify the uncertainty of key inputs identified through sensitivity analysis, etc. | [ | |||
| ② Apply sampling-based uncertainty propagation techniques (LHS, Monte Carlo, etc.) to quantitatively assess the impact of input variable uncertainty on final outputs (QOIs) | [ | |||||
| 9 | Comparator | Test samples | Quantity | ① Select two or more arbitrary sample quantities | [ | |
| ② Select sample quantities according to relevant standards or guidelines | [ | |||||
| ③ Select statistically significant sample quantities | [ | |||||
| 10 | Range characteristics | ① Select samples representing typical usage conditions, including those corresponding to the Nominal Value | [ | |||
| ② Select samples to include extreme values, such as those from high-risk groups or under severe conditions | [ | |||||
| ③ Select diverse samples to represent the variety of the product/patient population | [ | |||||
| 11 | Measurements | ① Measure some or all key characteristics of the test samples required for the comparator study | [ | |||
| 12 | Uncertainty of test sample measurements | ① Manage measurement uncertainty by performing the study using an already calibrated measurement system | [ | |||
| ② Measure test condition values repeatedly to reflect measurement uncertainty | [ | |||||
| ③ Estimate the combined measurement uncertainty for the test samples | [ | |||||
| 13 | Test conditions | Quantity | ① Establish two or more diverse test conditions representing the clinical usage environment or mechanical operating conditions the model aims to evaluate | [ | ||
| ② Select condition quantities according to relevant standards or guidelines | [ | |||||
| 14 | Range characteristics | ① Select a range including the most common or average usage conditions, or those specified by relevant standards or guidelines | [ | |||
| ② Include boundary conditions or extreme conditions that could most significantly impact safety | [ | |||||
| ③ Select conditions to cover the entire range, including various points between normal and extreme conditions | [ | |||||
| 15 | Measurements | ① Measure some key or all conditions required for comparator study | [ | |||
| 16 | Uncertainty of test condition measurements | ① Measure test condition values using an already calibrated measurement system | [ | |||
| ② Measure test condition values repeatedly to reflect measurement uncertainty | [ | |||||
| ③ Estimate the combined measurement uncertainty for the test conditions | [ | |||||
| 17 | Assessment | Equivalency of input parameters | ① If identical values or units (types) cannot be used, employ similar types | [ | ||
| ② Use the inputs from the comparator as inputs for the CM&S model | [ | |||||
| 18 | Output comparison | Quantity | ① Select QOIs for multiple parameters when choosing them | [ | ||
| 19 | Equivalency of output parameters | ① If identical variable values cannot be measured in the computational model, select a similar physical quantity and document the correlation | [ | |||
| ② Match the output parameters of the comparator to the outputs of the computational model | [ | |||||
| 20 | Rigor of output comparison | ① Compare visual similarities such as the shape and trend of output curves | [ | |||
| ② Compare arithmetic differences in key metrics such as maximum values, mean values, and mean root mean square error (NRMSE) | [ | |||||
| ③ Compare the uncertainty of computational model results or comparator results together (e.g., checking whether confidence intervals overlap) | [ | |||||
| 21 | Agreement of output comparison | ① Where quantitative comparison is impossible, conduct a qualitative comparison of output consistency | [ | |||
| ② Conduct a quantitative, statistical comparison of output consistency | [ | |||||
| ③ Confirm whether validation acceptance criteria exist in the same or similar field before analysis | [ | |||||
| 22 | Applicability | Relevance of the QOIs | ① Even if QOIs are not directly linked to the COU, explicitly state their mathematical/logical association with the COU and demonstrate this relationship | [ | ||
| ② Select parameters directly related to the core issues of the COU as QOIs for validation activities | [ | |||||
| 23 | Relevance of the Validation Activities to the COU | ① If validation activities cover only part of the COU, clearly describe these limitations | [ | |||
| ② Design validation activities to encompass as much of the COU as possible | [ | |||||
Model risk was categorized into five levels (L1–L5) in the selected literature.
QOI and COU were not explicitly stated in the original publication and were inferred by the authors based on the study objectives and methodology.
The heart-LVAD computational model may be used by design engineers to assist in the preclinical development of LVAD, by characterising aortic root, LVAD and intra-LV flows for a given pump speed. The goal of the heart-LVAD computational model is to provide a computational replica of a benchtop experiment for a quantitative analyses in parametric explorations. The heart-LVAD computational model by no means is replacing animal experiments or clinical trials, but augmenting the totality of evidence.
BBCT-hip is a methodology where a stochastic biophysics model provides an estimate, for a given subject, of the Absolute Risk of proximal femur Fracture upon falling at time zero (ARF0), from their height, weight, and a Quantitative Computed Tomography (QCT) scan of the hip region. This ARF0 is to be used as a response variable in multi-dose Phase II studies in place of the measured DXA-based aBMD. The average change in ARF0 over the period of treatment for all subjects treated with a given dose (AveΔARF0) can be used as response variable, by assuming the optimal dose amongst those tested is the one for which AveΔARF0 is most positive (or least negative).
The UISSTB-DR model will be used to support the decision about the most immunogenic dose of the new therapeutic vaccine against TB and inform phase II dose selection studies by predicting the human immune system response.