Two-Step Uncertainty Quantification Methodology for Medical Device Design: Influence of Input Parameter Probability Distribution on Output Uncertainty

When representing an input parameter for uncertainty quantification (UQ), use of a Gaussian distribution is ubiquitous, yet for some input parameters other types of probability distributions may be more appropriate. In this study, we compare the effect of input probability distribution choice on output uncertainty as predicted by nondeterministic finite element simulations of an implantable nitinol inferior vena cava (IVC) filter. Computational modeling and simulation are commonly combined with experimental fatigue testing data to predict the fatigue resistance of such cardiovascular devices. For applications where model credibility is crucial, characterization of model uncertainty is required to understand failure and safety factors. In our UQ methodology, data on three categories of input parameter – material properties, geometry, and experimental conditions – is gathered and statistically characterized across multiple manufacturing lots. After a sensitivity analysis step to exclude uninfluential parameters, three probability distributions – Gaussian, gamma, and uniform – are fit to the remaining input parameter datasets. Each of the three probability distributions are sampled using Latin hypercube sampling which serves as inputs to nondeterministic FE simulations that predict local strains and strain amplitudes. In this presentation, we will present this broadly applicable UQ methodology as applied to a medical device.