CM&S Innovators Circle: Beyond the Nominal—Practical Uncertainty Quantification for Simulations

Estimates of uncertainty can inform engineering and regulatory decisions throughout the medical device total product life cycle (TPLC). Geometric, material, and manufacturing variability anticipated to influence device safety or performance informs design specifications; physiological variability informs appropriately conservative targets for nonclinical testing; and anticipated uncertainty in clinical outcomes informs trial design and sample-size requirements. In this talk, we review practical uncertainty quantification (UQ) approaches, with a specific focus on propagating input uncertainties through numerical models to estimate their impact on model outputs.

We first examine how assumed input probability distributions influence uncertainty in finite element strain predictions for a nitinol surrogate specimen using Monte Carlo analyses with Latin hypercube sampling. We then consider laser-cut nitinol diamond specimens to demonstrate how sensitivity analysis and first-order second-moment methods can closely reproduce Monte Carlo estimates of strain variability with fewer simulations. Comparisons of predicted strain fields with digital image correlation measurements provide complementary evidence for assessing constitutive model assumptions.

Finally, we discuss broader applications of UQ throughout the medical device TPLC, including opportunities to inform device development, improve nonclinical testing, and enable in silico clinical trials by predicting how real-world variability in devices, patients, and clinical practice influences device performance.