Targeted computational probabilistic corroboration of experimental knee wear simulator: The importance of accounting for variability

Experimental testing is widely used to predict wear of total knee replacement (TKR) devices. Computational models cannot replace this essential in vitro testing, but they do have complementary strengths and capabilities, which make in silico models a valuable support tool for experimental wear investigations. For effective exploitation, these two separate domains should be closely corroborated together; this requires extensive data-sharing and cross-checking at every stage of simulation and testing. However, isolated deterministic corroborations provide only a partial perspective; in vitro testing is inherently variable, and relatively small changes in the environmental and kinematic conditions at the articulating interface can account for considerable variation in the reported wear rates. Understanding these variations will be key to managing uncertainty in the tests, resulting in a 'cleaner' investigation environment for further refining current theories of wear. This study demonstrates the value of probabilistic in silico methods by describing a specific, targeted corroboration of the AMTI knee wear simulator, using rigid body dynamics software models. A deterministic model of the simulator under displacement-control was created for investigation. Firstly, a large sample of experimental data (N> 100) was collated, and a probabilistic computational study (N> 1000 trials) was used to compare the kinetic performance envelopes for in vitro and in silico models, to more fully corroborate the mechanical model. Secondly, corresponding theoretical wear-rate predictions were compared to the experimentally reported wear data, to assess the robustness of current wear theories to uncertainty (as distinct from the mechanical variability). The results reveal a good corroboration for the physical mechanics of the wear test rig; however they demonstrate that the distributions for wear are not currently well-predicted. The probabilistic domain is found to be far more sensitive at distinguishing between different wear theories. As such we recommend that in future, researchers move towards probabilistic studies as a preferred framework for investigations into implant wear.