Sampling strategies for approximating patient variability in population-based finite element studies of total hip...

Total hip replacements must be robust to patient variability for long-term success in the population. The challenge during the design process is evaluating an implant in a diverse population but the computational cost of simulating a population of subject-specific finite element (FE) models is not practical. We examined five strategies to generate representative subsets of subjects from a cohort of 103 implanted hip joint FE models to approximate the variability in output metrics. Comparing with the median and distribution of the 95^th percentile composite peak micromotion (CPM) and polar gap in the full cohort (CPM median: 136 μm, interquartile range [IQR]: 74-230 μm) (Polar Gap median: 467 μm, IQR: 434-548 μm), the Anatomic Sampling strategy (12 subjects) achieved the best balance of computational cost and approximation of the output metrics (CPM median: 169 μm, IQR: 78-236 μm) (Polar Gap median: 469 μm, IQR: 448-537 μm). Convex hull sampling (41 subjects) more closely captured the output metrics (CPM median: 99 μm, IQR: 70-191 μm) (Polar Gap median: 456 μm, IQR: 418-533 μm) but required over three times the number of subjects. Volume reduction of the convex hull captured the extremes of variability with subsets of 5 to 20 subjects, while the largest minimum-distance strategy captured the variability toward the middle of the cohort. These strategies can estimate the level of variability in FE model output metrics with a low computational cost when evaluating implants during the design process.