Biofidelic finite element models for accurately classifying hip fracture in a retrospective clinical study of elderly women from the AGES Reykjavik cohort.

Clinical retrospective studies have only reported limited improvements in hip fracture classification accuracy using finite element (FE) models compared to conventional areal bone mineral density (aBMD) measurements. A possible explanation is that state-of-the-art quasi-static models do not estimate patient-specific loads. A novel FE modeling technique was developed to improve the biofidelity of simulated impact loading from sideways falling. This included surrogate models of the pelvis, lower extremities, and soft tissue that were morphed based on subject anthropometrics. Hip fracture prediction models based on aBMD and FE measurements were compared in a retrospective study of 254 elderly female subjects from the AGES-Reykjavik study. Subject fragility ratio (FR) was defined as the ratio between the ultimate forces of paired biofidelic models, one with linear elastic and the other with non-linear stress-strain relationships in the proximal femur. The expected end-point value (EEV) was defined as the FR weighted by the probability of one sideways fall over five years, based on self-reported fall frequency at baseline. The change in maximum volumetric strain (ΔMVS) on the surface of the femoral neck was calculated between time of ultimate femur force and 90% post-ultimate force in order to assess the extent of tensile tissue damage present in non-linear models. After age-adjusted logistic regression, the area under the receiver-operator curve (AUC) was highest for ΔMVS (0.72), followed by FR (0.71), aBMD (0.70), and EEV (0.67), however the differences between FEA and aBMD based prediction models were not deemed statistically significant. When subjects with no history of falling were excluded from the analysis, thus artificially assuming that falls were known a priori with no uncertainty, a statistically significant difference in AUC was detected between ΔMVS (0.85), and aBMD (0.74). Multivariable linear regression suggested that the variance in maximum elastic femur force was best explained by femoral head radius, pelvis width, and soft tissue thickness (R2 = 0.79; RMSE = 0.46 kN; p < 0.005). Weighting the hip fracture prediction models based on self-reported fall frequency did not improve the models' sensitivity, however excluding non-fallers lead to significant differences between aBMD and FE based models. These findings suggest that an accurate assessment of fall probability is necessary for accurately identifying individuals predisposed to hip fracture.