Identifying Individuals Predisposed to Hip Fracture

Abstract

The incidence of hip fracture in western society is becoming an increasingly serious concern as the population ages. While the economic and social costs are on the rise, the number of patients receiving treatment for bone fragility is in decline. Addressing this treatment gap should begin by investigating the relatively weak diagnostic technology currently being utilized in standard clinical practice. With problems including low sensitivity and a high rate of over diagnosis, it is clear that better tools are required for identifying individuals predisposed to hip fracture. Finite element (FE) models of the proximal femur derived from computed tomography have been proposed as a potential alternative. These simulations have outperformed areal bone mineral density (aBMD) measurements in predicting femoral strength within cadaver experiments, however FE models have not significantly outperformed conventional density measurements when predicting hip fracture from retrospective clinical data. The objective of this thesis was to address the limitations of state-of-the-art FE simulations of hip fracture in order to develop an improved hip fracture prediction model. A hip fracture is the result of a fall onto the hip, leading to an impact force greater than the load bearing capacity of the proximal femur. The three components of hip fracture, i.e. fall risk, fall severity, and bone strength, have typically been investigated individually, and no study thus far has attempted to fully characterize risk of fracture using a holistic understanding of all three aspects. This thesis attempts to address this knowledge gap, and was divided into three aims, each designed to enhance the biofidelity of FE simulations: In the first study, the anisotropy of trabecular bone was measured within numerous sub-volumes extracted from high-resolution scans using the conventional mean intercept length measurement. These anisotropic properties were subsequently mapped to FE models of the proximal femur, loaded in a sideways fall configuration. For the first time, isotropic and anisotropic models were compared using different material mapping strategies representing the extremes of previously published modeling techniques. The resulting whole bone stiffness and surface strains were then compared to cadaver experiments tested with similar boundary conditions. The addition of anisotropy had very little effect on whole bone stiffness and only a small effect on the resulting surface strain. Differences in principal compressive strain were identified in the femoral head, neck, and greater trochanter. The study concluded that anisotropic material properties, mapped using morphological measurements of high-resolution CT scans, has little impact on macroscopic, organ-level properties, but anisotropy could still affect localized internal strains, and could therefore be relevant depending on the modeling objectives. In the second study, a novel material mapping strategy was presented for explicit FE models of the proximal femur, validated with drop tower experiments. These non-linear material properties were designed to account for large deformations that could occur during an impact load on the hip. These properties included tensile damage, compressive densification, as well as tension/compression asymmetry and strain rate dependency. For the first time, the ultimate force predicted by the dynamic FE models was correlated with the ultimate force measured in dynamic experiments. Additionally, the simulated impulse response was strongly correlated with the force-time response measured in the drop tower experiments. Thus, these results represent the current benchmark in dynamic FE modeling of the proximal femur. Compressive strain rates over 100/s were observed in elements located in the femoral head, neck, and greater trochanter, suggesting that a better understanding of the strain rate dependency of bone tissue is still required for these loading rates. The dynamic models were able to simulate fracture patterns initiating in the sub-capital region, but were not able to predict inter-trochanteric fractures, suggesting that the material mapping strategy could still be improved in this anatomical region. The third study presented a biofidelic FE modeling technique that included models of the pelvis, soft tissue, and lower extremities that were morphed based on subject-specific biometrics. This approach was tested in a large, retrospective, clinical study to determine whether hip fracture classification models based on FE simulations provided a more accurate prediction of hip fracture incidence compared to clinical standard aBMD measurements. Logistic regression models based on simulated ultimate femur force, and surface strain in the femoral neck, marginally outperformed aBMD in terms of sensitivity and specificity, however the differences were not significant until subjects that did not report falling at baseline were excluded from the analysis. This resulted in a statistically significant difference between the simulated surface strain at the femoral neck and total femur aBMD, with AUC values of 0.85 and 0.74, respectively. These results indicate that subject-specific fall risk must be accurately estimated before mechanical assessments of fracture risk can be compared. This also suggested that fall risk could be a useful parameter for pre-screening hip fracture risk in the population. Additionally, the large number of biofidelic models tested in this study provided a novel measurement of subject-specific impact force transmitted to the proximal femur. A linear regression model found that the variance in ultimate femur force could be explained as a function of soft tissue thickness, pelvis width, and femoral head radius (R$^2$ = 0.79; RMSE = 0.46 kN). This thesis has made several important contributions to the field of hip fracture biomechanics. It has demonstrated that simulating anisotropic tissue-level properties had minimal effect on organ-level mechanics, relative to changes in the material properties. A novel material mapping strategy for explicit FE models of the proximal femur was designed in the absence of model tuning, and subsequently validated against drop tower experiments. The importance of subject-specific soft tissue thickness and fall risk was demonstrated using biofidelic FE models tested in a large retrospective study, suggesting that femoral strength alone is insufficient for accurately characterizing hip fracture risk. Finally, this thesis has presented the strongest evidence, to date, that FE simulations can predict hip fracture more accurately than conventional aBMD measurements. However, this observation was statistically significant only when subjects with high risk of falling were tested. Looking forward, the retrospective study presented in this thesis could serve as the baseline for future in silico trials simulating prophylactic interventions using biofidelic FE models. This could enable a novel estimate of the potential reduction in hip fracture incidence resulting from a preventative treatment. In conclusion, FE simulations of have the potential to improve both diagnostic and treatment technology, which could lead to meaningful change in the current standard of care for individuals predisposed to hip fracture.