In Silico evaluation of treatment strategies to reduce proximal femur fracture risk in the fragile elderly

This thesis tackles a persistent clinical and public health problem: predicting hip fracture incidence in ageing populations where risks remains high despite mature diagnostics and established therapies. Current practice relies on densitometry and multifactorial scores, yet these tools provide limited insight into the causal chain linking fall exposure, impact energy transfer, and the subject-specific capacity of the proximal femur to resist failure. The premise developed here is that hip-fracture incidence emerges from the interaction of these mechanisms and can be predicted more transparently by simulating the pathway from falls to structural failure, while propagating uncertainty end-to-end and keeping assumptions auditable. The work first summarises how ageing perturbs bone remodelling, geometry, and material quality, and how neuromuscular decline increases both fall frequency and impact severity. It then reviews risk-prediction approaches—from DXA-centred case finding to clinical algorithms and imaging-derived biomechanical surrogates—highlighting their value for diagnosis and monitoring, but motivating a mechanistic–stochastic approach for forecasting. BoneStrength is then introduced: a stochastic fall generator supplies exposure; a multiscale impact model maps falls to hip contact conditions with attenuation; and a finite-element model provides subject-specific femoral strength. Hip fracture is modelled as a threshold-crossing event during simulated falls, so cohort incidence becomes an emergent property of repeated events tested against individual capacity. Fall counts are sampled from an over-dispersed Negative Binomial process to reflect populations with many non-fallers and a minority of recurrent fallers. Credibility is assessed through retrospective external validation against placebo/control arms of concluded phase-III trials spanning heterogeneous baseline risk, with observed hip-fracture counts lying within predicted distributions, and a first set of applications is described. A tractable development path is outlined: refine subject-level fall propensity and functional measures as data allow, evolve skeletal modelling toward spatially resolved adaptation, and consolidate generalisability via cross-validation on additional datasets.