Femoral fractures are common injuries which, especially in the elderly, are often associated with substantial morbidity, worsening in quality of life, and increased chance of mortality. Leaving aside fractures caused by accidental falls, which are evidently unpredictable, a crucial issue is assessing accurately the fracture risk of pathological femurs, in which the loss of mechanical integrity is caused by clinical conditions such as osteoporosis or bone metastases. To this aim, clinicians currently rely on indices which, neglecting important subject-specific information related to the mechanical determinants of fracture (e.g., exact bone anatomy and loads), have shown a marked lack of specificity that not rarely has led to uncertainty and inadequacy in the selection of the most appropriate treatment. In the last two decades, patient-specific finite element (FE) modeling combined with diagnostic imaging from computed tomography (CT) has been shown to significantly improve the fracture risk prediction for pathological femurs compared to current clinical standards. However, the majority of existing computational studies focus on estimating femoral strength via linear elastic analyses, without replicating the complete failure process and therefore the fracture pattern evolution. This aspect is clinically relevant, as optimal management of patients highly depends on the fracture type and location. In this work, a modeling approach based on damage mechanics has been developed to investigate femoral failure mechanisms through advanced FE simulations. Patient-specific geometry and distributions of the material properties of two human femurs are derived from CT data. Progressive degradation of bone structural integrity is modeled using a quasi-brittle constitutive description coupled with an integral non-local regularization strategy, with the aim of mitigating pathological mesh dependence inherent in (local) damage-based approaches. The model predictions are validated against ex vivo mechanical tests conducted on both femurs, replicating a stance loading condition. The numerical results show good agreement in terms of load-displacement response as well as evolution of the fracture pattern.