Over 300,000 hip fractures occur each year in the U.S. and up to 25% of hip-fracture patients die within a year of their injury. Despite the importance of this clinical problem, the diagnostic screening rate for osteoporosis is only 5% of the eligible population, and the sensitivity of measuring bone mineral density (BMD) by DXA, the clinical standard test for diagnosis, is only 50%. Therefore, most patients are not being screened diagnostically for osteoporosis, and for those who are, about half who will experience a hip fracture are missed. Given that the current empirical approach is inadequate, we propose to pursue a more mechanistic approach, combining state-of-the-art biomechanics and machine learning approaches. Biomechanically, the three etiological elements of hip fracture are fall risk, femoral strength, and femoral impact force. In this project, our overall goal is to provide a deeper understanding of how all three biomechanical etiological elements interact in the event of a hip fracture and from that, directly improve clinical fracture risk assessment through the use of a single predictive ?Integral Biomechanical Risk (IBR)? parameter. In addition, we will also address the problem of low DXA screening rates by further developing our Biomechanical Computed Tomography (BCT) technology. This test estimates the breaking strength of the femur using finite element analysis of routine clinical CT scans previously acquired for any medical reason, and represents an improvement compared to the use of BMD alone. Since millions of patients are scanned with CT each year, this approach could double screening rates if offered as an alternative to DXA. The proposed study will investigate this biomechanical approach in a large incident hip fracture, case-cohort study (3,000 patients with hip fracture, 6,000 without). This retrospective study will include patients seen at Kaiser Permanente who had an abdominal CT scan as part of medical care prior to any hip fracture; and have standard geriatric measurements in their electronic medical records, which we will use to estimate fall risk. Specifically, our aims are to: 1) utilize electronic medical record data and CT scans to obtain patient-specific measurements related to fall risk, femoral strength, and fall severity, and 2) combine the different elements of hip fracture etiology into the IBR parameter to test the hypothesis that this metric predicts hip fracture independent of age, sex, BMI, race/ethnicity, and history of prior fracture and improves hip fracture prediction compared to the clinical standard (BMD with FRAX). Scientifically, a major novelty of this work is its use of contemporary machine learning algorithms to inform construction of a mechanistic model of the three etiological elements of hip fracture, which should better capture any interactions between these elements compared to a purely statistical-regression approach. In addition, the study cohort will be the largest and most diverse CT-based hip fracture cohort ever assembled. Importantly, positive results from this project would provide a compelling ?second front? to DXA that could be quickly translated to widespread clinical practice, profoundly impacting osteoporosis care.
STATEMENT OF RELEVANCE More than half of individuals who experience hip fracture do not have osteoporosis as assessed by DXA. This project draws together the latest technological advances in CT-based finite element modeling and combines it into a probabilistic model of fracture risk that uses as inputs, data typically available in medical records. By going beyond BMD, the fracture risk prediction tool developed by this work aims to significantly improve clinical fracture risk assessment and substantially impact the preventative care and treatment of osteoporosis.
