Endocrine Abstracts

Given that osteoporosis is asymptomatic until a fragility fracture occurs, identifying patients at high fracture risk is a critical component of osteoporosis management. Traditional methods have used enhanced case finding to identify patients with risk factors such as specific medications (glucocorticoids, aromatase inhibitors), specific comorbidities (rheumatoid arthritis), or presenting with a fragility fracture. Screening of the general older female population has also been tested, which reduces hip fracture risk. AI offers the opportunity to enhance patient identification through two broad strategies: clinical data and imaging. AI algorithms have been trained on electronic health record data to predict short-term fracture risk successfully. AI algorithms have been trained to identify vertebral fractures and estimate bone mineral density using multiple imaging modalities, including plain X-ray, CT, and DXA VFA. A key element in vertebral fracture identification is the definition used to train models, including semi-quantitative, fully quantitative approaches, and algorithm-based qualitative methods. In addition, algorithms are being trained to estimate bone density from images taken without synchronous calibration phantoms and algorithms that combine skeletal and non-skeletal features. However, identifying patients at high fracture risk is insufficient to provide clinical benefit. Clinical services introducing these tools into clinical practice need to address local data regulatory and IT infrastructure requirements and develop the clinical pathway capability and capacity needed to translate these AI innovations into real patient and societal benefits.