Where New and Old Technologies Converge: In Search of a Better Way to Predict Pathologic Fracture

Pathologic fractures due to metastatic disease are important predictors of worse survival1,2 and are associated with diminished quality of life and increased pain and disability. Pathologic fracture, as an initial symptom of metastatic bone disease, is observed in 10% to 15% of patients with cancer1 and signals a more advanced stage of disease. The number of patients who develop a long-bone metastasis and present with an “impending pathologic fracture” is a much smaller but important subgroup. Among these patients, 10% to 25% meet the criteria for prophylactic internal fixation due to “perceived risk” based on a model that relies on radiographic interpretation3. This is notable because patients who undergo internal fixation for impending versus completed pathologic fractures of the long bones have improved outcomes and better quality of life4. Detecting the presence of a pathologic fracture is, at times, challenging, and a valid artificial intelligence (AI)-based method to assist in the diagnosis of a pathologic fracture is clinically useful. In their study, Kim et al. asked the question: can AI accurately predict the presence of a neoplastic pathologic fracture versus nonpathologic fracture on hip radiographs with accuracy that is comparable with that of experienced observers? The answer was yes, and if they could package and franchise this AI methodology, the ability to predict that a fracture is a neoplastic fracture could be standardized and improved. Early identification of impending pathologic fractures is also important, and proper and timely intervention matters. Yet, no universally accepted definition of an impending fracture currently exists. Clinicians frequently use the Mirels score5, although this tool is limited by the observers’ experience in interpreting radiographs, which can vary across specialties and clinical settings. Even with expert radiographic interpretation, the classification of impending pathologic fractures tends to overestimate fracture risk, leading to unnecessary surgical intervention. The importance of choosing the appropriate time to intervene is not trivial because potential complications from operative intervention in this medically compromised group of patients include infection, hardware failure, nonunion or delayed healing, hemorrhage, thrombotic events, sepsis, the need for reoperation, disease progression, and persistent pain and limited function6. Going forward, the formidable challenge is how 2 technologies, one emerging and the other, old and established, can be combined to produce a more meaningful estimation of fracture risk. Having a tool that is available to all clinicians, especially those with limited experience, is valuable. Could the synergy of AI technology and radiographs advance patient care beyond matching experienced observers to also improve our ability to predict fracture risk? A comprehensive risk-assessment tool ideally should incorporate a variety of factors, including, but not limited to, age, sex, body mass index, bone mineral density, history of radiation or chemotherapy, activity level, and bone involved, in addition to standardized and enhanced radiographic criteria. While AI can learn and improve its risk-assessment capabilities, validating its predictions is a substantial hurdle. A potential solution could involve a large-scale database(s) containing clinical and radiographic data from all patients with long-bone metastatic disease. AI could then analyze both clinical and radiographic data to identify factors associated with pathologic fractures, helping us to better predict fracture risk. Another major challenge to this risk-assessment tool is the lack of standardized radiographic data. One solution is to leverage a standardized, commercial web-based system that converts radiographs into a consistent DICOM (Digital Imaging and Communications in Medicine) format for AI analysis, thereby eliminating variability in image quality. A more advanced tool would combine an AI-generated risk profile that analyzes clinical, historical, and radiographic parameters to produce a single validated and reproducible risk score. The authors should be commended for advancing our understanding of pathologic fracture risk through the application of AI. Their findings demonstrate that an AI platform can perform comparably to experienced clinicians in predicting the presence of a pathologic fracture. This raises the possibility of expanding the use of AI in conjunction with radiographic assessment to further refine risk stratification, improve clinical outcomes, and enhance quality of life for patients with metastatic bone disease.