New AI tool predicts hip fracture risk years ahead

Preventing hip fractures depends on identifying people at high risk early. Researchers at the University of Gothenburg have now developed a new clinical decision support tool that makes this possible-without requiring clinic visits or patient questionnaires.

Called FRACTURE-ML, the tool uses information already available in Sweden’s national health registers, including diagnoses, prescription medications, medical procedures, and demographic and socioeconomic data. It can serve as an initial screening tool to identify people who should be referred for further evaluation. Unlike existing fracture risk models, it does not rely on patient-reported information or physical examinations. 

The study is published in PLOS Medicine. It included more than 3.5 million people in Sweden aged 50 years and older. During the follow-up period, more than 142,000 people sustained a hip fracture.  

Early identification 

“Hip fractures often result in significant suffering, loss of independence, and increased mortality. At the same time, we know that many fractures can be prevented if people at high risk are identified early. Our model is very good at distinguishing between individuals at high and low risk and has the potential to become an important tool for preventive care,” says Mattias Lorentzon, Professor of Geriatric Medicine at Sahlgrenska Academy, University of Gothenburg. 

The researchers developed and compared several statistical models, including modern machine learning approaches. The final model was able to provide reliable individualized risk estimates one, two, five, and ten years into the future. 

“What makes the model unique is that it relies entirely on registry data, making it suitable for large-scale population screening without placing an additional burden on healthcare through extensive patient assessments. This creates opportunities for earlier intervention and more precise prevention,” says Kristian Axelsson, first author of the study and researcher at Sahlgrenska Academy, University of Gothenburg. 

Identifying more people at risk 

Today, people at high risk of fracture are typically identified only after they have already sustained a fracture through Fracture Liaison Services (FLS). By contrast, FRACTURE-ML can identify people at high risk before their hip fracture occurs. Compared with current practice, the tool could identify nearly seven times as many people at high risk of hip fracture within two years. 

“With this type of clinical decision support, healthcare providers could direct preventive measures to the right people at the right time. These measures may include bone density testing, fall prevention interventions, or medication for osteoporosis,” says Mattias Lorentzon. 

The next step is to evaluate the model in additional countries and determine how it can best be implemented in clinical practice.