Swedish researchers develop AI tool to predict hip fracture risk years in advance

Hip fractures carry a weight that extends far beyond the initial break. They often trigger a cascade of complications—loss of independence, chronic pain, diminished mobility—and in many cases, they accelerate decline toward death. Yet many of these fractures are preventable if the people most vulnerable to them can be identified before the break happens. Researchers at the University of Gothenburg have built a tool designed to do exactly that, using data that already exists in Sweden's health system.

The tool, called FRACTURE-ML, works without requiring patients to visit clinics or fill out questionnaires. Instead, it draws on information already recorded in Sweden's national health registers: diagnoses, prescription medications, medical procedures, and basic demographic and socioeconomic details. This approach sidesteps one of the major barriers to early screening—the burden of asking patients to come in for assessments. The model can flag people who should be referred for further evaluation, serving as a first filter in the system.

The research, published in PLOS Medicine, tested the approach on more than 3.5 million Swedes aged 50 and older. During the follow-up period, 142,000 of them sustained a hip fracture. The researchers developed and compared several statistical models, including modern machine learning approaches, before settling on a final version that could generate reliable risk estimates at one, two, five, and ten-year intervals. The specificity matters: knowing someone's risk not just in abstract terms but at concrete time horizons allows clinicians to time interventions more precisely.

Mattias Lorentzon, a professor of geriatric medicine at Sahlgrenska Academy, frames the stakes plainly. Hip fractures cause significant suffering and loss of independence, he notes, but many are preventable if high-risk people are caught early. The model, he says, is particularly good at separating those at high risk from those at low risk, and it has the potential to become a cornerstone of preventive care.

The current system identifies people at high risk only after they have already broken a hip, typically through Fracture Liaison Services that activate after the fact. FRACTURE-ML inverts this logic. It can identify people at high risk before the fracture occurs. The numbers are striking: compared with current practice, the tool could identify nearly seven times as many high-risk individuals within a two-year window. That difference translates directly into opportunity—more people who could receive bone density testing, fall prevention interventions, or osteoporosis medication before a fracture happens.

Kristian Axelsson, the study's first author, emphasizes what makes the approach distinctive. Because it relies entirely on registry data that healthcare systems already maintain, it can be deployed at scale without adding administrative burden to clinics. No new questionnaires, no new appointments required. This creates a pathway for earlier intervention and more targeted prevention.

The research team is now moving toward the next phase: testing the model in other countries and determining how it can be integrated into clinical workflows. The tool exists; the question now is how to make it part of routine practice, and whether its predictive power holds up beyond Sweden's particular health system and population.