A machine-learning algorithm can predict the risk of hip fracture from routinely collected health data better than current screening without needing to see the patient, according to a study in than 3.5 million older individuals.
The Fracture Risk Assessment and Classification Using Real-world Evidence and Machine Learning (FRACTURE-ML) tool could help shift hip-fracture care from reacting after an injury to preventing it in the first place.
The AI prediction model, described in PLOS Medicine, identified around seven times more high-risk individuals than current clinical practice, while maintaining relatively high precision.
“FRACTURE-ML accurately identified people at high risk of hip fracture using routinely collected healthcare and population data, without requiring an in-person clinical assessment,” said corresponding author Mattias Lorentzon, MD, from the University of Gothenburg in Sweden.
“This could make large-scale screening more efficient and help preventive care reach people before a hip fracture occurs.”
Hip fractures are common in older adults and often have serious consequences in terms of independence, health and survival. Existing tools that predict fracture risk usually require information provided by patients, making large-scale screening difficult.
In search of a better way forward, researchers led by Kristian Axelsson, MD, also from the University of Gothenburg, analyzed Swedish nationwide registry data from 3,542,647 individuals aged at least 50 years who had not been prescribed osteoporosis medication in the two years previously.
During follow up of up to 10 years, 142,327 of the participants sustained a hip fracture.
The team then used 139,980 variables encompassing diagnoses, medications, and procedures as well as demographic and socioeconomic data to develop and validate a deep-learning hip fracture prediction model based entirely on national registry data without patient assessment.
FRACTURE-ML, developed using Deep-Surv with 2500 variables, was the best model and had excellent performance with a one-year area under the curve (AUC) of 0.89 which sustained accuracy at five years, with an AUC of 0.85.
It therefore represented the most accurate registry-based hip fracture model to date.
A model in which the number of variables was reduced to 35 performed nearly as well, with a one-year AUC of 0.88, suggesting that it could strongly predict fractures in a practical and interpretable clinic tool.
The researchers conclude: “Implementing FRACTURE-ML could substantially improve efficient identification of high-risk patients, enabling targeted treatment and reducing hip fracture rates. strong predictive performance.”
