A clinical decision support tool that uses artificial intelligence has shown success in predicting hip fracture in adults, serving as a potential resource-efficient solution for population screening, according to a nationwide cohort study.
Fracture Risk Assessment and Classification Using Real-world Evidence and Machine Learning (FRACTURE-ML) demonstrates high accuracy that persists over longer follow-up and after accounting for competing mortality. [PLoS Med 2026;23:e1005190]
“Implementing FRACTURE-ML could substantially improve efficient identification of high-risk patients, enabling targeted treatment and reducing hip fracture rates,” the investigators said.
The study followed more than 3.5 million individuals aged ≥50 years living in Sweden at baseline (set between 2011 and 2013) who had not been prescribed osteoporosis medication within the last 2 years through the end of 2021. Of these, 142,327 sustained a hip fracture during follow-up.
The investigators defined a total of 139,980 variables comprising diagnoses, medications, procedures, demographics, and socioeconomic data with multiple historic windows and level of detail using a broad unconditional approach. They then divided the dataset into discovery (25 percent), development (65 percent), and holdout (10 percent) cohorts.
Finally, traditional Cox models and ML methods, XGBoost and DeepSurv, were used to examine the risk of fracture among participants.
Using 2,500 predictors, FRACTURE-ML delivered an area under the curve of 0.89 (95 percent confidence interval [CI], 0.88‒0.89) at 1 year and 0.88 (95 percent CI, 0.87‒0.88) at 2 years. A reduced model using 35 predictors generated comparable AUCs of 0.87 at 2 years and 0.85 at 5 years. Traditional Cox models with predictors ranging from 35 to 400 also yielded similar AUCs.
“Both the DeepSurv and the Cox models performed excellently at the individual level based on calibration plot analysis,” the investigators said.
Screening-based approach
Furthermore, screening using the fracture liaison services (FLS) secondary prevention approach (recent fracture) showed an AUC of 0.55 (95 percent CI, 0.54‒0.55) at 2 years.
Two-year prediction with FRACTURE-ML, which could be used as a complementary approach for primary prevention, detected nearly seven times more individuals at risk of hip fracture compared with the FLS approach (sensitivity 0.84 vs 0.12), with limited reduction in specificity (0.79 vs 0.98).
“FRACTURE-ML could be introduced for the primary prevention of hip fracture through a screening-based approach,” the investigators said. “One potential implementation would be to deploy the model via a national service integrated with registry data needed for FRACTURE-ML calculations.”
Patients with appropriate consent could avail of their FRACTURE-ML risk score through a healthcare provider, a health service mobile application, or an online form.
Those with an increased risk could then receive further examination in line with clinical practice, including evaluation of physical function and balance, as well as bone mineral density of the hip and lumbar spine via dual x-ray absorptiometry. [Osteoporos Int 2022;33:1631-1637]
“These assessments would inform appropriate interventions, such as fall prevention strategies and osteoporosis treatment,” the investigators said.
These findings are limited by a lack of external validation and implementation studies, which are needed to establish the clinical usefulness of FRACTURE-ML, according to the investigators.
Hip fractures may result in significant morbidity, disability, mortality and substantial healthcare costs. [Climacteric 2022;25:4-10; Arch Intern Med 2002;162:2053-2057; Inj Prev 2020;26(supp 1):i3-11; BMC Geriatr 2016;16:158]
“Therefore, it is of utmost importance to identify individuals at high risk of hip fracture in order to initiate effective preventive measures, such as treatment with osteoporosis medications and fall prevention,” the investigators said. [J Intern Med 2019;285:381-394]