Integrating Machine Learning into Fracture Liaison Services: Toward precision risk stratification in osteoporosis care: A prospective cohort study.

BACKGROUND

Traditional fracture risk assessment tools have limitations in accurately predicting re-fracture risk. Machine learning (ML) approaches offer a novel method for improving risk prediction. This study aimed to apply ML approach to identify key predictors of mortality, re-fractures, and falls among patients enrolled in a Fracture Liaison Service (FLS) program.

METHODS

This prospective cohort study analyzed data from 600 patients aged ≥50 years with a history of hip or vertebral fractures who were enrolled in an FLS program between 2014 and 2016. The patients underwent comprehensive assessments, and ML models, specifically a balanced random forest classifier with recursive feature elimination with cross-validation, were used to identify the most influential predictors for three outcomes: two-year mortality, re-fractures, and falls. Model performance was evaluated based on accuracy, precision, sensitivity, specificity, F1-score, and area under the curve.

RESULTS

Over two years, 14.2% of patients died, 6% sustained new fractures, and 33.2% experienced at least one fall. The ML analysis identified age, body mass index, self-care ability, and serum calcium and alkaline phosphatase levels as the strongest predictors of mortality. Re-fracture risk was primarily influenced by 10-year major fracture probability, body height, and BMI. Falls were significantly associated with BMI, age, and serum alkaline phosphatase levels.

CONCLUSION

While the ML models demonstrated modest predictive discrimination for re-fractures, they can analyze complex data to identify hidden risk factors that may be overlooked by traditional methods. These insights highlight the potential of AI-driven models for osteoporosis management, enabling more precise interventions for high-risk individuals.