Development and validation of common data model-based fracture prediction model using machine learning algorithm.

Area of Science:

• Gerontology
• Epidemiology
• Medical Informatics

Background:

• Osteoporotic fractures pose a significant health and economic burden globally.
• There is a growing need for accurate, real-world fracture prediction models tailored to specific populations.

Purpose of the Study:

• To develop and validate a user-friendly model for predicting major osteoporotic and hip fractures.
• To create accurate, country-specific fracture prediction scoring systems using real-world data.

Main Methods:

• Utilized data from 20,107 (discovery) and 13,353 (validation) participants aged ≥50 years from a common data model database (2008-2011).
• Employed DeepHit and Cox proportional hazard models to identify fracture predictors and construct scoring systems.
• Assessed major osteoporotic and hip fracture events over a mean follow-up of 7.6 years.

Main Results:

• The final models included predictors such as fracture history, age, lumbar spine T-score, total hip T-score, cardiovascular disease, cerebrovascular disease, and diabetes mellitus.
• Harrell's C-index values ranged from 0.762 to 0.860, indicating good predictive accuracy.
• Estimated 10-year risks for major osteoporotic fractures ranged from 2.0% to 68.8%, and for hip fractures from 0.2% to 18.8%.

Conclusions:

• Developed and validated simple scoring systems for osteoporotic fractures using hospital-based cohorts.
• These models demonstrated good predictive performance in an independent Korean cohort.
• The scoring systems offer a practical tool for predicting fracture risk in clinical practice.