Artificial Intelligence Approaches for Osteoporotic Fracture Risk Prediction Using Administrative Health Data: A...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Epidemiology

Background:

• Osteoporotic fractures pose a significant public health burden.
• Accurate fracture risk prediction is crucial for timely intervention.
• Administrative data offers a scalable resource for risk modeling.

Purpose of the Study:

• To systematically review machine learning (ML) models for osteoporotic fracture risk prediction using only administrative data.
• To evaluate model development, performance, risk of bias, and applicability concerns.
• To identify gaps for future research and clinical implementation.

Main Methods:

• Systematic literature search in PubMed, Embase, IEEE, and Web of Science (PRISMA guidelines).
• Inclusion of studies developing/validating ML models for adult osteoporotic fracture risk using administrative data.
• Risk of bias and applicability assessment using the PROBAST tool.

Main Results:

• Seven studies were included, utilizing various ML models (Random Forests, XGBoost, etc.).
• Moderate-to-good discriminative performance observed (AUC 0.818-0.905).
• Applicability concerns noted due to specific subpopulations/database features; limited external validation (2 studies) and no clinical utility evaluation.

Conclusions:

• ML models with administrative data show potential for automated, scalable fracture risk prediction.
• Clinical implementation requires enhanced external validation and formal utility assessment.
• Future research should focus on model recalibration and robustness over novel model development.