Artificial Intelligence Approaches for Osteoporotic Fracture Risk Prediction Using Administrative Health Data: A Systematic Review
Clinical Snapshot
PICO Framework
| P — Population | Adults at risk of osteoporotic fractures |
| I — Intervention | Machine learning (ML) models developed or validated using administrative health data alone (no clinical measurements) |
| C — Comparator | Implicit comparison across ML model types (Random Forest, XGBoost, LASSO-regularised Logistic Regression, Neural Networks); no head-to-head comparator stated |
| O — Outcomes | Discriminative performance (AUC), risk of bias (PROBAST), applicability concerns, external validation status, and clinical utility evaluation |
Bottom Line
This systematic review identified only seven studies examining ML models for osteoporotic fracture risk prediction built exclusively from administrative health data — a clinically important but methodologically immature field. Discriminative performance was moderate to good (best AUC 0.818–0.905), with the highest values reported for hip fracture prediction. However, the evidence base is too limited to support clinical adoption: only two studies used external validation, none evaluated clinical utility (e.g., via decision curve analysis), and five of seven raised applicability concerns due to population specificity or database-dependent features. The review is methodologically sound — PROBAST was appropriately applied and PRISMA guidelines followed — but the absence of pooled estimates, confidence intervals, and sensitivity analyses limits precision. For Australian clinicians, FRAX with or without DXA remains the recommended standard for fracture risk assessment per RACGP guidelines. Administrative ML tools represent a promising avenue for scalable, automated population screening, particularly leveraging linked MBS/PBS datasets, but require rigorous external validation and clinical utility assessment before any practice change is warranted. Future research should prioritise recalibration and external validation of existing models rather than developing further novel architectures.
Key Findings
P Value: Not reported
Effect Size: Best AUC ranged from 0.818 (osteoporotic fractures in general) to 0.905 (hip fractures only) across included studies
Primary Outcome: Discriminative performance of ML models for osteoporotic fracture risk prediction using administrative data only
Nnt Or Sensitivity: No NNT, sensitivity, specificity, or calibration metrics (e.g., Brier score, calibration slope) reported in abstract; AUC is the sole discriminative metric described
Confidence Interval: Not reported in abstract; individual study CIs not synthesised
Clinical Application
Conceptually feasible as a scalable, automated screening tool requiring no additional clinical input beyond routinely collected administrative data. However, current evidence does not support clinical deployment: only two studies used external validation, none demonstrated clinical utility, and five raised applicability concerns. Integration into electronic health record systems would require substantial further validation work. Australia possesses high-quality linked administrative datasets (MBS, PBS, AIHW hospital admissions, APDC) that are well-suited to this research paradigm. The PBS records bisphosphonate and denosumab prescriptions, and MBS data captures DXA scanning — both potentially useful ML features. However, no included studies appear to use Australian data. RACGP guidelines currently recommend FRAX with or without BMD for fracture risk assessment; ML administrative models are not yet endorsed. TGA has not approved any ML fracture risk tool. Osteoporosis Australia and the RACGP's 'Osteoporosis Prevention, Diagnosis and Management' guidelines would need to be updated before any such tool could be recommended in Australian primary care. The Australian Commission on Safety and Quality in Health Care's focus on preventive care and the National Osteoporosis Strategy provide a policy context in which validated administrative ML tools could eventually play a role in population-level case-finding. Adults at elevated risk of osteoporotic fractures who are identifiable through administrative health records; particularly relevant for population-level screening programmes where clinical measurements (BMD, FRAX inputs) are unavailable
Abstract
To systematically review Machine Learning (ML) fracture risk prediction models developed solely using administrative data, evaluating their development, performance, risk of bias, and concerns regarding applicability. A systematic search was conducted in PubMed, Embase, IEEE, and Web of Science following PRISMA guidelines (up to November 2025). We included studies developing or validating ML models for osteoporotic fracture risk in adults using administrative data without clinical measurements. Risk of bias and concerns for applicability were assessed using the PROBAST tool. Seven studies were included from 3435 initial records. A range of of ML models were utilized, including Random Forests, XGBoost, LASSO-regularized Logistic Regression and Neural Networks. Discriminative performance was moderate-to-good, with best Area Under the Curve (AUC) values ranging from 0.818 for osteoporotic fractures in general to 0.905 for hip fractures only. While five studies showed low risk of bias, five raised applicability concerns due to using specific subpopulations or features with database characteristics. Only two studies used external validation and none directly evaluated clinical utility of the models. ML models using solely administrative data show some promise as scalable, automated tools for fracture risk prediction without requiring manual clinical input. However, clinical implementation is hindered by limited external validation and a lack of formal utility evaluation of the models. Future research should prioritize more rigorous external validation and recalibration of the existing models, more so than the development of novel models, ensuring they are robust, interpretable, and integrable into future clinical workflows.
References
- 1.Hansen, B. B., Leth, K. W., Hansen, N. R., Abrahamsen, B., Fuggle, N. R., Brønd, J. C., & Rubin, K. H. (2026). Artificial intelligence approaches for osteoporotic fracture risk prediction using administrative health data: A systematic review. Calcified Tissue International. Advance online publication. PubMed ID: 42350824
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service