Enhanced osteoporotic fracture prediction in postmenopausal women using Bayesian optimization of machine learning models with genetic risk score
Authors' conclusion
This study accentuates the potential of combining genetic insights and optimized ML in strengthening fracture predictions, heralding new preventive strategies for postmenopausal women.
Study context
- Population
- Postmenopausal women from the Women's Health Initiative dataset; Genetic risk score (GRS) from 1,103 risk single nucleotide polymorphisms (SNPs) from genome-wide association studies (GWAS)
- Participants
- 25,772 women
- Comparison
- Machine learning (ML) models with GRS and FRAX clinical risk factors vs FRAX
- Outcome
- Prediction of major osteoporotic fractures (MOFs) and hip fractures (HFs): binary fracture outcome and 10-year fracture risk
- Follow-up
- 10-year follow-up
Methodological detailsmachine learning, Bayesian optimization, competing risk of death accounted for, sensitivity analysis including BMD, portability tests in non-European groups
- Machine learning
- Bayesian optimization
- Competing risk of death accounted for
- Sensitivity analysis including BMD
- Portability tests in non-European groups
Performance
As reported in the abstract
MOF binary prediction
| Test or model | Accuracy | AUC |
|---|---|---|
| XGBoost | 91.2% (95% CI: 90.4-92.0%) | 0.739 (95% CI: 0.731-0.746) |
XGBoost
Accuracy91.2% (95% CI: 90.4-92.0%)
AUC0.739 (95% CI: 0.731-0.746)
Further findings
| Outcome | XGBoost | Categorical net reclassification improvement (NRI) | P value |
|---|---|---|---|
| 10-year fracture risk modeling | 22.6% | P = 0.004 | |
| C-index | 0.795 (95% CI: 0.783-0.806) | ||
| Mean dynamic AUC | 0.799 (95% CI: 0.788-0.809) |
Sensitivity analysis including BMD but lacking GRS
- Reaffirmed the findings
Portability in non-European groups (Asians, African Americans)
- Model's robustness and adaptability underscored
Hyperparameter tuning
- Bayesian optimization displayed marked superiority over traditional grid search
Original publication
Enhanced osteoporotic fracture prediction in postmenopausal women using Bayesian optimization of machine learning models with genetic risk score. J Bone Miner Res. Published 4 February 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →