Ensemble-learning approach improves fracture prediction using genomic and phenotypic data.

Area of Science:

• Biomedical informatics
• Genomics
• Gerontology

Background:

• Current fracture risk models lack accuracy and fail to integrate genomic data effectively.
• Predicting major osteoporotic fractures (MOF) in older men requires advanced predictive capabilities.

Purpose of the Study:

• To develop and validate an ensemble machine learning (ML) model for enhanced MOF prediction in older men.
• To integrate diverse data types including clinical, lifestyle, skeletal, and genomic information.

Main Methods:

• Utilized data from 5130 participants in the Osteoporotic Fractures in Men cohort study.
• Developed a Super Learner (SL) ensemble model combining seven ML algorithms with 1103 genome-wide significant variants and conventional risk factors.
• Employed tenfold cross-validation and evaluated performance using AUC, accuracy, sensitivity, specificity, NPV, and PPV on a held-out testing set.

Main Results:

• The SL ensemble model achieved superior performance with an AUC of 0.76, accuracy of 95.6%, sensitivity of 94.5%, and specificity of 96.1%.
• The SL model outperformed individual ML models and baseline models across all performance metrics.
• Subgroup analyses showed the SL model maintained high accuracy in White (93.1%) and Minority (91.6%) populations.

Conclusions:

• Ensemble learning significantly enhances fracture prediction accuracy and model performance compared to individual ML approaches.
• Integrating genomic and phenotypic data through ensemble methods offers a promising strategy for personalized osteoporosis management.
• The developed SL model demonstrates potential for improving clinical decision-making in osteoporosis care.