# Advanced Bayesian BMD-Derived Genome-Wide Polygenic Scores Enhance Clinical FRAX-Based Fracture Risk Prediction in Postmenopausal Women

## TL;DR

Adding genetic data to a fracture risk tool improves predictions for postmenopausal women, especially those near treatment thresholds.

## Contribution

Integrating Bayesian-derived polygenic scores into FRAX enhances fracture risk prediction and clinical reclassification.

## Key findings

- Bayesian GPS-FRAX improved discrimination with a time-dependent AUC increase from 0.72 to 0.74.
- Approximately one-third of reclassified individuals had baseline FRAX scores within 15–25%, a zone of clinical uncertainty.

## Abstract

Current fracture prediction tools, including the Fracture Risk Assessment Tool (FRAX), rely primarily on clinical risk factors and bone mineral density measurements, yet omit genetic information. Given osteoporosis's substantial heritability (estimated at 50–80%), excluding genetic data may lead to clinical misclassification and uncertainty, particularly among postmenopausal women near clinical intervention thresholds. This study aimed to enhance clinical fracture risk prediction by integrating advanced Bayesian-derived genome-wide polygenic scores (GPS) into FRAX (Bayesian GPS-FRAX). Using genetic data from the Genetic Factors for Osteoporosis Consortium (GEFOS), we derived GPS using two advanced Bayesian methods, Polygenic Risk Score Continuous Shrinkage (PRS-CS) and Summary-data-based Bayesian Regression (SBayesR), selected for their ability to model complex genetic architectures and linkage disequilibrium. Bayesian GPS was integrated into FRAX, creating the Bayesian GPS-FRAX model, which was subsequently validated in 6932 postmenopausal women from the Women’s Health Initiative. We evaluated model performance in terms of both discrimination and clinical utility, with clinical utility primarily assessed using net reclassification improvement (NRI), threshold-based clinical reclassification, and decision curve analysis. Robustness and generalizability were confirmed through independent validation in a separate cohort of 3688 women. Bayesian GPS-FRAX modestly improved discrimination, increasing the time-dependent AUC from 0.72 to 0.74, and yielded clinically meaningful gains in risk classification, with a NRI of 4.55–5.07% compared with FRAX-CRF. Approximately one-third of reclassified individuals had baseline FRAX scores within 15–25%, a critical zone of clinical uncertainty. Bayesian GPS-FRAX successfully reclassified these women, correcting misclassification and enabling more accurate and timely clinical interventions. The greatest reclassification benefit was evident among women older than 70. Decision curve analyses demonstrated consistent net clinical benefit across clinically relevant thresholds (15–25%), notably at the widely used 20% treatment threshold. Independent validation further confirmed robust generalizability and clinical applicability. Integrating Bayesian GPS into FRAX substantially enhances discrimination and clinical fracture risk stratification, especially in older women and individuals near critical intervention thresholds. By addressing genomic complexity and clinical uncertainty through advanced Bayesian methodologies, Bayesian GPS-FRAX represents a meaningful advancement toward personalized osteoporosis management. Prospective validation in diverse cohorts, along with assessments of clinical implementation and cost-effectiveness, will be essential to facilitate the integration of this precision-based approach into routine clinical practice.

The online version contains supplementary material available at 10.1007/s00223-026-01497-8.