Abstract
IMPORTANCE: Current fracture risk prediction tools, including the Fracture Risk Assessment Tool (FRAX), do not incorporate genetic risk factors limiting accuracy and contributing to misclassification and suboptimal care. OBJECTIVE: To develop and validate a novel genome-informed fracture risk assessment tool (Bayes-FRAX) integrating Bayesian genome-wide polygenic scores (GPS) into the established FRAX to enhance major osteoporotic fracture (MOF) prediction. DESIGN SETTING AND PARTICIPANTS: This retrospective cohort study analyzed clinical and genetic data from 6,932 postmenopausal women enrolled in the Women's Health Initiative (1993-1998). EXPOSURES: Integration of GPS derived using Polygenic Risk Score Continuous Shrinkage (PRS-CS) and Summary data-based Bayesian regression (SBayesR) into FRAX. MAIN OUTCOMES AND MEASURES: Primary outcomes included the incidence of MOF. Predictive performance metrics assessed were area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), calibration slopes, Hosmer-Lemeshow goodness-of-fit tests, net reclassification improvement (NRI), diagnostic sensitivity, decision curve analysis (DCA), and external validation metrics. RESULTS: Of 6,932 women, 513 (7.4%) experienced MOF. Bayes-FRAX significantly improved prediction over standard FRAX based solely on clinical risk factors, increasing AUROC from 0.662 to 0.680 for both PRS-CS and SBayesR. AUPRC improved from 0.120 (FRAX) to 0.140 (PRS-CS) and 0.138 (SBayesR). Calibration slopes were ideal (GPS-PRS-CS: 1.00 [95% CI: 0.8569-1.1431]; GPS-SBayesR: 1.00 [95% CI: 0.8560-1.1437]). Bayes-FRAX reclassified 3.5% of women, 34% near the intervention threshold. NRI improved by 4.59% (SBayesR) and 4.34% (PRS-CS), largely from better classification of women who fractured (5.85% and 5.65%). Decision curve analyses demonstrated greater net clinical benefit at clinically relevant thresholds, notably at the 20% threshold. External validation in 852 independent White postmenopausal women confirmed robust generalizability, with GPS significantly associated with fracture risk (PRS-CS OR = 0.148, 95% CI: 0.052-0.411; SBayesR OR = 0.116, 95% CI: 0.040-0.324). Likelihood ratio tests also supported improved model fit after GPS inclusion (PRS-CS: P < 0.001; SBayesR: P <0.001). Sensitivity analysis without BMD demonstrated stable AUROC (0.74). CONCLUSIONS AND RELEVANCE: Integrating GPS into FRAX using Bayesian methods improved fracture risk prediction, reclassification, and decision-making. Bayes-FRAX provides a generalizable tool for personalized osteoporosis care, especially for women near treatment thresholds.