Title: Causal Analysis of Backdoor-Adjusted Skeletal Phenotypes from DXA and Their Impact on Hip Fracture Risk

Abstract

Purpose: This study aimed to evaluate the association between dual-energy X-ray absorptiometry (DXA)-derived hip skeletal phenotypes and hip fracture risk, utilizing prespecified confounder adjustment. Additionally, it sought to determine whether ranking these phenotypes based on their backdoor-adjusted average treatment effects (ATEs) could enhance risk stratification capabilities.

Methods: The analysis included 21,098 participants from the UK Biobank who possessed linked health records, hip DXA-derived skeletal metrics, and prespecified covariates. Sixteen distinct phenotypes were assessed, covering bone mineral content (BMC), bone mineral density (BMD), and T-scores across various hip-related anatomical regions. The selection of confounders was directed by a prespecified directed acyclic graph (DAG). Backdoor-adjusted ATEs were calculated on the absolute risk-difference scale for each standard deviation (SD) increase in phenotype value. The study further examined effect heterogeneity regarding total femur BMD and evaluated downstream prediction performance by combining clinical variables with phenotypes ordered by ATE magnitude.

Results: Of the 21,098 participants, 115 experienced hip fractures. Every one of the 16 phenotypes exhibited negative backdoor-adjusted ATEs for every SD increase. The most substantial ATEs were recorded for total femur BMC and total femur BMD, both showing a risk difference of -0.0047. This equates to approximately 4.7 fewer hip fractures per 1,000 participants for every SD increase in the respective phenotype value. Analysis of conditional effects indicated that the impact of total femur BMD was more pronounced in older individuals and those with lower body mass index (BMI). In terms of predictive performance, a model integrating clinical variables with the top 11 ATE-ranked phenotypes outperformed FRAX using femoral neck BMD, achieving a higher AUC (0.842 versus 0.709), greater sensitivity (0.748 versus 0.443), and comparable specificity (0.793 versus 0.777).

Conclusion: The backdoor-adjusted ATEs varied significantly among DXA-derived hip skeletal phenotypes. Conducting causal evaluations at the phenotype level may assist in pinpointing the most informative DXA measures for effective risk stratification.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC