Age-related macular degeneration (AMD) is a major cause of irreversible central vision loss in older adults, and its disease burden continues to increase with population aging. AMD is heterogeneous and progresses through multiple stages, from drusen deposition and retinal pigment epithelium dysfunction in early disease to neovascular or atrophic changes in advanced disease, ultimately impairing independence and quality of life. Therefore, accessible systemic markers for early identification and risk stratification are needed. Bone mineral density (BMD) reflects not only skeletal strength and osteoporosis risk, but also broader biological aging processes linked to endocrine, metabolic, immune, and inflammatory regulation. Emerging evidence has suggested a possible “bone–eye axis,” in which skeletal health and retinal degeneration may share mechanisms such as chronic low-grade inflammation, oxidative stress, mitochondrial dysfunction, microvascular changes, lipid metabolic disturbance, and extracellular matrix remodeling. “However, previous epidemiological evidence linking BMD and AMD remains inconsistent, often limited by small sample sizes, cross-sectional designs, or single-population analyses, leaving causality, generalizability, and molecular mechanisms unclear.” said the author Xuehao Cui, a researcher at University of Cambridge, “This study therefore aimed to systematically evaluate the association between BMD and AMD risk, as well as its potential biological basis, by integrating multi-cohort epidemiological analyses, Mendelian randomization, multi-omics profiling, machine learning, and a low-BMD animal model.”
This study established a multinational and multilayered analytical framework to evaluate the relationship between bone mineral density (BMD) and age-related macular degeneration (AMD) risk. Three datasets were included: UK Biobank as a prospective cohort, NHANES as an imaging-based cross-sectional survey, and a Tianjin hospital-based case-control cohort, providing complementary evidence for the BMD–AMD association. For statistical analyses, Cox regression was used in UK Biobank to assess incident AMD risk, while multivariable logistic regression was used in NHANES and Tianjin. Restricted cubic splines, generalized additive models, and subgroup analyses were applied to examine nonlinear patterns and population differences. For prediction, Boruta and LASSO were used for feature selection, and regularized logistic regression, support vector machine, random forest, and XGBoost models were compared. Model performance and BMD contribution were evaluated using AUC, calibration, and SHAP interpretation. To explore causal and mechanistic evidence, the study further performed two-sample Mendelian randomization, multivariable MR, and two-step mediation MR, together with UK Biobank proteomic and metabolomic analyses to identify shared molecular signatures and pathways. Finally, a glucocorticoid-induced low-BMD rat model was used, with OCT, retinal vascular analysis, and Morris water maze testing to observe retinal structural, vascular, and visual-spatial changes under low-bone-mass conditions.
The results showed a consistent association between lower bone mineral density (BMD) and higher age-related macular degeneration (AMD) risk across three cohorts. In UK Biobank, NHANES, and the Tianjin cohort, participants with AMD were generally older, had lower BMD, and showed more vascular, metabolic, and inflammatory risk features. Multivariable models further indicated that higher BMD was associated with lower AMD risk, while quartile-based and nonlinear analyses suggested a protective trend at higher BMD levels. In predictive modeling, age and BMD were repeatedly identified as important contributors by Boruta, LASSO, and multiple machine learning models. XGBoost performed best in UKB, logistic regression in NHANES, and random forest in the Tianjin cohort, although the authors emphasized that these findings should be interpreted as cohort-specific predictive performance, and the incremental clinical value of BMD still requires further validation. Mendelian randomization provided supportive genetic evidence that higher BMD may contribute to lower AMD risk and prioritized GZMA, NELL1, and COL2A1 as candidate molecular intermediates. Proteomic and metabolomic analyses suggested shared molecular signatures involving extracellular matrix remodeling, lipid transport, amino acid metabolism, and inflammatory pathways. In the low-BMD rat model, the researchers also observed outer retinal thinning, retinal vessel narrowing, and delayed visual-spatial performance, but these changes only support retinal degeneration-related alterations under low-bone-mass conditions and should not be interpreted as direct validation of human AMD pathology.
The significance of this study lies in revealing a systemic link between bone mineral density (BMD) and age-related macular degeneration (AMD) risk from the perspective of the “bone–eye axis.” By integrating UK Biobank, NHANES, and the Tianjin cohort, the study found that lower BMD was consistently associated with higher AMD risk across different populations and measurement strategies. Mendelian randomization, multi-omics analyses, and animal experiments further provided supportive evidence, suggesting that skeletal aging and retinal vulnerability may share pathways involving extracellular matrix remodeling, lipid transport, amino acid metabolism, and inflammation. Overall, BMD does not replace canonical AMD mechanisms such as complement activation, RPE lipid handling, and Bruch membrane changes, but may serve as an accessible marker of systemic biological aging and retinal susceptibility. However, this study cannot establish direct causality, and AMD ascertainment and BMD measurement differed across cohorts. The machine learning results also mainly reflect cohort-specific predictive contributions. “In the future, we will further validate the incremental clinical value of BMD in AMD risk stratification through standardized AMD typing, unified BMD measurement, long-term fundus imaging follow-up, eye tissue-specific molecular data, and larger scale experimental models, and elucidate the temporal sequence and biological mechanisms between bone aging and retinal degeneration.” said Xuehao Cui.
Authors of the paper include Xuehao Cui, Qiuchen Zhao, Jingwen Hui, Zheya Han, Patrick Yu-Wai-Man, and Quanhong Han.
This study was supported by Tianjin Key Medical Discipline Construction (No. TJYXZDXK-3-004A-3), Nankai University Eye Institute Key Science and Technology Fund (No. NKSGZ202305), and Nankai University Eye Institute Open Fund Incubation Project (No. NKSGP202504).
The paper, “Decoding the Bone-Eye Axis: Machine Learning for Age Related Macular Degeneration Risk Prediction” was published in the journal Cyborg and Bionic Systems on Sept 14, 2026, at https://doi.org/10.34133/cbsystems.0676.