A Comprehensive Study on the Impact of Hypertension on Bone Metabolism Abnormalities Based on NHANES Data and Machine Learning Algorithms

Abstract

Background: Hypertension (HTN), a globally prevalent chronic condition, poses a significant public health challenge. Concurrently, abnormalities in bone metabolism, such as reduced bone mineral density (BMD) and osteoporosis (OP), profoundly affect the quality of life of affected individuals. This study aims to comprehensively investigate the relationship between HTN and bone metabolism abnormalities using data from the National Health and Nutrition Examination Survey (NHANES) and advanced machine learning techniques. Methods: Data were sourced from the NHANES database, covering the years 2009 to 2018. Specifically, femur and spine BMD measurements were obtained via dual-energy X-ray absorptiometry (DXA) for the 2009-2010 period, given the lack of full-body data. A predictive model was developed to estimate total body BMD from femur and spine measurements. The initial dataset comprised 49,693 individuals, and after rigorous data cleaning and exclusion of incomplete records, 7,566 participants were included in the final analysis. Data were processed and analyzed using SPSS, which facilitated descriptive statistical analysis, multivariate logistic regression, and multiple linear regression, alongside subgroup analyses to explore associations across different demographic groups. Machine learning algorithms, including neural networks, decision trees, random forests, and XGBoost, were utilized for cross-validation and hyperparameter optimization. The contribution of each feature to the model output was assessed using SHAP (Shapley Additive Explanations) values, enhancing the model's accuracy and robustness. Results: Baseline characteristic analysis revealed that compared to the non-HTN group, the HTN group was significantly older (44.37 vs. 34.94 years, p < 0.001), had a higher proportion of males (76.8% vs. 60.7%, p < 0.001), higher BMI (31.21 vs. 27.77, p < 0.001), a higher smoking rate (54.4% vs. 41.2%, p < 0.001), and notably lower BMD (1.1507 vs. 1.1271, p < 0.001). When comparing the low bone mass group with the normal bone mass group, the former was older (36.02 vs. 34.5 years, p < 0.001), had a lower proportion of males (41.8% vs. 63.3%, p < 0.001), lower BMI (25.28 vs. 28.25, p < 0.001), and a higher incidence of HTN (10.9% vs. 8.6%, p = 0.006). Overall logistic and multiple linear regression analyses demonstrated a significant negative correlation between HTN and bone metabolism abnormalities (adjusted model Beta = -0.007, 95% CI: -0.013 to -0.002, p = 0.006). Subgroup analysis revealed a more pronounced association in males (Beta = -0.01, p = 0.004) and in the 40-59 age group (Beta = -0.01, p = 0.012). The machine learning models corroborated these findings, with SHAP value analysis consistently indicating a negative impact of HTN on BMD across various feature controls, thus demonstrating high explanatory power and robustness across different models. Conclusion: This study comprehensively confirms the significant association between HTN and bone metabolism abnormalities, utilizing NHANES data in conjunction with machine learning algorithms.