Factors Associated With Trabecular Bone Score and Bone Mineral Density; A Machine Learning Approach

Abstract

Introduction: Bone indexes including trabecular bone score (TBS) and bone mineral density (BMD) have been shown to be associated with wide spectrum of variables including physical activity, vitamin D, liver enzymes, biochemical measurements, mental and sleep disorders, and quality of life. Here we aimed to determine the most important factors related to TBS and BMD in SUVINA dataset.

Methods: Data were extracted from the Survey of Ultraviolet Intake by Nutritional Approach (SUVINA study) including all 306 subjects entered this survey. All the available parameters in the SUVINA database were included the analysis. XGBoost modeler software was used to define the most important features associated with bone indexes including TBS and BMD in various sites.

Results: Applying XGBoost modeling for 4 bone indexes indicated that this algorithm could identify the most important variables in relation to bone indexes with an accuracy of 92%, 93%, 90% and 90% respectively for TBS T-score, lumbar Z-score, neck of femur Z-score and Radius Z-score. Serum vitamin D, pro-oxidant-oxidant balance (PAB) and physical activity level (PAL) were the most important factors related to bone indices in different sites of the body.

Conclusions: Our findings indicated that XGBoost could identify the most important variables with an accuracy of >90% for TBS and BMD. The most important features associated with bone indexes were serum vitamin D, PAB and PAL.

Keywords: Bone mineral density; Machine learning; Trabecular bone score; Vitamin D; XGBoost.