Methodological Clarification and Analysis of Demographic and Anthropometric Determinants in the Calculation of REMS Bone Mineral Density.

Abstract

Radiofrequency echographic multi spectrometry (REMS) derives bone mineral density (BMD) using age- and anthropometry-dependent spectral models and conversion equations. The aim of this paper was to quantify the fraction of REMS BMD that can be explained by demographic and anthropometric variables. We analyzed 8000 proximal-femur and 8000 lumbar-spine REMS scans from Caucasian women, with independent development pools and fixed test sets of 4000 scans each and no significant development-versus-test differences in age, anthropometry, or REMS BMD. In 100 repeated runs, linear regression equations were fitted on random training subsets of 100-400 development scans. The primary anthropometric model equation included age and weight only. A further hypothetical maximum-explainability expanded model additionally included height, squared terms, and pairwise interactions, to explore the upper bound of explainability obtainable from anthropometric inputs alone. At train size 400, primary-model train/test R² values were 0.77/0.58 at the lumbar spine, 0.83/0.57 at the total hip, and 0.93/0.70 at the femoral neck; corresponding expanded-model values were 0.84/0.66, 0.89/0.69, and 0.95/0.75, respectively. Thus, although an association between REMS BMD and demographic/anthropometric variables was expected by design and resulted relevant even under independent testing, the primary model left 30-43% of REMS BMD unexplained by age and weight. The expanded model increased test R², but still left a substantial fraction of REMS BMD unexplained by anthropometric inputs. The highest R² values were consistently observed only in training data. Additional sensitivity analyses using alternative anthropometric models yielded similar findings. In conclusion, anthropometric explainability of REMS BMD was strongly sample-dependent and systematically much lower under independent testing than suggested by same-cohort analyses.