Application of recursive partitioning method (RPM) to select the multi-frequency bioimpedance analysis (MF-BIA) raw parameters predicting appendicular skeletal muscle mass index (SMI)

Introduction: Anthropometric and BIA parameters are used as covariates in linear regression equation to estimate total body water and fat free mass. More recently, the research is focused on understanding the capacity of bioelectrical raw variables by itself to detect nutritional status. Objective of the study is to identify, among the multi-frequency BIA (MF-BIA) parameters, the best predictors of SMI (kg/m2) estimated by dual-energy X-ray absorptiometry (DXA). Methods: 148 women (age 45.6 ± 14.8 years; BMI 37.3 ± 6.7 kg/m2) have been enrolled at CASCO. Z, AP, Rx and Xc at 5, 10, 50, 100, 250 kHz frequencies (Human im Touch, DSMedica) and SMI (Hologic 4500 RDR) were measured according to the standardized procedures. A set of MF-BIA covariates was selected a priori (PA50, Z5, Z50, Z250) and the recursive partitioning method (RPM) was applied to identify the best predictors of SMI. The RPM was performed using the party package of the free statistical software R which provides significance level for multiple test procedure (p-value <0.05 was assumed as significant). Results: RPM selected at the first decision step Z250 as the covariate having the greatest association with SMI identifying a split at 446Ohm. In the subsequent decisions step the covariates selected were Z250 (373Ohm) and then PA50 (5.3°) in one branch of the decision tree and Z250 (499Ohm) in the other branch. The value Z250≤373 together with PA50>5.3° and Z250>499 Ohm identify respectively the subjects with higher and lower SMI values (p-value < 0.001). Conclusion: The study shows that multi-frequency raw data combined with each other can be used to predict SMI measured by DXA. We believe that this approach will allow to identify the cut-offs of the MF-BIA specific raw data useful to screening sarcopenia in various categories of subjects.