# A neural network approach to sarcopenia prediction based on bioelectrical impedance in community-dwelling older adults

## TL;DR

This study shows that a neural network can accurately predict sarcopenia using raw bioelectrical impedance data from older adults.

## Contribution

A novel neural network model is proposed for sarcopenia prediction using raw bioelectrical impedance data.

## Key findings

- A model using 5, 50, and 250 kHz frequencies achieved high accuracy (AUC: 0.96) in predicting sarcopenia.
- A model using 50 kHz frequency showed strong performance (AUC: 0.88) in an alternative dataset.
- High sensitivity and negative predictive value suggest the model's potential for sarcopenia screening.

## Abstract

This study aimed to apply a neural network to raw bioelectrical impedance analysis data and to test whether sarcopenia could be predicted with high accuracy. The study population comprised 727 community-dwelling older adults aged 65–85 years who participated in the Kasama Study from 2015 to 2018. Sarcopenia was determined using the standard values set by the Asian Working Group for Sarcopenia 2019. Skeletal muscle mass index, grip strength, and five-times sit-to-stand test (Dataset 1) or skeletal muscle mass index, grip strength, and gait speed (Dataset 2) were used. The characteristic variables were sex, age, height, and body mass index, as well as parameters from bioelectrical impedance analysis, such as reactance, resistance, and impedance for six frequencies (1, 5, 50, 250, 500, and 1000 kHz) in six body parts measured using a multi-frequency body composition analyzer (MC-980A, Tanita). For analysis, a neural network was used to construct a model. For verification of the model’s accuracy, a receiver operating characteristic analysis was performed to calculate the sensitivity, specificity, area under the curve, and positive and negative predictive values. Among the participants analyzed, 21 (3.3%) in Dataset 1 and 24 (3.7%) in Dataset 2 had sarcopenia. In Dataset 1, the model that used 5, 50, and 250 kHz showed the highest prediction accuracy (sensitivity: 1.00, specificity: 0.91, area under the curve: 0.96, accuracy: 0.91, positive predictive value: 0.28, negative predictive value: 1.00). In Dataset 2, the model that used 50 kHz exhibited the highest prediction accuracy (sensitivity: 0.91, specificity: 0.84, area under the curve: 0.88, accuracy: 0.84, positive predictive value: 0.17, negative predictive value: 1.00). In conclusion, highly accurate predictions are possible by applying a neural network to the raw data obtained from bioelectrical impedance analysis. As a highly accurate sarcopenia screening method, it is expected to be used in various settings, ranging from clinical practice to local communities.