HUMAN BIOLOGICAL AGE: REGRESSION AND NEURAL NETWORK MODELS

Abstract

The purpose of this study is to develop methods for determining the biological age (BA) and the pace of human aging based on anthropometric and biochemical laboratory biomarkers, comparing the accuracy of BA determination using regression and neural network analysis. In 735 practically healthy people aged from 20 to 79, we determined the blood plasma concentrations of glucose and insulin and the blood serum concentrations of total cholesterol, cholesterol of high, low and very-low density lipoproteins, triglycerides, urea, creatinine, transaminases, and alkaline phosphatase. Also, we conducted anthropometric measurements and a standard oral glucose tolerance test and calculated the HOMA index. Age recognition was carried out using regression and neural network analysis. The multiple regression equation, which connects the examinees age with their biochemical laboratory parameters, allows to calculate the metabolic age of a person with an absolute error of 6.92 years. This accuracy is sufficient to reveal people at risk of accelerated aging. The use of a neural network algorithm with deep learning allows to determine the metabolic age with an error of 4.57 years, which is sufficient to distinguish between physiological and accelerated aging. The use of a neural network algorithm with deep learning increases the accuracy of determining a person’s metabolic age, the error of its determination is reduced by 40%.