Derui AI Pharma Files for IPO Following Breakthrough in Machine Learning-Based Biological Age Assessment Tailored for Chinese Population

Recently, the MindRank team, in collaboration with Zhejiang University and Duke Kunshan University, published in "Frontiers in Medicine" (Impact Factor5.1) Jointly published paper. This study used machine learning (ML) methods to establish a biological age measurement method based on the middle-aged and elderly population in China, and demonstrated a close relationship between machine learning-based biological age and disease incidence as well as mortality. It also found that the machine learning-based biological age has advantages over previous methods. The MindRank team was responsible for the modeling and data analysis work in this study.

01 Research Background

According to the results of the seventh national census in China, the population aged 60 and above has reached 260 million, accounting for more than 18% of the total population [1]. With the intensification of aging comes the increase in the incidence of various diseases. By the end of 2018, 75% of the elderly in China suffered from at least one chronic disease, with nearly 190 million patients [2]. Therefore, formulating measures for aging, such as early identification of high-risk disease patients, is crucial for addressing the issue of aging in China.

An important indicator for assessing the risk of disease is the degree of an individual's aging, biological age (BA), which refers to the age calculated based on the physiological and anatomical developmental state of a normal body. It is considered a more accurate aging indicator than an individual’s chronological age. In recent years, the role of machine learning (ML) in evaluating BA has gained increasing recognition. However, most studies have focused on Western populations, lacking research on ML-based BA assessments specific to the Chinese population. Therefore, further validation is needed to assess the accuracy of ML-BA in evaluating individual aging levels and its application value in the Chinese population.

02 Establishment of Biological Age Based on Machine Learning

This latest research, brought by MindRank and the university scientific research team, aims to apply several ML algorithms to develop new and more accurate measures of aging, such as Gradient Boosting Regression Trees, Gradient Boosting Machines, Random Forests, CatBoost, Support Vector Machines, and AdaBoost Regressors. The data used in the study originated from a survey that began in 2011—the China Health and Retirement Longitudinal Study—covering 28 provinces, 150 counties/districts, and 450 villages/urban communities across China. Participants were followed up once every two years, receiving three follow-ups over a six-year period, reporting their Basic Activities of Daily Living (BADL), Instrumental Activities of Daily Living (IADL), and upper and lower limb functional status.

After screening, this new study analyzed data from 9,771 participants aged between 45 and 85 years. The average age of the participants was 59.1 years (SD = 9.2), with the average age for males and females being 59.8 (SD = 9.1) years and 58.5 (SD = 9.2) years, respectively. Among them, 44.6% of the participants were aged 60 years or older, and 53.5% were female. The researchers constructed a binary variable to assess the mortality rate of participants over a six-year follow-up period from baseline.

The research team based on 19 itemsBiomarkerUsing the methods in Table 1 for training, it was found that the gradient boosting regression tree performed the best (Table 1). Based on this model, the participants' ML-BA was calculated.

03 Validation of Biological Age Based on Machine Learning

To validate ML-BA, the research team used regression models to determine the association between ML-BA and disease incidence and mortality. The team also established a classical BA based on the Klemera-Doubal (KD) method and compared its strength of association with outcomes to that of ML-BA.

In terms of the association with health, both ML-BA and KDM-BA were significantly associated with the rate of physical disability among participants within a 4-year follow-up period, although the association was stronger for ML-BA. In models without adjustment for any covariates, each additional year of ML-BA in participants was associated with a 6% increase in the likelihood of BADL and IADL impairments, a 4% increase in the likelihood of upper limb mobility impairment, and a 7% increase in the likelihood of lower limb mobility impairment (Table 2). Regarding the correlation with mortality within a 6-year follow-up period, both ML-BA and KDM-BA were positively correlated with the risk of death, with each additional year increasing the mortality risk by 16% and 10%, respectively.

After adjusting for age covariates in the model, ML-BA and KDM-BA remained significantly associated with the rate of physical disability within 4 years and mortality risk within 6 years of follow-up (Table 3). For each one-year increase in ML-BA and KDM-BA, participants' mortality risk increased by 7% (OR=1.07, 95%CI=1.05, 1.09) and 5% (OR=1.05, 95%CI=1.04, 1.07), respectively.

04 Conclusion

This study established the first ML-BA based on the Chinese population and demonstrated that this indicator can be used to predict disability and mortality in the elderly. This conclusion further supports the potential of ML-calculated BA as a biomarker of biological aging, as well as its role in helping to classify health risks among the general elderly population in China. In the future, ML-BA could serve as an indicator for evaluating healthy aging, providing important reference for addressing aging issues in China and other countries. The study was funded by the Kunshan Municipal Government.BioValleyBioon.com）