Abstract
For a person to continue a healthy life, it is important for him/her to become aware of the risk of illness as soon as possible and take appropriate countermeasures. To help people live a healthy life, NTT is researching and developing technology for predicting disease risk by using machine learning and analyzing the risk factors from data on health-diagnosis results, genomic information, and data obtained from large-scale cohort surveys. Analysis of genomic data targeting lifestyle-related diseases and analysis of cohort-survey data targeting locomotive syndrome (which is attracting attention as a factor requiring long-term care) are introduced in this article.
Introduction
As global populations age—particularly in Japan’s "super-aged society"—the burden of lifestyle-related diseases, such as Type 2 diabetes, and "locomotive syndrome" (the decline of mobility functions) is skyrocketing. Early intervention is the only sustainable solution to maintain quality of life and reduce healthcare costs. However, effective prevention faces two major hurdles: the immense complexity of high-dimensional biological and lifestyle data, and the friction involved in traditional medical assessment models. NTT researchers are addressing these challenges by leveraging machine learning to move healthcare from reactive treatment to proactive prevention.
The "Superhero" Model: Identifying Protective Factors
A primary focus of this research is the "Superhero" model. This paradigm shift involves identifying individuals who possess genetic markers for high disease risk—based on Single Nucleotide Polymorphisms (SNPs)—yet remain healthy. By using statistical analysis to compare these "superheroes" against those who developed the disease, researchers can isolate specific protective behaviors and lifestyle choices that mitigate innate genetic risks.
Streamlining Prediction for Locomotive Syndrome
Locomotive syndrome, characterized by the risk of requiring long-term care due to mobility decline, is a major health priority. Traditional analysis of this condition involves massive, high-dimensional datasets. To make risk prediction accessible to the general public, the team utilized a 15-year cohort survey containing over 1,000 analytic elements.
Through feature engineering, the researchers used correlation coefficients to prune redundant data, reducing the dataset from over 1,000 elements to approximately 20 key factors. This refined dataset was used to train a multi-layer neural network capable of predicting the probability of a user reaching a locomotive syndrome risk level of 1 or higher within three years.
Datasets and Methodology
The research integrates several high-quality data sources to ensure clinical relevance:
- Genomic & Employee Data: Genetic test results combined with time-series health checkups from NTT employees.
- Long-term Cohort Survey: A 15-year study of approximately 4,400 people conducted in collaboration with the University of Tokyo.
- Standardized Mobility Tests: Physical assessments including the "stand-up test," "two-step test," and the "Locomo 25" questionnaire.
By optimizing these inputs, the team developed a research application that allows users to receive a high/low risk assessment for mobility decline through a simple mobile interface, requiring only basic metrics like age, weight, and lifestyle habits.
Future Outlook: The Bio-Digital Twin
The ultimate goal of this research is the creation of a "bio-digital twin"—a digital simulation of an individual incorporating real-time data from wearables and Electronic Medical Records (EMR). This twin would allow for "what-if" simulations, demonstrating how specific lifestyle changes today could alter a person's health trajectory a decade into the future, thereby empowering individuals to make informed decisions for a healthier life.
Summary updated