Updated on 2024/05/26

写真a

 
Wang, Jinfang
 
Affiliation
Faculty of International Research and Education, School of International Liberal Studies
Job title
Professor
 

Syllabus

 

Internal Special Research Projects

  • 最適な血糖値予測モデルの構築について

    2023   Shigetoshi Hosaka

     View Summary

    Purpose:In data science, two distinct cultures exist: model-based statistics and algorithm-based machine learning. Statistical models offer interpretability, whereas machine learning algorithms are known for their predictive prowess. This research is part of a larger project of combining statistical inference with machine learning prediction to make optimal personalized management for subjects with specific conditions such as prediabetes. Our framework generalize a recently proposed health improvement framework [1]. Methods:An iterative algorithm that fuses state-of-the-art machine learning predictors like XGBoost with statistical inference using Bayesian linear model is proposed. We have done comprehensive case studies focusing on prediabetes management, leveraging health checkup data from a large hospital in Edogawa Ward, Tokyo. Clinically, an individual is diagnosed with type 2 diabetes if the glucose level exceeds 126 mg/dL and with prediabetes if the level is between 100-125 mg/dL. For glucose levels under 100 mg/dL, the individual is considered normal. To effectively manage prediabetes, we proposed XGBoost to make both precise and robust forecasts of glucose levels using sequentially modified risk factors associated with lifestyle, like Body Mass Index (BMI) and systolic blood pressure. Main Conclusions:In a Spanish cohort study [2], 41% of prediabetes remained prediabetic, 23% progressed to diabetes, and 36% returned to normoglycemia. Employing our model initially constructed for glucose level, but based on a closely related outcome of HbA1c, significant improvements were observed: merely 3.5% remained prediabetic, 7% progressed to diabetes, and a remarkable 89.5% reverted to normoglycemia. The main findings were presented in Wang and Hosaka (2023) [3]. We will explore applications of the proposed method for personalized management in other areas such as constructing strategies for the prevention and reduction of juvenile delinquency. [1] Nakamura, K., et al., Nature Communications, 2021, 12:3088; https://www.nature.com/articles/s41467-021-23319-1[2] Miquel B., et al., Nutrients 2020, 12(5), 1538; https://doi.org/10.3390/nu12051538[3] Wang, J. and Hosaka, S. (2023). The 12th conference of the Asian Regional Section of the International Association for Statistical Computing (IASC-ARS), 6–8 December, 2023, Macquarie University, NSW, Australia.