Development of a computer-assisted predictive tool for facilitating selective screening for undiagnosed diabetes mellitus in Vietnam

本研究项目旨在开发一种机器学习（ML）模型来预测越南农村中年人中未诊断的糖尿病（DM），有3个具体目标：1）开发用于DM预测的ML模型; 2）外部验证所提出的DM预测模型;及3）利用外部验证的模型，开发一个用于DM预测的计算机辅助工具。在2021财政年度，我专注于完成第一个研究项目的目标。简言之，庆和心血管研究（KHCS）的基线数据（n = 3000）用于模型开发，该研究是一项基于越南中部人群的心血管疾病队列研究。评估了四种候选ML算法（逻辑回归[LR]，支持向量分类[SVC]，线性SVC和随机梯度下降[SGD]）的预测性能，其中SVC表现出最高的性能（AUC：71.7 ; 95%CI，65.0，78.2）和最佳校准。这些模型使用非侵入性预测因子，即，人口统计学（年龄和性别），人体测量学（例如，腰高比），生活方式（例如，体力活动），和饮食（例如，然而，自2022年8月1日起，我已经离开日本，不再为国家全球健康和医学中心（NCGM）工作。因此，继续该项目不可行，因此，该项目已被终止。

The present research project aims to develop a machine learning (ML) model to predict undiagnosed diabetes mellitus (DM) among middle-aged adults in rural Vietnam, with 3 specific objectives: 1) to develop a ML model for DM prediction; 2) to externally validate the proposed DM predictive model; and 3) to develop a computer-assisted tool for DM prediction, using the externally validated model.During the fiscal year of 2021, I focused on the fulfillment of the first research project's objective. Briefly, the baseline data (n = 3000) of the Khanh Hoa cardiovascular study (KHCS), a population-based cohort study on cardiovascular disease in the Central Vietnam, were used for the model development. Four candidate ML algorithms (Logistic regression [LR], support vector classification [SVC], Linear SVC, and stochastic gradient descent [SGD]) were evaluated for their predictive performance, among them SVC showed the highest performance (AUC: 71.7 ;95% CI, 65.0, 78.2) and the best calibration. The models used non-invasive predictors, i.e., demographic (age, and sex), anthropometric (e.g., waist-to-height ratio), lifestyle (e.g., physical activity), and dietary (e.g., fruit and vegetable consumption) variables.However, since the 1st of August 2022, I have moved out of Japan, and no longer work for the National Center for Global Health and Medicine (NCGM). Because of this, the continuation of the project is not feasible, and thus, the project has been terminated.