Development of a Simple Youth Diabetes Screening Tool Using Machine Learning

点击翻译按钮获取中文摘要

Project Summary The number of youth with type 2 diabetes in the U.S. is projected to increase by a staggering 49 percent by 2050. However, simple screening tools to reliably identify diabetes risk and prevent the adverse effects of this serious disease are only available for adults, not for youth. Indeed, our preliminary studies using nationally representative data from the National Health and Nutrition Examination Survey (NHANES) found that published pediatric clinical guidelines performed relatively poorly in capturing youth with diabetes or its precursor condition (prediabetes). In response to this urgent health challenge, this R21 research proposal aims to bring together clinical, epidemiology and data science experts to develop and validate a youth diabetes risk screener. We will develop a user-friendly screening tool to identify youth with prediabetes or diabetes by leveraging state of the art machine learning techniques and rich data from NHANES. Our final product will be a web-based screener that can be integrated into both digital health platforms and traditional public health surveillance efforts. This translational product will aid parents, community-based organizations, schools, and primary care providers in accurately identifying youth at risk of diabetes who can benefit from subsequent definitive diagnostic testing, as well as prevention and medical management programs. Specific Aims: 1. We will develop an initial candidate screener to distinguish between normal and prediabetic/diabetic youth by applying parsimonious predictive modeling-oriented machine learning techniques to NHANES data. 2. We will develop additional candidate screeners that integrate the various domains of NHANES data, and will identify the best-performing screener for youth by comparing the performances of all the candidates. 3. To address the importance of sociodemographic factors for prediabetes/diabetes screening, we will also develop candidate screeners specific to sociodemographic subgroups based on age, sex and race/ethnicity. We will validate all these candidates to develop an integrated screener that is accurate and personalized to individuals based on their sociodemographic characteristics (age, sex and race/ethnicity). Finally, we will implement a user-friendly web-based version of the integrated screener than can help identify youth at risk of diabetes, and become part of a community-based youth diabetes prevention strategy for future implementation in high-risk communities.