Youth Diabetes Crisis

Youth diabetes represents one of the most pressing public health crises facing American families today. Diabetes among youth in the United States has increased significantly over the last two decades; from 2001 to 2017, the number of youth living with Type 1 diabetes increased by 45%, while the number of youth living with Type 2 diabetes nearly doubled, increasing by 95%. If current trends continue, the number of young people with diabetes could increase as much as 700% by 2060, particularly for Type 2.

Most alarmingly, one in three American teenagers, ages 12-19, has prediabetes, according to CDC research. Prediabetes represents adolescents with blood glucose levels higher than usual but not yet high enough to be classified as Type 2 diabetes. That’s over 8.4 million young people on a dangerous path toward a lifetime of chronic disease!

Our Story

YouthDiabetes.AI was created to tackle the growing youth diabetes crisis by making AI-powered health tools accessible to everyone. Our goal is to empower young people and their families with early risk detection and actionable guidance, completely free and available to anyone with internet access.

This project combines two powerful AI approaches: machine learning for risk prediction and generative AI for personalized recommendations, transforming complex health data into an easy-to-understand tool for the public.

Technical Approach

Machine Learning Risk Prediction

Our prediction model was trained on the most recent comprehensive youth diabetes dataset from Mount Sinai School of Medicine, released in July 2024. This dataset draws from the National Health and Nutrition Examination Survey (NHANES), incorporating health indicators from national surveys spanning 1999-2018 and covering over 15,000 individuals aged 12-19.

The development process involved extensive data preparation, a hybrid feature selection pipeline that reduced over 100 variables to approximately 30-40 key predictors, and evaluation of multiple algorithms including XGBoost, AdaBoost, Random Forest, and Logistic Regression. All models were assessed using weighted metrics to ensure accuracy across diverse populations.

Generative AI Recommendations

Once risk factors are identified, generative AI creates personalized daily meal plans, weekly exercise schedules, and practical lifestyle recommendations tailored specifically to each user's profile. The system is carefully engineered to provide actionable guidance while avoiding specific medical claims.

To ensure accuracy, AI outputs were validated using the LLM-as-a-Judge method, where we cross-checked responses across multiple AI systems (ChatGPT, Deepseek, Gemini, Grok) to verify consistency and accuracy for general lifestyle guidance.

Responsible AI in Healthcare

AI is transforming healthcare by enabling better diagnoses, personalized treatments, and improved patient outcomes. However, these technologies require careful implementation. AI can produce inaccurate information and suffers from biases that affect vulnerable populations.

YouthDiabetes.AI is designed with responsibility in mind:

• Educational focus: We provide general health guidance, not medical diagnoses
• Complementary care: We encourage users to work with healthcare professionals
• Accessible design: Free and easy to understand for users of all health literacy levels

When integrated responsibly with strong governance and collaboration with healthcare professionals, AI can support healthier lives without compromising safety.