Type 2 diabetes is a chronic condition that affects the way your body regulates blood sugar (glucose), which is the main source of energy for your cells. In Type 2 diabetes, your body either resists the effects of insulin - a hormone that regulates the movement of sugar into your cells - or doesn't produce enough insulin to maintain normal glucose levels. As a result, glucose builds up in the bloodstream, leading to high blood sugar levels.
Despite common misconceptions, Type 2 diabetes isn't just about lifestyle choices or poor habits. It's a complex, multifaceted condition influenced by genetics, hormones, and various metabolic processes. In fact, the pathways leading to high blood sugar can vary significantly from person to person.
Stanford Medicine's latest AI research uncovers hidden subtypes of Type 2 diabetes, opening the door to personalized care.
About 95% of diabetes cases are Type 2, but new research from Tracey McLaughlin, MD, and her team uses AI and machine learning to challenge conventional wisdom and show there isn't just one way to develop it.
Scientists have long known that different biological processes contribute to high blood sugar, which is a key indicator of diabetes. Some people develop insulin resistance, where their muscles and liver don't respond properly to insulin; others have something called beta-cell dysfunction, meaning their pancreas struggles to produce enough insulin; and some have issues with gut hormones - called incretins - that normally help regulate blood sugar.
Traditionally, diagnosing diabetes and prediabetes has relied on measuring fasting glucose levels or the "A1C test," which tracks average blood sugar over three months. But these conventional methods don't capture the complexity of how blood sugar rises and falls throughout the day.
That's where machine learning comes in. Machine learning is a process where computers learn from data and improve their performance over time without being explicitly programmed. It involves feeding large amounts of data into algorithms that identify patterns and make predictions or decisions based on that data.
By analyzing how your blood glucose goes up and down over time, McLaughlin and team, in collaboration with data scientists from Michael Synder, PhD's lab, trained an AI model to predict which metabolic issue is most likely to lead to development of diabetes in a given individual.
Using a simple test, the study found that AI could accurately detect insulin resistance 95% of the time, beta-cell dysfunction 89% of the time, and incretin deficiency 88% of the time.
The researchers tested their AI model using glucose data from people wearing continuous glucose monitors (CGMs) at home. These small, wearable sensors track blood sugar levels in real time, providing a detailed picture of how glucose fluctuates throughout the day. The results were nearly as accurate as gold-standard metabolic tests performed in a research clinic.
With this AI-powered approach to analyzing continuous glucose monitoring data, doctors can pinpoint and even predict each patient's unique subtype of Type 2 diabetes or prediabetes, paving the way for more customized treatments and personalized care. The research shows: AI can transform early detection of Type 2 diabetes, and lead to improved health outcomes like never before.
