For decades, a diabetes diagnosis has relied largely on measuring blood sugar and seeing whether it crosses a clinical threshold. But researchers increasingly worry that approach misses millions of people already progressing toward disease.
Globally, diabetes has become one of the defining health crises of the modern era. According to the World Health Organization, 14 percent of adults were living with diabetes in 2022, up from 7 percent in 1990. In the US, more than 40 million people have diabetes, but around 11 million remain undiagnosed. More than 115 million Americans are estimated to have prediabetes, and roughly 80 percent do not know it. In the UK, around 5.8 million people are living with diabetes, with up to 1.3 million thought to be undiagnosed.
“We’re talking about an epidemic that, in my mind, is way worse than the Covid pandemic,” says Michael Snyder, professor of genetics at Stanford University. “We need new ways of approaching this.”
The danger is not just diabetes itself, but the damage that accumulates silently for years before diagnosis. Persistently elevated blood sugar increases the risk of heart disease, stroke, kidney failure, blindness, and nerve damage. The earlier the disease is identified, the greater the chance of preventing those complications—or avoiding diabetes entirely.
Diagnosis still relies heavily on measuring glucose levels in the blood, most commonly using the HbA1c test, which estimates average blood sugar over the previous few months. While widely used and generally reliable, it is not infallible. Results aren’t able to reflect certain medical conditions or physiological factors that can impact blood sugar levels.
Researchers are increasingly concerned that existing diagnostic tools are also less effective in some populations. Recent studies suggest HbA1c can read falsely low in some Black and South Asian people, delaying diagnosis until the disease is more advanced.
That disparity has triggered growing interest in more personalized and data-rich approaches to diabetes detection: ones that combine biomarkers, wearable devices, and artificial intelligence to identify risk earlier and understand the disease in greater detail.
At Stanford University, Snyder and colleagues have been exploring whether continuous glucose monitors (CGMs)—wearable sensors that track glucose levels in real time—can reveal hidden metabolic patterns long before conventional diagnosis of Type 2 diabetes, which accounts for around 95 percent of cases. While often associated with obesity—which is an important risk factor—slimmer people can also develop Type 2. Snyder himself developed Type 2 diabetes despite not fitting the stereotypical profile for the disease.
“Glucose regulation involves many organ systems: your liver, your muscle, your intestine, your pancreas, even your brain,” Snyder says. “There are lots of biochemical pathways, and it stands to reason that glucose dysregulation may not just be one bucket.”
The Stanford team developed an AI-powered algorithm that analyzes patterns in CGM data to identify different forms of Type 2 diabetes. In tests, the system identified some of these patterns with around 90 percent accuracy.
The researchers believe that the findings could help identify people who are already developing metabolic problems long before a conventional diabetes diagnosis. “It’s a tool that people can use to take preventative measures,” Snyder says. “If the levels trigger a prediabetes warning, dietary or exercise habits could be adjusted, for example.”
CGMs are also becoming cheaper and more accessible, with many now available over the counter in the US. Snyder believes they could eventually become part of routine preventative health care. “In an ideal world, people would wear them once a year,” he says. “The goal from our standpoint is to keep people healthy versus try to fix them later.”