HMN 2026: How Smartwatch and blood test data combine to better predict insulin resistance and diabetes

Around 20–40% of the general population are estimated to have insulin resistance—a condition where insulin begins to be less effective in the body, and glucose regulation becomes more difficult. Eventually, this can lead to diabetes. Often, insulin resistance goes undetected until prediabetes or diabetes develops. But now, a more reliable and accessible method of detecting insulin resistance may be on the way. A new study, published in Nature, outlines a method combining smartwatches and routine blood tests to better detect insulin resistance in its early stages, and its early validation testing has shown good accuracy.

Detecting insulin resistance

When detected early, insulin resistance is reversible and progression to diabetes is preventable through lifestyle interventions, like weight loss, exercise and a healthy diet. However, most tests for detecting insulin resistance are not implemented routinely, and many people don’t have obvious symptoms to spur doctors to do testing. On top of that, the “gold-standard” test for insulin resistance is expensive, time-consuming, and may not be available in many settings. Tests that are administered aren’t always reliable.

“Instead, a focus on snapshots of glucose levels, fasting glucose, HbA1c or glucose levels after a two-hour oral glucose tolerance test (OGTT) represents the typical screening approach, and can be insensitive to those in the early stages of insulin resistance,” say the authors of the new study.

Because early detection of insulin resistance is important for preventing diabetes and its related complications, a more scalable, affordable, and accessible method of detection could improve health outcomes for many people. And so, the authors of the new study decided to look to devices that many people already have—smartwatches.

More accessible insulin resistance detection with wearables

The research team put together the Wearables for Metabolic Health (WEAR-ME) study, incorporating data from wearable smartwatches and routine blood testing of cholesterol, insulin and glucose, along with health and lifestyle questionnaires. Data from 1,165 individuals was gathered remotely and analyzed using deep neural networks. Then the team validated results using cross-validation and an independent cohort of 72 people.

Results showed that the multimodal model predicted insulin resistance with high accuracy. When the model was fine-tuned with a wearable foundation model (WFM) pretrained on 40 million hours of sensor data, accuracy improved even further.

“The results provide further evidence that data from wearables provide considerable added value in predicting IR, even when the model is applied to previously unseen data. A model integrating WFM-derived representations with demographics surpassed a demographics-only baseline (AUROC?=?0.75 versus 0.66). Furthermore, adding WFM representations to an optimal model that included demographics, fasting glucose and a lipid panel substantially improved predictive performance over an identical model without data from wearables (AUROC?=?0.88 versus 0.76),” the study authors explained.

An AI agent for communicating results

The team also designed an AI agent for interpreting and communicating insulin resistance assessment results to users. To ensure accuracy and reliability, the team had endocrinologists assess and rate the AI agent’s responses. According to the endocrinologists, 79% of responses were completely factually accurate and 96% of responses considered
safe. The agent was able to accurately reference and interpret blood test values.

“Our proposed agent, called the insulin resistance literacy and understanding agent (IR agent), uses a reason and act (ReAct) framework that is built on top of an LLM—in our case, Gemini 2.0 Flash. Our agent combines the language understanding of an LLM with the ability to perform actions, such as searching the web for up-to-date information, accessing specialized tools like a calculator and using our IR prediction models. This allows the IR agent to dynamically plan its response to a user’s query about their metabolic health, grounding its answers in real-world data and verifiable calculations, rather than relying solely on the LLM’s pre-existing knowledge,” the study authors wrote.

After some further validation testing, this new method for testing insulin may offer a more scalable, at-home screening option, making early diabetes risk detection more accessible for more people.

The content is provided for information purposes only.