Researchers have developed a new method to detect insulin resistance earlier and more accurately by combining data from smartwatches with routine blood tests. The approach can help to identify the risk of diabetes at an early stage.
Insulin resistance is estimated to affect 20 to 40 percent of the general population. In this condition, the body responds less well to insulin, disrupting glucose regulation. Without intervention, this can eventually lead to diabetes. However, in many cases, insulin resistance goes unnoticed for a long time. In addition, recent research showed that insulin resistance is a risk factor for no fewer than twelve different forms of cancer.
Limitations of current diagnostics
When insulin resistance is diagnosed early, the development of diabetes can often be prevented through lifestyle measures such as weight loss, increased exercise and a healthier diet. Yet many diagnostic tests are rarely used in practice.
Current screening methods often focus on snapshots of glucose levels, such as fasting glucose, HbA1c or an oral glucose tolerance test. These tests are not always sensitive enough to detect early stages of insulin resistance. The so-called gold standard test is also expensive, time-consuming and not available everywhere.
Combining data from wearables with blood tests
To develop a more accessible screening method, researchers started the ‘Wearables for Metabolic Health’ (WEAR-ME) study. In this study, data was collected from 1,165 participants. The researchers combined smartwatch data, from sensors that record movement and physiological signals, among other things, with routine blood tests for cholesterol, insulin and glucose, as well as health and lifestyle questionnaires.
The dataset was analyzed using deep neural networks. The researchers then validated the model with an independent group of 72 participants. The results show that the multimodal model can predict insulin resistance with high accuracy. When the model was combined with a so-called wearable foundation model that had been previously trained on 40 million hours of sensor data, the accuracy increased further. The results of the research have been published in Nature.
AI-assist
In addition to the prediction model, the researchers also developed an AI assistant that helps users understand the results of their insulin resistance score. This digital agent can interpret blood test values and provide additional information about metabolic health. The AI assistant’s answers were assessed by endocrinologists. According to the researchers, 79 percent of responses were completely factually correct and 96 percent were considered safe.
According to the researchers, the combination of wearable sensor data, blood tests and AI analysis could lead to a scalable screening method in the future that can also be applied at home. This would make it possible to identify insulin resistance and the risk of diabetes earlier, so that preventive measures can be implemented in a timely manner.