A simple online screening tool aims to catch diabetes risk earlier
Researchers in Denmark say they have developed an at-home screening tool designed to identify people who may have type 2 diabetes or prediabetes before they ever book a doctor’s appointment. The project addresses a familiar public-health problem: many people with early metabolic disease do not know they are at risk, which means diagnosis often comes late, after damage has already begun.
According to the Danish Diabetes Association, about 100,000 people in Denmark are unaware they have type 2 diabetes, while roughly half a million are estimated to have prediabetes. That gap matters because type 2 diabetes can be confirmed with a blood test, but many people never seek one unless symptoms become noticeable or a clinician flags the possibility. The new tool is intended to work upstream of that process by nudging people toward follow-up care if their risk appears elevated.
The screening system is called MEDWACS. It was developed by postdoctoral researcher Daniel Yoo at DTU Sustain, together with DTU professor Olivier Jolliet and Italian physician Umberto Maggiore, whose work focuses on kidney disease, a frequent complication of diabetes. Their study was published in the Journal of Clinical Epidemiology.
How the test works
The core idea behind MEDWACS is that a diabetes-risk screen does not have to begin with lab work. Instead, the online test uses information people can provide on their own at home. The source report specifically highlights body measurements, age, sex, and other accessible inputs, including the length of the upper leg, as part of the model’s risk assessment.
That makes the tool different from traditional clinical screening pathways, which usually begin in a doctor’s office or through formal health checks. The researchers’ argument is not that an online questionnaire can replace a diagnosis. It cannot. Instead, they frame it as a triage layer: a low-friction, accessible prompt that may persuade people with an elevated risk profile to seek the blood testing needed for confirmation.
Yoo said the logic is straightforward. If people do not suspect they have diabetes, they are unlikely to visit a doctor for testing. A credible online screen, by contrast, can lower the barrier to first action. That may be especially useful for people in the gray zone of prediabetes, where disease progression can still potentially be slowed or interrupted through treatment and lifestyle changes.
Built on long-run health data and AI methods
The research team says MEDWACS was developed using artificial intelligence and large volumes of U.S. health data from the National Health and Nutrition Examination Survey. The dataset spans 30 years and includes adults aged 18 and older of both sexes. That long time horizon gave the model a broad base of physiological and demographic variation from which to learn patterns associated with diabetes and prediabetes risk.
In practical terms, the use of AI here appears less about futuristic automation than about pattern recognition across many variables. Screening rules built from a handful of conventional markers can miss subtler combinations that still signal elevated risk. By training on a large historical dataset, the researchers sought to create a tool that can extract those combinations while still relying on inputs ordinary users can supply themselves.
The source text does not provide performance numbers, but it does say the team conducted two external validation studies and found the tool highly effective at detecting people with the condition. That matters because external validation is one of the more important steps in evaluating whether a model can generalize beyond the data it was originally trained on.
Why external validation matters
Medical prediction tools often look promising in development and then weaken when tested in other populations. The researchers tried to address that problem by validating MEDWACS on two separate populations in the United States and South Korea where the number of people with and without the condition was already known.
Testing across different populations is an important signal of seriousness in digital health. It does not guarantee the model will work equally well everywhere, but it helps answer a basic question: is the system picking up durable risk patterns, or just fitting quirks in one dataset? The fact that the team chose two external populations, rather than relying only on internal testing, suggests an effort to move beyond a proof of concept.
There are still obvious limits. A tool developed from historical health datasets will reflect the populations represented in those data, and any broad public rollout would still need careful monitoring in other countries, ethnic groups, and care settings. The source material also does not say whether the tool has yet been integrated into routine care pathways, reimbursed programs, or national screening systems.
The broader significance for preventive care
The bigger story is not only about one screening website. It is about the growing use of lightweight AI tools to extend the reach of preventive medicine. Type 2 diabetes is common, costly, and often silent in its earlier stages. The Danish Diabetes Association says the disease costs Danish society at least DKK 13.3 billion per year, while also raising the risk of serious complications including cardiovascular disease, kidney failure, and vision loss.
That makes early detection unusually valuable. Unlike some conditions where earlier awareness changes little, diabetes risk can often be managed with relatively familiar interventions once it is identified. For health systems, that creates an incentive to experiment with cheap front-end screening tools that can push people into formal care sooner.
The association welcomed efforts to detect type 2 diabetes and prediabetes earlier, emphasizing that earlier identification improves the chances of preventing severe complications. In that sense, MEDWACS fits a broader health-tech trend: using digital tools not to replace clinicians, but to widen the funnel of people who enter clinical care before disease becomes harder and more expensive to manage.
Whether this specific tool becomes widely adopted will depend on how it performs in real-world use, how clearly it communicates risk, and whether users who score high actually follow through with medical testing. But the project shows where practical AI in health may have some of its strongest near-term value: not in dramatic diagnostic claims, but in simple, scalable tools that help more people realize they should get checked.
This article is based on reporting by Medical Xpress. Read the original article.
Originally published on medicalxpress.com
