Upper Leg Length Joins The List Of Diabetes Warning Signs

Free Online Tool Estimates Risk Without A Needle

Nearly one in three American adults with elevated blood sugar has no idea anything is wrong, going undiagnosed and untreated. That gap between silent disease and medical attention is exactly where diabetes causes damage, and a newly published study is taking direct aim at it. Researchers have developed a free, AI-powered online tool that can estimate the likelihood that someone currently has prediabetes or diabetes, using seven inputs and no blood draw.

Diabetes already costs the United States an estimated $412 billion a year and is projected to affect 700 million people globally by 2045, according to figures cited in the study. Yet standard screening still depends heavily on lab tests and clinic visits, creating real barriers for millions of people. This new tool, called MEDWACS, was built to serve as a no-cost, at-home first step. Unlike many risk calculators that estimate a person’s future odds, it focuses on whether someone already has undiagnosed prediabetes or diabetes, and it needs only a tape measure, a blood pressure cuff, and a few basic body measurements, though it is not a replacement for a doctor’s diagnosis.

Published in the Journal of Clinical Epidemiology, the study found that MEDWACS outperformed the established screening guidelines from two of the country’s leading health authorities at identifying high-risk people within the study’s own validation analyses. Its seven inputs are simple measurements, most obtainable with a tape measure, though a blood pressure reading calls for a cuff and body mass index calls for height and weight.

Building a Diabetes Risk Tool From 30 Years of Health Data

To build MEDWACS, researchers analyzed 30 years of data from the National Health and Nutrition Examination Survey, a large, ongoing federal study that tracks the health and diet of Americans across the country. From an initial pool of more than 135,000 participants, the team focused on 17,458 adults who had complete records for the 159 predictors the researchers retained, plus two blood-sugar markers used to define the condition: a fasting blood sugar reading and a separate marker that reflects average blood sugar levels over time. Anyone who tested above the threshold on either or both blood-sugar measures was classified as having prediabetes or diabetes, totaling about 53% of the included participants.

Using a type of artificial intelligence that learns patterns from large amounts of data, the research team evaluated 159 possible predictors of diabetes risk, ranging from body measurements to dietary habits. A statistical sorting process identified 32 meaningful predictors, and from those, the team selected the seven that were both highly informative and realistically measurable at home without any medical equipment beyond a tape measure, a scale, a way to measure height, and a blood pressure monitor.

Those seven variables are: age, waist circumference, the top number in a standard blood pressure reading, gender, upper leg length, arm circumference, and body mass index (BMI, calculated from height and weight). Among them, age, waist size, and blood pressure carried the most predictive weight. Upper leg length and arm circumference also proved surprisingly important. Researchers note these measurements can reflect how the body developed in early life and how fat is distributed across the body, both of which are connected to the body’s ability to regulate blood sugar.

Diabetes Risk Tool Performance Against Standard Screening

Researchers tested six different AI model types before selecting a neural network, a type of AI loosely modeled on the way the human brain processes information, as the best performer. On internal testing, MEDWACS achieved a score of 0.804 on a scale of 0 to 1 (where 1 is perfect) on a standard measure of how well a test separates people with a condition from those without it. The authors described that result as strong discrimination, meaning the model was fairly good at telling people with abnormal blood sugar apart from those without it.

To confirm the tool was not tailored to only one population, the team then tested it on two completely separate, more recent datasets: survey data from 2021 to 2023 covering 3,043 Americans, and a 2023 health survey from South Korea that included 5,492 participants. MEDWACS held up well in both cases, scoring 0.773 in the US external test group and 0.780 in the Korean group. Performance remained solid across different age groups, genders, income levels, education levels, and body types, evidence that the tool held up across the populations tested.

Perhaps the most pointed finding involved a side-by-side comparison with existing clinical guidelines. Using a method that weighs the real-world benefit of identifying true cases against the cost of false alarms, MEDWACS came out ahead of the US Preventive Services Task Force recommendation and a simplified version of the American Diabetes Association screening strategy used by the researchers, who relied only on the recommendation to screen adults 35 and older rather than the full ADA framework. At a 30% risk threshold, the specific decision point examined, MEDWACS detected five additional true positive cases per 100 people screened compared to the American Diabetes Association strategy, without increasing unnecessary follow-up testing. That figure comes from the study’s decision curve analysis and should not be taken as a guaranteed real-world outcome.

Who This Diabetes Screening Tool Is For and What It Can’t Do

MEDWACS is designed for adults 18 and older who have not yet been diagnosed with diabetes. Entering the seven measurements into the online tool produces a risk probability displayed on a gauge chart. For people younger than 65, a predicted probability above roughly 45% is flagged as warranting a follow-up visit for clinical testing. For those 65 and older, the threshold sits higher.

Researchers are candid about what the tool cannot do. Because it was built on data collected at a single point in time rather than tracking people over years, it predicts whether someone likely has prediabetes or diabetes right now, not whether they will develop it in the future. It also cannot distinguish between prediabetes and diabetes. A high-risk result is a prompt to see a doctor, not a diagnosis. Authors also acknowledge that the dataset used to build the tool required participants to have complete records across all variables, which could mean certain groups were underrepresented. Physical activity data, for example, had to be excluded due to inconsistent tracking across the full 30-year survey period.

Funding disclosures note that portions of the research were supported by the Novo Nordisk Foundation, and one co-author reported advisory and speaking relationships with separate outside organizations unrelated to this study. The authors state that the funders played no role in the study’s design, data collection, analysis, or interpretation.

With diabetes affecting tens of millions who don’t yet know it, a free tool that asks only for a few body measurements is not a cure. But it could be the nudge that gets someone into a doctor’s office before real damage is done.

---

Disclaimer: This article describes a cross-sectional machine-learning study. MEDWACS estimates the likelihood that a person currently has prediabetes or diabetes; it does not diagnose either condition, cannot predict future onset, and cannot distinguish prediabetes from diabetes. Because the model was built on data collected at a single point in time, it cannot establish cause and effect. A high-risk result is a prompt to seek clinical testing, not a substitute for evaluation by a licensed medical professional.

---