Beyond the Blood Draw: New AI Tool Detects Prediabetes Using Heart Signals

January 20, 2026

TOKYO — Researchers have developed a breakthrough artificial intelligence model capable of detecting prediabetes using only electrocardiogram (ECG) data, potentially eliminating the need for invasive blood tests in early screenings. The model, dubbed DiaCardia, marks a significant shift in metabolic monitoring by identifying subtle “heart signatures” of impaired blood sugar long before traditional symptoms appear. Published in Cardiovascular Diabetology, the study suggests that this technology could soon bring clinical-grade screening to consumer wearables like smartwatches, allowing for “anytime, anywhere” health monitoring.

The Silent Window of Opportunity

Prediabetes is often described by clinicians as a “silent” condition. It occurs when blood glucose levels are higher than normal but haven’t yet crossed the threshold into type 2 diabetes. Because it rarely presents physical symptoms, millions of people remain unaware of their risk until permanent damage to the heart, kidneys, or nerves has already begun.

“Early detection is our best weapon against the diabetes epidemic,” says Junior Associate Professor Chikara Komiya of the Institute of Science Tokyo, who led the research. “The challenge has always been participation. People avoid blood tests because they are inconvenient or require a needle stick. DiaCardia changes that equation.”

Type 2 diabetes is a global health crisis characterized by the body’s inability to produce or effectively use insulin. However, the prediabetes stage offers a critical window where lifestyle interventions—such as diet and exercise—can often reverse the condition or significantly delay its progression.

How AI “Hears” High Blood Sugar

While an ECG is a standard tool for measuring the heart’s electrical activity, its link to blood sugar is subtle. The Japanese research team, including Dr. Ryo Kaneda and Professor Tetsuya Yamada, utilized a machine learning algorithm called LightGBM to analyze 16,766 health records.

The AI was trained to look at 269 specific waveform features from standard 12-lead ECGs. Unlike previous models that required a patient’s age or sex to make a guess, DiaCardia achieved high accuracy using heart signals alone.

Key Predictors Identified by DiaCardia:

• Increased R-wave Amplitude: Higher peaks in certain electrical leads, often associated with increased heart mass—a common side effect of insulin resistance.

• Reduced Heart Rate Variability (HRV): A sign of autonomic neuropathy, where high blood sugar begins to affect the nerves controlling the heart.

“The model isn’t just finding patterns; it’s identifying physiological changes,” explains Komiya. “It captures how impaired glucose regulation physically alters the heart’s electrical pathways.”

From the Clinic to the Wrist

One of the most significant findings of the study is DiaCardia’s performance with single-lead ECG data. Traditional clinical ECGs use 12 leads (sensors) placed across the chest and limbs. However, most modern smartwatches use only a single lead (the contact between the back of the watch and the wearer’s opposite finger).

Even when the researchers stripped the data down to just one lead, the AI’s accuracy remained nearly as high as the full clinical version.

Note: An AUROC score of 0.851 is considered “excellent” in diagnostic modeling, where 1.0 is a perfect prediction.

Expert Perspectives and Public Health Impact

Independent experts see this as a potential game-changer for public health, though they urge cautious optimism.

“The ability to screen for metabolic issues through a non-invasive, passive device like a watch could capture millions of ‘missing’ patients,” says Dr. Elena Rossi, a cardiologist not involved in the study. “However, we must ensure these tools lead to clinical follow-ups rather than self-diagnosis. An AI alert should be the start of a conversation with a doctor, not the final word.”

The study also addressed concerns about “confounding factors.” Even when accounting for variables like age, smoking status, and blood pressure, DiaCardia maintained its predictive power, suggesting it is truly sensing glucose-specific changes rather than just general poor health.

Limitations and Next Steps

Despite the promising results, the researchers note that DiaCardia was validated primarily on a specific demographic in Japan. Further global studies are needed to ensure the AI’s accuracy across different ethnicities and age groups. Additionally, while the model is “interpretable” (meaning scientists understand why it makes a decision), it is not a replacement for a formal diagnosis by a physician.

For now, the team is looking toward integrating DiaCardia into wearable software platforms. If successful, the next “low heart rate” or “irregular rhythm” notification on your smartwatch might actually be an early warning to check your blood sugar.

“DiaCardia has the potential to make screening scalable and accessible,” concludes Komiya. “By moving screening out of the lab and into daily life, we can catch prediabetes before it becomes a life-altering disease.”

Reference Section

Primary Study:

• Koga, D., Kaneda, R., Komiya, C., et al. (2025). “Artificial intelligence identifies individuals with prediabetes using single-lead electrocardiograms.” Cardiovascular Diabetology. DOI: 10.1186/s12933-025-02982-4.

Medical Disclaimer: This article is for informational purposes only and should not be considered medical advice. Always consult with qualified healthcare professionals before making any health-related decisions or changes to your treatment plan. The information presented here is based on current research and expert opinions, which may evolve as new evidence emerges.