AI Model DiaCardia Detects Prediabetes Using ECG Data, Eliminating Need for Blood Tests

The model has shown strong accuracy using both clinical 12-lead ECGs and simplified single-lead signals, suggesting future applications in wearable and home-based screening.

DiaCardia, an artificial intelligence model, has been developed to detect prediabetes using electrocardiogram (ECG) data alone, without relying on blood tests.

The model has shown strong accuracy using both clinical 12-lead ECGs and simplified single-lead signals, suggesting future applications in wearable and home-based screening.

“DiaCardia has the potential to make prediabetes screening scalable, accessible, and available anytime, anywhere, without a blood test,” said Chikara Komiya, Junior Associate Professor, Institute of Science, Tokyo. “By promoting widespread screening of prediabetes, this work will ultimately contribute to the prevention of diabetes.”

Routine screening for prediabetes typically depends on fasting plasma glucose or HbA1c blood tests, which require clinical visits and may discourage participation due to cost, inconvenience, or lack of symptoms. As a result, many individuals progress to type 2 diabetes without awareness of their risk status.

Researchers have increasingly explored whether ECG data, a non-invasive and widely available clinical measurement, could serve as an alternative screening signal. Prediabetes and diabetes are known to influence cardiac structure and autonomic nervous system function, changes that can subtly alter electrical activity in the heart.

Earlier studies have used deep learning to identify diabetes by combining ECG signals with demographic data, but accurate detection of prediabetes using ECG alone has remained unproven.

In a study published in Cardiovascular Diabetology, a research team from the Institute of Science Tokyo introduced DiaCardia, an AI model designed specifically to address this gap. The team was led by Junior Associate Professor Chikara Komiya, graduate student Dr Ryo Kaneda, and Professor Tetsuya Yamada from the Department of Molecular Endocrinology and Metabolism.

“This is the first robust, interpretable, and generalizable AI model capable of identifying individuals with prediabetes using either 12-lead or single-lead ECG data,” explains Komiya.

During internal testing, DiaCardia achieved an area under the receiver operating characteristic curve of 0.851 using ECG data alone. The model also performed well when evaluated on external data from another institution, without retraining, indicating strong generalizability.

Explainability analysis showed that higher R-wave amplitudes and reduced heart rate variability were key predictors, findings consistent with known effects of insulin resistance and autonomic dysfunction.

Notably, the researchers found that even a simplified version of the model using single-lead ECG data and only 28 features delivered nearly comparable performance. This result supports the feasibility of integrating DiaCardia into consumer wearables such as smartwatches.

Stay tuned for more such updates on Digital Health News