AI predicts heart risks years before symptoms via sleep data

American scientists have engineered an artificial intelligence system capable of forecasting cardiovascular disease risks several years prior to the onset of clinical symptoms. The innovation relies on interpreting electrocardiogram signals captured during sleep, potentially revolutionizing early detection methods for heart and vascular disorders.

Study methodology and data sources

The research, backed by the National Institutes of Health (NIH), demonstrated that deep learning algorithms can process cardiac electrical activity across various sleep stages. This technology identifies individuals at heightened risk for atrial fibrillation and heart failure while estimating mortality probabilities over the subsequent decade.

Investigators utilized records from 38,195 patients who underwent sleep studies at three US medical centers. The dataset included single-channel ECG recordings alongside detailed information on sleep phases, which served to train and validate the predictive capabilities of the AI model regarding future health outcomes.

Validation across independent cohorts

Model development drew on data from 15,809 patients at Massachusetts General Hospital in Boston. Subsequent testing occurred on two separate groups: 9,810 patients from Emory University Hospital in Atlanta and 12,576 patients from Beth Israel Deaconess Medical Center in Boston.

Results indicated the system successfully stratified patients into varying risk categories. These predictions remained effective even after accounting for traditional risk factors such as age, gender, obesity, diabetes, hypertension, and sleep disorders including apnea.

Predictive accuracy and limitations

The study revealed promising capacity for predicting atrial fibrillation, heart failure, and all-cause mortality. However, accuracy levels for forecasting heart attacks and strokes still require further refinement.

According to findings published in the journal SLEEP, integrating AI estimates with known risk factors improved prediction performance for atrial fibrillation by approximately 4 to 6 percentage points. For heart failure, the improvement ranged from 8 to 9 percentage points, while mortality predictions saw a gain of 4 to 7 percentage points based on statistical metrics evaluating model precision.

Clinical implications and future integration

Dr. David Gov, acting director of the National Heart, Lung, and Blood Institute in the United States, stated that this approach could help identify people susceptible to heart disease years before symptoms appear. He noted that the next step involves assessing whether the technique enhances existing predictive tools, paving the way for potential inclusion in hospital sleep tests.

Gary Clifford, a co-researcher from Emory University School of Medicine, emphasized that recorded sleep data can pinpoint high-risk patients, enabling earlier preventive measures and better medical decision-making.

Cost-effective monitoring potential

A key advantage of the new method is its reliance on a single channel for recording cardiac electrical activity, contrasting with traditional ECGs that use 12 leads. Researchers suggest this simplicity may allow future use of basic sensors or low-cost medical patches to gather necessary data during a single night’s sleep, avoiding complex diagnostic procedures.