To date, lack of an accurate mechanism for measuring the QTc interval has represented a significant challenge for clinicians, especially among high-risk patient groups such as those with atrial fibrillation (A-fib). New research from NYU Langone Health suggests estimating the QTc interval using artificial intelligence (AI) could allow clinicians to rule out prolongation of the QTc interval associated with risk of torsades de pointes.

A research team led by cardiologist Lior Jankelson, MD, PhD, director of the Inherited Arrhythmia Program at NYU Langone’s Heart Rhythm Center, developed the AI-driven QT correction algorithm to rule out prolongation of the QTc interval in A-fib. A report published in JACC: Clinical Electrophysiology shows the tool could be useful in cases where a concurrent sinus electrocardiogram (ECG) is not available, potentially addressing both QT measurement and correction problems.

“Determination of the QTc interval is of critical importance. And in A-fib, it is very difficult to accurately measure the QTc,” Dr. Jankelson says.

Leveraging the Dataset

The researchers’ convolutional neural network (CNN)–based model was designed to maximize objectivity and clinical applicability. Their approach incorporated the ECG signal as well as general patient and cardiac-specific data, such as sex, race, QRS duration, heart rate, and more into the predictions.

“We identified patients with a 10-second 12-lead ECG in A-fib within 10 days of a sinus ECG with similar QRS durations,” Dr. Jankelson explains.

The full dataset included 6,432 patients in A-fib with an average age of 71 years. The data were randomly divided to generate subsets, with 60 percent allotted to training, 20 percent to validation, and 20 percent to testing. Dr. Jankelson and his team evaluated the model for predicting the QTc value in the sinus ECG based on an input of the A-fib ECG waveform and its associated features.

High Degree of Accuracy

After training, the model performed best for ruling out QTc prolongation, exhibiting high negative predictive value (0.82 in males and 0.92 in females) and specificity (0.92 in males and 0.97 in females). In addition, approximately 84 percent and 97 percent of the predictions were contained within 1 standard deviation (SD) and 2 SD from the sinus QTc interval.

The model also outperformed the AFQTc method, exhibiting significantly narrower error ranges.

“Our model applied to A-fib ECGs accurately predicted the sinus QTc interval, outperforming current alternatives and exhibiting a high negative predictive value.”

Lior Jankelson, MD, PhD

“To our knowledge, this is the first published deep neural network developed for QTc prediction in A-fib ECGs,” Dr. Jankelson says. “Our model applied to A-fib ECGs accurately predicted the sinus QTc interval, outperforming current alternatives and exhibiting a high negative predictive value.”

Improvements Over Time

As with most AI-driven models, performance improves over time. Dr. Jankelson says the model might therefore be used to inform clinical decision-making in various patient cases.

“For example, QT-prolonging medications for the management of A-fib or comorbid conditions could potentially be safely prescribed, using data derived from the AI tool,” Dr. Jankelson explains.

“The clinical impact of this AI tool is potentially great with integration into existing EMRs or commercially available ECG software,” he adds.