Model may offer more efficient way to detect congenital long QT syndrome

An artificial intelligent-enhanced ECG (AI-ECG) was able to distinguish patients with long QT syndrome (LQTS) from those who did not have LQTS, researchers reported, potentially offering a “simple and inexpensive method for early detection of congenital long QT syndrome.”

In the diagnostic study that used a deep neural network, the 12-lead AI-ECG successfully pinpointed patients with LQTS (n=967) from those who were evaluated for LQTS but were discharged without a diagnosis (n=1,092), all of whom presented to a specialized arrhythmia clinic, according to Michael J. Ackerman MD, PhD, of the Mayo Clinic in Rochester, Minnesota, and co-authors.

The model performed better than the corrected QT (QTc) alone, even in the setting of a normal QT interval, they wrote in JAMA Cardiology.

Specifically, based on the ECG-derived QTc alone, patients were classified with an area under the curve (AUC) value of 0.824 (95% CI 0.79 to 0.858). When AI-ECG was used, the AUC was 0.900 (95% CI 0.876 to 0.925).

Also, in the subset of patients who had a normal resting QTc (<450 milliseconds), the QTc alone distinguished those with LQTS from those without LQTS with an AUC of 0.741 (95% CI 0.689 to 0.794), while AI-ECG increased this discrimination to an AUC of 0.863 (95% CI 0.824 to 0.903), they reported.

“Although QT prolongation is LQTS’ pathognomonic feature, approximately 40% of patients with proven (genotype-positive) LQTS have a normal QT, highlighting a need for additional tools to identify at-risk patients,” the authors explained.

The aim of the current study was to determine if using the AI-ECG was better than a normal QTc alone in distinguishing patients with genetically confirmed LQTS who presented with a normal QTc value at rest (concealed LQTS) from those with normal QTc, they wrote.

In an editorial accompanying the study, Geoffrey H. Tison, MD, MPH, of the University of California San Francisco, noted that congenital LQTS is usually recognized by its namesake prolongation of the QT interval on the 12-lead ECG, and that underlying pathophysiologic mechanisms capable of causing this syndrome are heterogeneous, complex, and not fully understood.

“Despite the numerous advances made in elucidating the various genetic causes of LQTS, the everyday tool used by physicians to screen for LQTS remains the standard ECG,” Tison said. “What distinguishes modern machine learning algorithms, such as CNNs [convolutional neural networks], from the algorithms currently providing preliminary ECG interpretations is that they do not necessarily mimic and so are not limited to the same sets of rules and criteria used by physicians.”

“Instead, they can learn for themselves from the training data providing a wide range of possibilities by which to potentially advance ECGs’ utility including, but not limited to, improved diagnostic performance,” he said.

For this diagnostic case-control study, the authors used all available ECG data from January 1999 through December 2018 from 2,059 (57% men; mean age 21.6 at first ECG) patients. Participants presented to a specialized genetic heart rhythm clinic, which was a study limitation so the results may not be generalizable to other settings.

“Patients were included if they had a definitive clinical and/or genetic diagnosis of type 1, 2, or 3 LQTS (LQT1, 2, or 3) or were seen because of an initial suspicion for LQTS but were discharged without this diagnosis,” they wrote.

They trained the CNN using 60% of the patients, validated it with 10% of the patients, and then tested it on the remaining 30%.

Ackerman’s group found that the AI-ECG capably discriminated between three main genotypic subgroups, LQT1, LQT2, and LQT3:

  • LQT1 versus LQT2 and LQT3: AUC 0.921 (95% CI 0.890-0.951).
  • LQT2 versus LQT1 and LQT3: AUC 0.944 (95% CI 0.918-0.970).
  • LQT3 compared with LQT1 and LQT2: AUC 0.863 (95% CI 0.792-0.934).

Other study limitations included the fact that there was no external validation, and that the AI-ECG model will require training and validation with a larger, unselected patient population before it can be deployed outside a clinical study.

“Ultimately, the real-world clinical utility achieved by neural network–based ECG interpretation will be largely dictated by the success of their integration into the clinical workflow and the scale of their adoption: key directions requiring collaboration between the investigators developing these technologies and the clinicians interpreting and deploying them in daily medical practice,” Tison stressed.

  1. Artificial intelligence-based ECG (AI-ECG) successfully distinguished patients with long QT syndrome (LQTS) from those who were evaluated for LQTS but discharged without this diagnosis at a specialized arrhythmia clinic.

  2. Currently just a research tool, AI-ECG could offer a streamlined and inexpensive method for early detection of congenital LQTS.

Shalmali Pal, Contributing Writer, BreakingMED™

The study was supported by the Mayo Clinic Windland Smith Rice Comprehensive Sudden Cardiac Death Program and Mayo Clinic Center for Clinical and Translational Research/National Center for Advancing Translational Sciences.

Ackerman and co-authors reported relationships with AliveCor, Boston Scientific, Daiichi Sankyo, Invitae, LQT Therapeutic, Medtronic, MyoKardia, Abbott, Audentes Therapeutics, Biotronik, and UpToDate. A co-author is an employee of AliveCor and reported having a patent pending to AI for QT assessment.

Tison reported support from, and/or relationships with, the NIH, General Electric, Janssen Pharmaceuticals, Myokardia, and Cardiogram.

Cat ID: 914

Topic ID: 74,914,730,914,192,925