More people were correctly identified as having long QT syndrome (LQTS) when their electrocardiograms were analyzed with artificial intelligence (AI) rather than the QTc metric alone, researchers reported.
The task of differentiating people with LQTS from peers, evaluated for the condition but not diagnosed, was performed better by a convolutional neural network — with an area under the receiver operating characteristic curve (AUC) of 0.900 — than by ECG-derived QTc classification (AUC 0.824), according to Michael Ackerman, MD, PhD, of Mayo Clinic in Rochester, Minnesota, and colleagues.
Similarly, AI outperformed the QTc in distinguishing concealed LQTS, or cases with genetically confirmed LQTS despite a normal resting QT interval (AUC 0.863 vs 0.741), from cases dismissed as normal, Ackerman’s group reported in a study published online in JAMA Cardiology.
“This model may aid in the detection of LQTS in patients presenting to an arrhythmia clinic and, with validation, may be the stepping stone to similar tools to be developed for use in the general population,” the authors said.
Their neural network had an overall accuracy of 78.7% and was even able to distinguish among the three main genotypic subgroups of LQTS (AUCs ranging from 0.863 to 0.944).
The problem of diagnosing LQTS today is that although QT prolongation is the hallmark feature of the condition, about 40% of genetically confirmed LQTS patients will have a normal-looking QT on a resting ECG. This group remains at risk for serious LQTS-associated arrhythmias, but their normal QTc does not qualify them for further evaluation.
“Although individuals with QTc values within the normal limits are at significantly lower risk of such arrhythmias, identification of these patients is still important to, at minimum, implement the potential lifesaving measure of avoiding drugs that prolong the QT interval and perhaps institute prophylactic β-blocker therapy (class IIa recommendation) and help to identify family members who might be at (possibly higher) risk,” according to Ackerman and colleagues.
Their diagnostic case-control study included all 12-lead ECGs from 2,059 patients evaluated for LQTS at a specialized genetic heart rhythm clinic from 1999 to 2018. Participants were 57% men, with average age at first ECG being 21.6 years.
The cohort was split between the 967 diagnosed with LQTS and 1,092 dismissed as not having LQTS.
Ackerman’s group trained their AI on 60% of the patients, validated it in 10% of the patients, and tested it on the remaining 30%.
“This study contributes to the important body of work demonstrating how machine learning algorithms, properly applied, can expand the utility of existing medical data sources, such as ECGs, to extract novel insights and improve performance, in many cases without significant additional invasiveness or cost,” according to Geoffrey Tison, MD, MPH, of the University of California San Francisco, writing in an accompanying editorial.
“Most prior algorithmic approaches could not feasibly accept complete raw ECG data as their input, forcing investigators to derive summaries using a priori constraints, such as calculating T wave characteristics or measuring the QT interval. Instead, a CNN [convolutional neural network] can derive any of these summary characteristics and countless more, as they are all contained within the raw voltage data,” Tison noted.
A properly trained neural network can therefore do what humans cannot.
“This goal could be achieved through the use of as-yet unrecognized features or subtle, visually imperceptible elements of the ECG waveform. The CNN in this study appears to have achieved this, since it identified concealed LQTS from ECGs that otherwise appeared to be normal to expert cardiologists,” the editorialist said.
Limiting the generalizability of the study’s results were its single-center nature as well as the selection of patients with suspicion of possible LQTS but discharged without that diagnosis, Ackerman and colleagues cautioned. They noted that their findings require further internal validation within the study clinic as well as external validation and calibration in a different center or larger, unselected population.
Previously, Ackerman’s group announced that they had trained an AI model to detect concealed LQTS from a single ECG lead in a collaboration with AliveCor.
“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,” according to Tison.
Last Updated February 10, 2021
The study was funded by an NIH grant.
Ackerman is a consultant for Abbott, Audentes Therapeutics, Biotronik, Boston Scientific, Daiichi Sankyo, Invitae, LQT Therapeutics, Medtronic, MyoKardia, and UpToDate; and has an equity/royalty relationship with AliveCor.
Tison disclosed receiving an NIH grant; prior research support from General Electric, Janssen Pharmaceuticals, and MyoKardia; and is an unpaid adviser to Cardiogram.