A new study reported last month found that computer-based rhythm interpretation, a standard feature of most modern electrocardiograms – often is inaccurate, and without physician intervention, could lead to a false heart rhythm diagnosis. This year alone, nearly 100 million Americans will have an ECG, which is used to diagnose heart disease – including heart attack, acute angina and hypertrophy.
The New York-Presbyterian Hospital/Weill Cornell Medical Center (New York) study was published in the Journal of Electrocardiology. Based on an analysis of 4,297 consecutive ECG recordings at New York-Presbyterian/Weill Cornell, 13.2% of computer-based rhythm statements required physician revision. Nearly half (45.7%) of all inaccurate rhythm statements involved the 343 patients with pacemakers. Of these, three quarters (258) had computer interpretations that required physician revision. Excluding patients with pacemakers, the required revision rate for computer-based rhythm statements was 7.8%.
“Despite improvements in the computer algorithm, our study finds that computer rhythm statements can be flawed and therefore must remain secondary to the expertise and judgment of a physician, and physician checking – or ‘over-reading’ – to verify or correct computer-based electrocardiogram rhythm diagnoses remains mandatory,” said Paul Kligfield, MD, the study’s senior author. The professor of medicine at Weill Medical College of Cornell University and attending physician at NewYork-Presbyterian/ Weill Cornell added: “It is very possible to miss conditions like atrial fibrillation, atrial flutter, etc., if an experienced cardiologist does not look at the ECG. It is also possible for the computer to state that these irregular rhythms are present when they are not.”
Kligfield told Cardiovascular Device Update that while the computerized ECG is “good at measuring some things, it’s very bad at measuring other things.” While he said computers were particularly suited for reading automated measurements of signals that it can average such as amplitude and duration of waves, “The problem is identifying the different wave forms.” He noted that some of these waves are difficult to see, particularly the P wave. “If you can’t accurately see P waves and measure them reliably or if there is noise that looks like a P wave, then the algorithm that tries to determine what the rhythm is won't work.”
Kligfield said that a misdiagnosis of cardiac rhythm could delay treatment, such as blood thinners to prevent blood clots with some rhythms such as atrial fibrillation, or suggest the wrong treatment. “The computer is good, and it is getting better, but alone it is not good enough,” he said, adding that, “unfortunately, there sometimes is an over-reliance on computer interpretation of the ECG.” By contrast, he said that for more than 25 years, every computer-based ECG at NewYork-Presbyterian/Weill Cornell has been checked by an attending cardiologist and all diagnostic statements modified when appropriate.
The study found a significant degree of false-negative rhythm diagnoses by computer: 10% of primary rhythms other than normal heart rhythm – including atrial fibrillation, atrial flutter, and atrial tachycardia – were incorrectly stated to be normal rhythms by the computer algorithm. Importantly, it was found that the failings identified by the study for patients without pacemakers were a result of the ECG software algorithm – not a hardware problem, said Kligfield.
“In the case of patients with pacemakers, the problem is predominantly one of not recognizing the modern pacemaker signal, which is very small in both amplitude and duration compared with the electrical signal of the heart,” he said. “Sometimes, even when the pacemaker signal is detected, it is just not interpreted correctly in terms of the overall rhythm of the heart.”
He said the study findings highlight “the importance of current efforts to understand the clinical implication of, and to improve, these diagnostic algorithms.” While Kligfield noted that some components of the algorithm, such as enhancing the pacemaker detection component, could be easily solved with existing technology, tweaking other components could have a negative affect on the entire cascade that goes into developing the computer algorithm.
The study made use of the most recent version of GE Healthcare’s (Waukesha, Wisconsin) MUSE 12 SL software. All results were checked by one of two cardiologists, and all disagreements with the initial computer rhythm statement were reviewed by the second cardiologist to achieve physician consensus, which was used as the “gold standard” for rhythm diagnosis. Half or more of U.S. hospitals and cardiologists use that company’s ECG system, Kligfield said. “GE is aware of this situation and is actively working to improve this, as they have been all along.”
He said the computer programs for all companies involved in this sector, as far as the current state of the art allows, are nearly optimized. “The reality is these programs are honestly meant to be used as adjuncts to the cardiologist, not a substitute for the physician.” The one big thing that could revolutionize the computer diagnosis, Kligfield said, “would be the ability for the algorithms to very accurately determine what is a P wave and what is not a P wave and to relate them to the complexes around it.”