ECG 'Noise' Predicts Death After MI
By Chris Kaiser, Cardiology Editor,
October 03, 2011
Action Points
Explain that three computational biomarkers analyzing data from long-term Holter monitoring identifies individuals with non-ST-segment elevation acute coronary syndromes (NSTEMI ACS) at risk for cardiovascular death.
Note that these biomarkers rely on data already being collected and appear to independently add to traditional predictors.
Review
Certain ECG signals that were once considered "noise" are associated with a higher risk of death following a non-ST-segment elevation acute coronary syndrome (NSTEMI ACS), according to computational ECG analysis of the MERLIN-TIMI36 trial.
Each of three computational biomarkers derived from long-term Holter ECG signals were strongly associated with cardiovascular death, Zeeshan Syed, PhD, from the University of Michigan in Ann Arbor, and colleagues reported.
After adjusting for TIMI risk score, ejection fraction, and other ECG-based metrics, the computationally-generated biomarkers were still independently associated with cardiovascular death, according to the study published online in Science Translational Medicine.
"The main point of our study is that we are able to make better use of data that we're already collecting," Syed told MedPage Today.
"There's prognostic information buried in the noise, and it's almost invisible because of the sheer volume of the data," Syed said. "The sophisticated computational techniques allow us to home in on truly abnormal ECG signals. These patients with unstable hearts are at a greater risk of dying. Identifying them would allow physicians to initiate more aggressive treatment."
Researchers noted that current metrics such as echocardiography and left ventricular ejection fraction (LVEF) don't always correctly stratify ACS patients into high and low risk. The three computational biomarkers, when added to existing predictors, improved classification by 7% to 13%, they wrote.
About 2.5% of patients die within 90 days following an NSTEMI ACS. The rate increases to about 6% within a year.
The use of echo and a conservative threshold of less than 40% LVEF identifies 31.7% of these deaths, according to the study.
Although the majority of the patients identified by echo are at high risk, we are still missing more than two-thirds of these patients who will die," Syed said.
To better help identify these patients, Syed and colleagues used data mining and machine learning techniques to retrospectively search through 24-hour continuous ECGs from 4,557 heart attack patients enrolled in the MERLIN-TIMI36 trial.
They found that the ECG signals from many of the patients who later suffered cardiovascular death contained similar errant patterns that until now were dismissed as noise or simply undetectable.
The three computational biomarkers are:
Morphologic variability, which assesses myocardial instability by quantifying low-amplitude probabilistic variability in the shape of the ECG waveform over long periods of time
Symbolic mismatch, which quantifies the degree to which long-term ECG signals of individual patients are anomalous relative to those of other patients with a similar clinical history
Heart rate motif, which integrates the frequency with which high- or low-risk heart rate patterns reflecting autonomic function appear in a patient's ECG over long time periods.
In the unadjusted analysis, morphologic variability was the strongest predictor of death (HR 3.31, 95% CI 2.49 to 4.40, P<0.001), followed by symbolic mismatch (HR 2.36, 95% CI 1.73 to 3.22, P<0.001) and heart rate motif (HR 2.21, 95% CI 1.65 to 2.97, P<0.001).
Even after adjusting for traditional risk stratifying techniques, the computational biomarkers were still independently associated with dying compared with the other techniques and as well as compared with each other.
But Syed is not ready to do away with conventional risk stratifying techniques.
"Just as disease is multifactorial, we need to continue to apply multiple risk stratifying approaches including functional and electrical activity," Syed told MedPage Today.
One potential area where these biomarkers might help is in identifying candidates for implantable cardioverter defibrillators, he said.
But there is a spectrum of treatments, including drugs, stents, and devices, which are initiated based on assessment from traditional metrics. The hope of Syed and colleagues is that the computational biomarkers will help "to better stratify patients and define where they should fall within the spectrum of treatment."
Syed said he wants to test these biomarkers prospectively on patients in the hospital, and also on asymptomatic people at an increased risk for heart disease.
The study was limited because about 2,000 patients in the MERLIN-TIMI36 trial did not have continuous ECG data available.
In addition, some of the conventional risk metrics could not be assessed in the patient data, which reduced the size of the multivariate model.
Also, these findings are not generalizable to other populations, such as those with STEMI or heart failure, for example.