Study: Buried data in ECGs predict risk of death after ACS
A multidisciplinary team of engineers and cardiologists reported that what appears to be noise on electrocardiograms (ECGs) contains useful information for estimating the risk of death for patients who experience acute coronary syndrome (ACS). In a study published in the Sept. 28 issue of Science Translational Medicine, the team proposed that sophisticated computational techniques applied to routinely collected ECG data could be integrated into clinical practice to provide more accurate risk stratification.
Zeeshan Syed, PhD, an electrical engineering and computer science assistant professor at the University of Michigan in Ann Arbor, Mich., collaborated with cardiologists and engineers in Massachusetts to develop what they termed computational cardiac biomarkers. Their goal was to supplement biomarkers such as B-type natriuretic peptides with biomarkers derived from time-series analyses of continuous ECG data. By using data-extracting methods, they hypothesized that they could detect anomalies that otherwise are overlooked in the wash of data.
“We’re reaching the point in medicine where our ability to collect data has far out-stripped our ability to analyze or digest it,” John V. Guttag, PhD, co-author and electrical engineering and computer science professor at Massachusetts Institute of Technology in Cambridge, Mass., said in a statement. “You can’t ask a physician to look at 72 hours worth of ECG data.”
Working with cardiologists at Brigham and Women’s Hospital and Harvard Medical School in Boston, the researchers used two different data sets to first identify biomarkers and then assess the biomarkers’ prognostic capability.
In the first stage, Syed et al applied machine learning and data mining techniques to data from the TIMI-DISPERSE 2 clinical trial, which used ECG in a comparison of the safety and initial efficacy of the reversible oral adenosine diphosphate receptor antagonist AZD6140 with clopidogrel (Plavix, Bristol-Myers Squibb, Sanofi Aventis) in patients with non-ST-segment elevation acute coronary syndromes (NSTE-ACS). The researchers identified three computational biomarkers for potential risk stratification for post-ACS patients, all based on long-term analyses: morphologic variability (MV), symbolic mismatch (SM) and heart-rate motif (HRM).
MV assessed myocardial instability; SM quantified anomalous signals; and HRM addressed the integration of heart-nervous system signaling.
“Each of these computational biomarkers uses time-series analytical techniques to extract new types of information that are presently unappreciated in large volumes of continuous cardiovascular data,” the authors wrote. “These computational biomarkers measure subtle features of long-term ECG data that are outside the scope of human visualization but are consistently associated with increased or decreased future risk.”
In the second stage, the researchers validated the biomarkers using the MERLIN-TIMI 36 (Metabolic Efficiency with Ranolazine for Less Ischemia in NSTE-ACS Thrombolysis in Myocardial Infarction 36) trial, which included continuous ECG data as well as baseline clinical and follow-up data on more than 4,500 patients. The results showed a strong association between all three computational biomarkers and cardiovascular death (CVD) in a two-year period after ACS.
“These biomarkers were independent of information in clinical risk scores … echocardiographic measurements such as LVEF [left ventricular ejection fraction] and an extensive panel of other ECG-based metrics,” Syed and colleagues wrote. “The use of computationally generated biomarkers simultaneously improved discrimination, net reclassification, precision and recall over existing approaches of predicting risk of CVD after NSTE-ACS.”
They concluded that the optimal model for risk stratification should combine computational biomarkers and existing metrics, pointing out that the ECG data are routinely captured for patients with ACS already.
“By incorporating the algorithms to measure MV, SM and HRM in monitoring devices by the bedside, or in systems where data from continuous monitoring may be uploaded, these computational biomarkers can be automatically assessed without imposing additional burden on either patients or caregivers,” they wrote. “Ultimately, we envision a strategy of translating these methods into the clinic.”
Zeeshan Syed, PhD, an electrical engineering and computer science assistant professor at the University of Michigan in Ann Arbor, Mich., collaborated with cardiologists and engineers in Massachusetts to develop what they termed computational cardiac biomarkers. Their goal was to supplement biomarkers such as B-type natriuretic peptides with biomarkers derived from time-series analyses of continuous ECG data. By using data-extracting methods, they hypothesized that they could detect anomalies that otherwise are overlooked in the wash of data.
“We’re reaching the point in medicine where our ability to collect data has far out-stripped our ability to analyze or digest it,” John V. Guttag, PhD, co-author and electrical engineering and computer science professor at Massachusetts Institute of Technology in Cambridge, Mass., said in a statement. “You can’t ask a physician to look at 72 hours worth of ECG data.”
Working with cardiologists at Brigham and Women’s Hospital and Harvard Medical School in Boston, the researchers used two different data sets to first identify biomarkers and then assess the biomarkers’ prognostic capability.
In the first stage, Syed et al applied machine learning and data mining techniques to data from the TIMI-DISPERSE 2 clinical trial, which used ECG in a comparison of the safety and initial efficacy of the reversible oral adenosine diphosphate receptor antagonist AZD6140 with clopidogrel (Plavix, Bristol-Myers Squibb, Sanofi Aventis) in patients with non-ST-segment elevation acute coronary syndromes (NSTE-ACS). The researchers identified three computational biomarkers for potential risk stratification for post-ACS patients, all based on long-term analyses: morphologic variability (MV), symbolic mismatch (SM) and heart-rate motif (HRM).
MV assessed myocardial instability; SM quantified anomalous signals; and HRM addressed the integration of heart-nervous system signaling.
“Each of these computational biomarkers uses time-series analytical techniques to extract new types of information that are presently unappreciated in large volumes of continuous cardiovascular data,” the authors wrote. “These computational biomarkers measure subtle features of long-term ECG data that are outside the scope of human visualization but are consistently associated with increased or decreased future risk.”
In the second stage, the researchers validated the biomarkers using the MERLIN-TIMI 36 (Metabolic Efficiency with Ranolazine for Less Ischemia in NSTE-ACS Thrombolysis in Myocardial Infarction 36) trial, which included continuous ECG data as well as baseline clinical and follow-up data on more than 4,500 patients. The results showed a strong association between all three computational biomarkers and cardiovascular death (CVD) in a two-year period after ACS.
“These biomarkers were independent of information in clinical risk scores … echocardiographic measurements such as LVEF [left ventricular ejection fraction] and an extensive panel of other ECG-based metrics,” Syed and colleagues wrote. “The use of computationally generated biomarkers simultaneously improved discrimination, net reclassification, precision and recall over existing approaches of predicting risk of CVD after NSTE-ACS.”
They concluded that the optimal model for risk stratification should combine computational biomarkers and existing metrics, pointing out that the ECG data are routinely captured for patients with ACS already.
“By incorporating the algorithms to measure MV, SM and HRM in monitoring devices by the bedside, or in systems where data from continuous monitoring may be uploaded, these computational biomarkers can be automatically assessed without imposing additional burden on either patients or caregivers,” they wrote. “Ultimately, we envision a strategy of translating these methods into the clinic.”