Session S24.1

Comparing Symbolic Representations of Cardiac Activity to Identify Patient Populations with Similar Risk Profiles

Z Syed*, BM Scirica, CM Stultz, JV Guttag

Massachusetts Institute of Technology
Cambridge, MA, USA

The focus of many existing techniques to analyze ECG is to calculate a particular feature from the raw signal, and to use it to rank patients along a risk scale. In our work, we propose a different approach to risk-stratification, by identifying patients at increased risk of adverse cardiovascular outcomes through a comparative rather than a feature based strategy. We directly compare the long-term signals for every pair of patients to determine how similar they are, and use hierarchical clustering to find patients whose electrocardiographic characteristics are substantially different from the rest of the population. The central component of this work is the process of quantifying the differences in the long-term electrocardiographic recordings of two patients. We propose a new metric, called electrocardiographic mismatch (ECGM) between two patients, which is computed by first creating a symbolic representation of the electrocardiographic signal for each patient, and by then carrying out a weighted inter-patient comparison of the symbol distributions. We assessed the ability of ECGM-based clustering to identify patients at increased risk of adverse cardiac events as individuals whose long-term electrocardiogram is substantially different from the rest of the population. When evaluated on 686 patients who suffered NSTE-ACS, hierarchical clustering on an ECGM distance matrix based on ECG the day following ACS was able to find patients at increased risk of death and myocardial infarction over a 90 day follow-up period. The 20% patients most different from the dominant population group were at 5.3 times increased risk of death (p-value: 0.003) relative to other members of the population, and had a 2.8 times increased risk for the combined endpoint of death and myocardial infarction (p-value: 0.003). Our results show that patients at high risk for future cardiovascular events may be identifiable as individuals whose long-term ECG signals are different from the dominant group of the population. Our strategy of using a comparative approach for risk-stratification also allows for a more fine-grained method of determining future risk, i.e., by using ECGM to match new patients with previously seen ones for whom histories of major adverse cardiac events are available.

(Abstract Control Number: 76)