In Echocardiography (Mount Kisco, N.Y.)
INTRODUCTION AND AIM : Patients undergoing exercise echocardiography with no evidence of myocardial ischemia are considered a low-risk group; however, this group is likely heterogeneous in terms of short-term adverse events and subsequent cardiac testing. We hypothesized that unsupervised cluster modeling using clinical and stress characteristics can detect heterogeneity in cardiovascular risk and need for subsequent cardiac testing among these patients.
METHODS : We retrospectively studied 445 patients who had exercise echocardiography results negative for myocardial ischemia. All patients were followed for adverse cardiovascular events, subsequent cardiac testing, and nonacute coronary syndrome (ACS) revascularization. The heterogeneity of the study subjects was tested using computational clustering, an exploratory statistical method designed to uncover invisible natural groups within data. Clinical and stress predictors of adverse events were extracted and used to construct 3 unsupervised cluster models: clinical, stress, and combined. The study population was split into training (357 patients) and testing sets (88 patients).
RESULTS : In the training set, the clinical, stress, and combined cluster models yielded 5, 4, and 3 clusters, respectively. Each model had 1 high-risk and 1 low-risk cluster while other clusters were intermediate. The combined model showed a better predictive ability compared to the clinical or stress models alone. The need for future testing mirrored the pattern of adverse cardiovascular events. A risk score derived from the combined cluster model predicted end points with acceptable accuracy. The patterns of risk and the calculated risk scores were preserved in the testing set.
CONCLUSIONS : Patients with no evidence of ischemia on exercise stress echocardiography represent a heterogeneous group. Cluster-based modeling using combined clinical and stress characteristics can expose this heterogeneity. The findings can help better risk-stratify this group of patients and aid cost-effective healthcare utilization toward better diagnostics and therapeutics.
Omar Alaa Mabrouk Salem, Ramirez Roberto, Haddadin Faris, Sabharwal Basera, Khandaker Mariam, Patel Yash, Argulian Edgar
cluster analysis, coronary artery disease, exercise echocardiography, machine learning, stress testing