Machine learning may personalize care for patients with severe cardiomyopathy

Machine learning trained on heart imaging data can effectively predict one-year cardiovascular events in patients with a severe form of cardiomyopathy, according to research published June 11 in the European Journal of Radiology.

Dilated cardiomyopathy (DCM) is associated with a 5-year mortality ratee as high as 20%, wrote lead author Rui Chen, with South China University of Technology’s School of Medicine in Guangdong Province, China, and colleagues. Machine learning (ML) has been used to predict events in patients with coronary artery disease, acute myocardial infarction and pulmonary hypertension.

With this in mind, Chen et al. created and performed 10-fold cross-validation on a ML-based risk model using features taken from baseline, laboratory, electrocardiography (ECG), echocardiography and cardiovascular resonance (CMR) imaging data to predict 1-year outcomes in patients with severe DCM.

The ML model was created based on data from 98 patients (18 years or older) who were diagnosed with severe DCM (left ventricular ejection fraction  greater than 35%) at one of two hospitals between October 2014 and March 2017.

In total, 32 clinical data features were input, with those highly relevant to the cardiovascular events chosen by Information gain (IG), which measures how much information a feature provides for classification.

Twenty-two patients met the one-year end-point criteria—defined as any cardiovascular events—and the machine learning platform performed well, according to the researchers, achieving an area under the curve (AUC) of the receiver operating characteristics of 0.887. The top features selected by the IG included left atrial size, QRS duration and systolic blood pressure.

In fact, ML recorded a higher AUC score than left ventricular ejection fraction (0.504) and Meta-analysis Global Group in Chronic Heart Failure (MAGGIC) score (0.599), also used to predict mortality in patients with DCM, the authors noted.

Machine learning also had a higher specificity and sensitivity compared to both LVEF and MAGGIC.

A new (DCM) risk-predicting system with better predictive performance is still needed. To build such a system, ML is a useful tool, because it can handle a large number of features and better focus on predicting events in each individual,” the authors wrote. “Indeed, it achieved better performance than LVEF and MAGGIC Score in the present study.”

Chen and colleagues also believe their model will be easy to apply to other institutions and may help clinicians with risk stratification and individual patient management, they concluded.