Discriminatory power of Global Longitudinal Strain and Left Ventricular Ejection Fraction for Identification of Dilated Cardiomyopathy

Dilated cardiomyopathy (DCM) is one of the leading causes of heart failure. Left ventricular ejection fraction (LVEF) is one of the most used features for assessing heart health and predicting outcomes in DCM patients. However, it does have numerous pitfalls. Recent studies suggest that global longitudinal strain (GLS) and heart rate variability (HRV) can be used to predict DCM. Furthermore, numerous studies have demonstrated how essential it is to deploy interpretable machine learning models to aid clinicians in the cases when the diagnosis is challenging. Therefore, we aimed to investigate discriminatory power of GLS and LVEF as a feature of logistic regression models for identification of DCM. The study encompassed 138 DCM and 138 healthy controls (HC). The models were produced by logistic regression algorithms considering the set of selected HRV features and GLS (LogRegGLS) or LVEF (LogRegLVEF). The results showed that the accuracy of produced LogRegGLS model was 86%, higher than the one observed in case of LogRegLVEF model (83%). The produced nomograms also supported the hypothesis of the relevance of the GLS and LVEF features, indicating that these two measurements are the most useful in identification of DCM. In conclusion, our findings highlight the value and efficacy of interpretable machine learning models and suggest that GLS may have more discriminatory power in differentiating between DCM and healthy participants than LVEF.