Enhancing differential diagnosis of IHD and DCM using interpretable machine learning in mildly reduced ejection fraction

Aim: Etiological diagnosis is critical in patients with left ventricular dysfunction, as both dilated cardiomyopathy (DCM) and ischemic heart disease (IHD) can present similarly in the early stages. This study aims to evaluate the discriminative power of global longitudinal strain (GLS) and heart rate variability (HRV) parameters using interpretable machine learning models to differentiate between DCM and IHD patients with left ventricular ejection fraction (LVEF) of between 40% and 50%. Methods: In this retrospective exploratory study, we included consecutive patients with LVEF 40-50% who had a recent (<3 months) 24-h Holter ECG and no history of acute myocardial infarction or heart failure hospitalization. HRV features and GLS were extracted by the processing of Holter ECG and echocardiographic imaging, respectively. Feature selection was performed through the ReliefF method and interpretable predictive models were produced using HRV features, sex, age, and GLS to differentiate between DCM and IHD patients. Results: The study population included 97 DCM patients (63 males and 34 females, aged 57 ± 15 years) and 91 IHD patients (73 males and 18 females, aged 71 ± 11 years). The logistic regression model achieved a classification accuracy of 76% in distinguishing the populations with an area under the curve of 83%. Sex, age, mean RR, FD, HFn, GLS, pNN50, SD1/SD2, SD1, and LFn were identified as the most important features in distinguishing between IHD and DCM. Conclusion: This study highlights the added value of a novel approach based on a predictive model that integrates HRV metrics with myocardial deformation parameters to support the differential diagnosis between DCM and IHD in patients with mildly reduced ejection fraction.