A random forest machine learning model using mechanical dyssynchrony indicators (EF, ESV, EDV) predicted LV synchrony with an AUC of 0.925 and 85.5% accuracy, outperforming traditional indices.
Cohort (n=412)
Does a machine learning model using mechanical dyssynchrony indicators improve the prediction of LV synchrony compared to traditional electrical markers in patients with chronic heart failure?
Mechanical dyssynchrony indicators such as EF, ESV, and EDV are superior to traditional electrical markers in predicting LV synchrony, providing new insights for personalized cardiac resynchronization therapy candidate selection.
Effect estimate: AUC 0.925
Chronic heart failure (CHF) patients often present with heterogeneous patterns of cardiac dyssynchrony. Although QRS prolongation (>150 ms) and left bundle branch block (LBBB) are classical markers of electrical dyssynchrony, their direct association with mechanical dyssynchrony remains controversial. This study aimed to identify key determinants of left ventricular (LV) synchrony using machine learning and explainable artificial intelligence techniques. A cohort of 412 CHF patients was stratified using integrated echocardiographic and electrocardiographic criteria: synchronous group (SG, n = 239) with standard deviation of time to peak longitudinal strain in 18 LV segments (SD18STE) ≤ 33 ms, QRS duration ≤ 120 ms, and left ventricular end-diastolic volume (LVEDV) ≤ 150 mL; asynchronous group (AG, n = 173) with SD18STE > 33 gt; 33 ms typically accompanied by QRS duration > 120 gt; 120 ms and/or LVEDV > 150 gt; 150 mL. All patients underwent speckle tracking echocardiography (STE) to assess LV and left atrial function, along with electrocardiographic evaluation of QRS duration. Key parameters included electrical dyssynchrony markers (QRS duration, LBBB status) and mechanical dyssynchrony markers (LV ejection fraction EF, end-diastolic volume EDV, end-systolic volume ESV, and segmental strain timing). A random forest model was used to identify predictors of mechanical dyssynchrony, and SHapley Additive exPlanations (SHAP) analysis was employed to quantify feature contributions. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1-score. Compared with the SG group, AG patients had significantly higher QRS duration (132 vs. 116 ms, p 100 gt; 100 mL increased the probability of dyssynchrony. Posterior wall delays were strongly associated with dyssynchrony. LBBB presence increased the likelihood of dyssynchrony 3-fold. The model demonstrated excellent performance (AUC = 0.925, accuracy = 85.5%, F1-score = 0.878), outperforming traditional dyssynchrony indices. Mechanical dyssynchrony indicators such as EF, ESV, and EDV are superior to electrical markers in predicting LV synchrony. Dysfunction in posterior wall segments significantly contributes to mechanical asynchrony. These findings provide new insights into CHF pathophysiology and support the use of personalized criteria for cardiac resynchronization therapy candidate selection.
Kong et al. (Mon,) conducted a cohort in Chronic heart failure (CHF) (n=412). Random forest machine learning model vs. Traditional dyssynchrony indices was evaluated on Prediction of mechanical dyssynchrony (AUC 0.925). A random forest machine learning model using mechanical dyssynchrony indicators (EF, ESV, EDV) predicted LV synchrony with an AUC of 0.925 and 85.5% accuracy, outperforming traditional indices.