Myocardial deformation assessed by machine learning is associated with heart failure, atrial fibrillation, and mortality in the elderly
J O Simonsen, F O Olsen, D M Modin, J C Christensen, S D Diederichsen5, L K Koeber, S H Hoejberg, C K Kronborg, A B Brandes, K H Haugan, C G Graff, S R Rasmussen, J T H Tfelt-Hansen, J S Svendsen, T B S Biering-SoerensenAbstract
Background
Over the last decade, novel echocardiographic parameters, based on deformation curves, have been established as reliable parameters to assess left ventricular function and demonstrated predictive utility for adverse cardiovascular outcomes. However, these parameters are single measures of the curves, which may not capture the complexity of the entire strain and tissue doppler imaging (TDI) curves. Signal analysis and machine learning (ML) have previously demonstrated a promising capability to assess the complexity of biological signals, including electrocardiograms and electroencephalography. Consequently, the same techniques might be applicable in myocardial deformation curves, including strain and TDI curves.
Purpose
We sought to investigate whether a new ML-derived variable, “myocardial deformation score,” based on echocardiographic myocardial deformation curves, associates with cardiovascular events and all-cause mortality independently of known clinical and conventional echocardiographic parameters in an elderly population.
Methods
A previously developed ML model for predicting heart failure (HF) or cardiovascular (CV) death, based on entire deformation curves from strain and tissue doppler imaging (the CCHS-HF-ML model), was deployed. We investigated the association between the model output of normalized probabilities, which we termed myocardial deformation score (ranging from 1: low risk to 10: high risk), and the composite of HF and CV death, all-cause mortality, atrial fibrillation, and stroke in an elderly population. Associations were evaluated with Cox proportional hazard models adjusted for clinical parameters, including age, sex, body mass index, smoking pack years, systolic blood pressure, diabetes, ischemic heart disease, and randomization group (if randomized to a loop recorder or not). Furthermore, the models were adjusted for echocardiographic variables, including left ventricular ejection fraction (LVEF), global longitudinal strain (GLS), left ventricular mass index, relative wall thickness, E/e’, E/A, and left atrial volume index.
Results
The study included 1363 participants from the LOOP study, a randomized controlled clinical trial that enrolled participants aged 70 years or older with cardiovascular risk factors but without atrial fibrillation. Myocardial deformation score was associated with increased risk of HF and CV death [HR: 1.44, CI: 1.08:1.93, p = 0.014, per 1 increase in score] (Figure 1A), all-cause mortality [HR 1.27, CI: 1.03:1.56, p = 0.023] (Figure 1B), atrial fibrillation [HR: 1.16, CI: 1.02:1.32, p = 0.024] (Figure 1C) but not stroke (Figure 1D) when adjusted for clinical and echocardiographic confounders.
Conclusion
A ML-derived myocardial deformation score based on strain and tissue doppler imaging curves was associated with increased risk of the composite of HF and CV death, all-cause mortality, and atrial fibrillation, but not of stroke in an elderly at-risk population.Figure 1