Prognostic value of right atrial strain derived from cardiovascular magnetic resonance in non-ischemic dilated cardiomyopathy

Patients with DCM who underwent CMR imaging from June 2012 to March 2019 were prospectively enrolled in this study. DCM was diagnosed in accordance with the classification of cardiomyopathies from the European Society of Cardiology Working Group on Myocardial and Pericardial Diseases [14]. Exclusion criteria included significant coronary artery disease, defined as > 50% diameter stenosis of epicardial coronary artery on x-ray coronary angiography or coronary computed tomography (CT), a history of myocardial infarction or infarct patterns of late gadolinium enhancement (LGE); primary heart valvular disease; congenital heart disease; arrhythmia related cardiomyopathy; infiltrative heart disease; constrictive cardiomyopathy; acute myocarditis; and peripartum cardiomyopathy. In addition, patients with excessive acholic consumption, contradiction to CMR, atrial fibrillation or significant arrhythmia during CMR, and poor image quality were also excluded in this study. 120 sex- and age-matched healthy volunteers were selected from our database [15]. This study was approved by the institutional ethics of West China Hospital of Sichuan University. All subjects provided a written informed consent.

CMR was performed on a 3T CMR scanner (MAGNETOM, Tim Trio system; Siemens Healthineers, Erlangen, Germany) with a 32-channel phased array cardiac coil. A balanced steady-state free precession (bSSFP) cine images were acquired covering the entire LV, continuously from the base to the apex on short-axis views, and long-axis views (2-, 3- and 4-chamber) by breath-holding and electrocardiographic gating. The scan parameters were as follows: field of view, 320–340 mm²; repetition time, 3.4 ms; echo time, 1.3 ms; slice thickness, 8 mm with no gap; flip angle, 50°; acquisition matrix, 256 × 144; temporal resolution, 42 ms; spatial resolution, 1.4 × 1.3 mm²; and number of frames, 25 per cardiac cycle. A phase-sensitive inversion recovery sequence was used to obtain LGE images, 10–15 min after gadolinium (Magnevist; Bayer Healthcare, Berlin, Germany) injection with 0.15 mmol/kg per bolus. The scan parameters were as follows: field of view, 260 × 340 mm²; repetition time, 700 ms; echo time, 2.0 ms; delay time after the inversion pulse, 300–380 ms; slice thickness, 8 mm; flip angle, 20°; and matrix size, 116 × 192.

Biventricular volume, function, and LV mass were measured by Qmass (version 8.1, Medis, Leiden, Netherlands) according to the standardized protocol of the Society for Cardiovascular Magnetic Resonance post-processing guideline [16]. Ventricular mass and volume were indexed to body surface area. The presence of LGE was evaluated by 2 independent readers who were blinded to clinical data. LA volume and function were calculated using the biplane area-length method by tracing LA endocardial contour exclusion of pulmonary veins and LA appendage.

RA maximum volume, RA volume prior to atrial contraction, and RA minimum volume were assessed using area-length method in the 4-chamber during various RA phase with the exclusion of RA appendage using Qmass (version 8.1, Medis). The RA total empty fraction, RA passive fraction, and RA active fraction were calculated using corresponding RA volumetric measurements.

The fast long-axis strain was performed by semi-automated tracking of the length between the mid-posterior RA wall and right atrioventricular junctions on CMR 4-chamber view. The mid-posterior RA wall was selected at the intersection point of RA posterior wall and the RA long-axis, and the right atrioventricular junctions were defined as the tricuspid valve insertion points at the septal and free wall borders of the annuals. The three points were marked at the ventricular end-diastolic phase and tracking was applied through the cardiac cycle using the method of template matching. Manual adjustments were performed when necessary for optimal tracking. RA longitudinal strain then was calculated according to the Lagrange strain equation: (L_(t) – L₀) × 100/L₀. L_(t) is the distance between the mid-posterior point and the atrioventricular points at any time (t) during the cardiac cycle. L₀ is the distance at RV end-diastole phase. RA reservoir strain, passive strain and active strain were measured at different RA phase. Illustrative example of measurement is shown in Fig. 1.

Patients’ follow-up was conducted until October 2020 by review of medical records, telephone interviews, and contact with the patients’ physician by a cardiologist blinded to CMR data. The cause of hospitalization and death was carefully analyzed according to patients’ symptoms, signs and medical documents. The predefined primary endpoint was all-cause mortality. The composite heart failure (HF) endpoint included HF death, HF readmission, and heart transplantation. The follow-up duration was calculated from the CMR date until endpoint occurred or last contact. Only the first event for patients was included in analysis for composite HF endpoint. Patient data were censored at the time of last follow-up for patients without predefined events.

Normally distributed data were presented as mean ± standard deviation and compared with independent t tests, and non-normally distributed data were reported as median (interquartile range) and compared with Mann–Whitney U test. Chi-square test or Fisher’s exact test were performed to compare categorical variables as appropriate. Pearson’s and Spearman’s correlation coefficients were used to determine the correlation between RA strain parameters and other clinical or CMR variables, as appropriate. N-terminal pro-brain natriuretic peptide (NT-ProBNP) was analyzed using log transformed values. Cox proportional hazards model was performed to examine the prognostic value of variables in predicting the clinical endpoints. We included variables with P < 0.05 in the univariable Cox analysis as covariates in the multivariable model to determine the independent predictors of clinical endpoints. Variance inflation factor (VIF) was calculated to avoid collinearity. One variable was incorporated into the multivariable analysis among parameters with VIF > 3. Survival curves were drawn by Kaplan–Meier method and compared by log-rank test. The incremental prognostic value of RA strain was evaluated by comparing the improvement of Chi-square value of Cox analysis after adding the strain parameters to baseline model. The baseline model included variables with P < 0.05 in the univariable Cox analysis. Models were compared by using likelihood ratio test. Receiver operating characteristic (ROC) curves were used to assess the discriminating power between patients with and without clinical endpoints for LA function and volume parameters. The statistical comparison of ROC curves was computed by using the nonparametric DeLong test.

For reproducibility analysis, we randomly selected 20 patients. Interobserver reproducibility was measured by two independent operators blinded to the clinical data. Intraobserver reproducibility was performed by a single operator 2 months later blinded to the results from the first measurements. Then, intraclass correlation coefficient (ICC) and Bland–Altman analysis were performed.

The results of univariable Cox regression analysis are summarized in the Additional file 1: Table S1. All three RA strain parameters were predictors of the all-cause mortality and composite HF endpoint. We incorporated the following covariates in the multivariable regression model: age, systolic BP, NYHA class, Log transformed NT-ProBNP, LVEF, LV end-diastolic volume index, RVEF, LGE, LA end-diastolic volume index, RA minimal volume index, and RA strain. The results showed that RA reservoir strain [hazard ratio (HR) (per 5% decrease) 1.19, 95% confidence interval (CI) 1.03–1.37, P = 0.022] and conduit strain [HR (per 5% decrease) 1.37, 95% CI 1.03–1.84, P = 0.033] were independent predictors of all-cause mortality, while RA booster strain was not statistically significant in multivariable analysis. Likewise, RA reservoir strain [HR (per 5% decrease) 1.16, 95% CI 1.04–1.30, P = 0.008] and RA conduit strain [HR (per 5% decrease) 1.40, 95% CI 1.13–1.74, P = 0.002] were independently associated with composite HF endpoint. The results are shown in Tables 3 and 4. Kaplan–Meier analysis stratified by median of RA strain values demonstrated that patients with RA strain better than median (RA reservoir strain > 22.6%, RA conduit strain > 9.9%, and RA booster strain > 11.9%) presented significant lower risk of all-cause mortality and composite HF endpoint than those with RA strain worse than median (All Log rank P < 0.001). Results are shown in Fig. 2. The relationship between RA strain and relative hazard ratio is shown in Fig. 3. The cubic splines presented that mortality and HF endpoint risk increased with the decrease of RA strain value. Adding RA strain value to model with clinical and CMR risk factors resulted in a significant increase of chi-square value in the prediction of adverse clinical outcomes (all P < 0.05). The results are shown in Fig. 4.

ROC curve analysis showed that RA reservoir and conduit strain demonstrated significantly greater area under the curve (AUC) than RA volume and empty fraction for predicting the all-cause mortality and composite HF endpoint (all P < 0.05). RA booster strain had similar AUC compared with the RA empty fraction in predicting the adverse events (P = 0.06 for all-cause mortality, and P = 0.17 for HF endpoint), but still greater than RA volume parameters (P < 0.05) (Fig. 5).

In a subgroup analysis among patients with LVEF < 35%, RA reservoir strain and conduit strain remained independent predictors of all-cause mortality and composite HF endpoint (Additional file 1: Table S2).

RA strains were moderately correlated with RA volume and empty fraction (r = 0.35 to 0.66, all P < 0.001), LVEF (r = 0.41 to 0.56, all P < 0.001), RVEF (r = 0.54 to 0.62, all P < 0.001), LA emptying fraction (r = 0.61 to 0.70, all P < 0.001), and reversely correlated with LV end-diastolic volume index (r = − 0.25 to − 0.30, all P < 0.001) and Log (NT-ProBNP) (r = − 0.43 to − 0.51, all P < 0.001). The results are shown in Table 5.

Reproducibility results are presented in Table 6. The intra-observer and inter-observer reproducibility was excellent for RA strain derived from fast long-axis method with all ICC greater than 0.90 and relatively low bias.