Speckle-tracking echocardiography in comparison with ejection fraction for prediction of cardiovascular mortality in patients with end-stage renal disease

Abstract

Background: Cardiovascular disease is the major cause of death in end-stage renal disease (ESRD). To develop better means to assess cardiovascular risk in these patients, we compared conventional echocardiography-derived left ventricular ejection fraction (EF) with the novel method of 2D speckle-tracking echocardiography to determine cardiac strain.

Methods: Predictive performances of conventional EF and speckle-tracking echocardiography-derived global longitudinal strain (GLS) were compared using receiver-operator curve (ROC) analyses and calibration by calibration plots. We also took into account other known cardiovascular risk factors through multivariable logistic regression analysis.

Results: The study comprised 171 ESRD patients (mean age 64 years, 64% male) on maintenance dialysis therapy (93% haemodialysis, 7% peritoneal dialysis) for an average period of 39 months. During 2.1 years of follow-up, 42 patients (25%) died from cardiovascular disease. ROC analysis of GLS resulted in an area under the curve of 0.700 [95% confidence interval (CI) 0.603-0.797] compared with an area under the curve of EF of 0.615 (95% CI 0.514-0.716) (P = 0.059 for difference). The total absolute deviation between predicted and observed outcome frequencies obtained by calibration plots were 13.8% for EF compared with only 6.4% for GLS. Best results of ROC analysis (area under the curve = 0.759; P = 0.06), calibration and goodness-of-fit (χ² = 28.34, P ≤ 0.0001, R ² = 0.25) were achieved for GLS added to a baseline model consisting of known cardiovascular risk factors in a multivariate regression analysis.

Conclusions: In summary, in chronic dialysis patients, GLS is a more precise predictor of cardiovascular mortality than conventional echocardiography-derived EF.

Keywords: ESRD; cardiovascular; dialysis; echocardiography; prognosis.

FIGURE 1:

ROC curves showing the predictive performance of GLS and EF concerning cardiovascular mortality. GLS obtained a better predictive result (area under the curve: 0.700; 95% CI 0.603–0.797) than EF (area under the curve: 0.615; 95% CI 0.514–0.716). Longitudinal strain parameters show negative values as ventricular myocardium shortens along the longitudinal axis during systolic activity. For easier comparison, values of EF were modified by changing the algebraic sign.

FIGURE 2:

ROC curves representing predictive performance of the baseline model including four relevant clinical variables (age, sex, diabetes mellitus and dialysis vintage), as well as ROC curves of EF and GLS when added separately on top of the baseline model. ROC analysis of the baseline model itself yields an area under the curve of 0.690 (95% CI 0.600–0.780). Both, EF and GLS added separately to the baseline model improve the area under the curve. But, additive predictive value of GLS is higher (area under the curve: 0.759; 95% CI 0.673–0.845) than that of EF (area under the curve: 0.735; 95% CI 0.641–0.829). Baseline model = age, sex, diabetes mellitus and dialysis vintage

FIGURE 3:

Bar chart summarizing the results of calibration plots. It shows the absolute difference between predicted and observed frequencies of cardiovascular mortality dependent on different predictors and their discriminative ability. GLS alone and GLS added to the baseline model (Base + GLS) carries the lowest absolute difference (0.077 and 0.064). EF causes higher discrepancy between observed and predicted mortality rates when used as predictor (0.138).